pcbjam/tests/screenshot-manifest.json

1973 lines
40 KiB
JSON
Raw Normal View History

feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
{
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"_note": "Generated from baseline-screenshots/<engine>/ — run `npm run screenshots:manifest`. The name list is authoritative for removed-screenshot detection.",
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
"screenshots": [
{
"name": "01-loading.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "03-after-load.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "04-controls-tab.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "05-text-input-tab.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "06-text-input-typed.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "07-drawing-tab.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "08-drawing-done.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "09-lists-tab.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "10-lists-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "10b-choice-selected.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "11-file-menu.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "12-help-menu.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "13-final.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "3d-viewer-00-board-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "3d-viewer-models-pic_programmer-render.png",
"engine": "chromium"
},
{
"name": "3d-viewer-models-pic_programmer.png",
"engine": "chromium"
},
{
"name": "3d-viewer-pic_programmer-render.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "3d-viewer-pic_programmer.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "3d-viewer-titlebar.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
{
"name": "appearance-00-layers.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "appearance-01-objects.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "appearance-02-nets.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "appearance-03-layers-again.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "appearance-10-layers-scrolled.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "appearance-11-layers-scrolled-back.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "appearance-20-objects-scrolled.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "aui-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "aui-02-panels.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "aui-03-close-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "aui-04-dragged.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "aui-05-multi-panel.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "aui-resize-01-hover.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "aui-resize-02-mid-drag-live.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "aui-resize-03-mid-drag.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "aui-resize-04-after.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "auinotebook-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "auinotebook-02-tab-switch.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "auinotebook-03-add-tab.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "auinotebook-04-remove-tab.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "auinotebook-05-tab-style.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "bitmapbuttons-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "bitmapbuttons-02-toolbar-click.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "bitmapbuttons-03-toggle.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "bitmapbuttons-04-multi-toggle.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "bitmapbuttons-05-enable-toggle.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "bitmapbuttons-06-shapes.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "bitmapbuttons-07-artprovider.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "bitmask.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "calculator-before-switch.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "calculator-color-code.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "calculator-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "calendar-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "calendar-02-select-date.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "calendar-03-next-month.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "calendar-04-prev-month.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "calendar-05-today.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "clipboard-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "clipboard-02-copy-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "clipboard-03-paste-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "clipboard-04-check-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "clipboard-05-clear-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "clipboard-06-full-flow.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "collapse-relayout-01-loaded.png",
"engine": "chromium"
},
{
"name": "collapse-relayout-02-expanded.png",
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "collapsible-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "collapsible-02-panes.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "collapsible-03-expanded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "collapsible-04-expand-all.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "collapsible-05-collapse-all.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "contextmenu-01-open.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "coroutine-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "coroutine-02-summary.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "coroutine-nested-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "coroutine-nested-02-summary.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataview-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataview-02-list-populated.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataview-03-list-selected.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataview-04-list-add.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataview-05-tree-tab.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataview-06-tree-selected.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataview-07-tree-expanded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataview-08-tree-collapsed.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataview-09-column-click.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataview-10-scrolled.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataviewvirtual-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataviewvirtual-02-large-dataset.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataviewvirtual-03-scroll.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataviewvirtual-04-selection.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dataviewvirtual-05-zone-manager.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialog-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialog-02-info-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialog-03-yesno-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialog-04-error-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialog-05-custom-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-00-initial.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-01-tab-selected.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-custom-closed.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-custom-open.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-full-flow.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-msgbox-error-closed.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-msgbox-error-open.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-msgbox-info-closed.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-msgbox-info-open.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-msgbox-yesno-closed.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-msgbox-yesno-open.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-timer-initial.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-timer-multiple.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-timer-running.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-timer-started.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dialogs-timer-stopped.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dnd-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dnd-02-handlers.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dnd-03-dragenter.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dnd-04-dragleave.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dnd-05-drop.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dnd-06-file-written.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dnd-07-event-fired.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dnd-08-multiple-files.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "dnd-09-with-file.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "earlysize-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "earlysize-02-result.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "eefpsel-01-boot.png",
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "eeschema-crosshair-00-small.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "eeschema-crosshair-01-full.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "eeschema-crosshair-02-45.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "eeschema-crosshair-03-small-again.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "eeschema-draw-wires-00-before-tool-click.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "eeschema-draw-wires-01-after-click.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "eeschema-draw-wires-02-after-drawing.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "eeschema-load-rendered.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "eeschema-loaded-css.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "eeschema-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "eeschema-sim-frame.png",
"engine": "chromium"
},
{
"name": "eeschema-sim-plot.png",
"engine": "chromium"
},
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
{
"name": "eeschema-url-regex.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "filedialog-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "filedialog-02-open-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "filedialog-03-save-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "filedialog-04-multiple-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "filedialog-05-all-buttons.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "filedlg-folder-nav.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "fontenum.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "frame-runtime-eeschema.html.png",
"engine": "chromium"
},
{
"name": "frame-runtime-footprint_editor.html.png",
"engine": "chromium"
},
{
"name": "frame-runtime-pcbnew.html.png",
"engine": "chromium"
},
{
"name": "frame-runtime-symbol_editor.html.png",
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gerbview-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gerbview-02-metrics.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gerbview-print-00-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gerbview-print-01-dialog.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gerbview-print-02-reopened.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "grid-tab-final.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gridedit-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gridedit-02-select-cell.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gridedit-03-edit-cell.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gridedit-04-add-row.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gridedit-05-delete-row.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gridrenderers-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gridrenderers-02-color-cells.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gridrenderers-03-icon-text.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gridrenderers-04-striped.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "gridrenderers-05-checkbox-toggle.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "htmlwin-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "htmlwin-02-basic-content.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "htmlwin-03-tables.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "htmlwin-04-long-content.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "htmlwin-05-kicad-about.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "htmlwin-06-link-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "htmlwin-07-scrolled.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "htmlwin-08a-basic.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "htmlwin-08b-tables.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "htmlwin-08c-about.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "import-settings-modal-stack.png",
"engine": "chromium"
},
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
{
"name": "infobar-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "infobar-02-info.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "infobar-03-warning.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "infobar-04-error.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "infobar-05-dismiss.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "layout-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "layout-02-splitter.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "layout-03-sash-dragged.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "layout-04-scrolled-left.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "layout-05-scrolled-right.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "layout-06-combined.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "listctrl-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "listctrl-02-virtual.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "listctrl-03-columns.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "listctrl-04-selected.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "listctrl-05-scrolled.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "load-pcb-microwave-00-pcbnew-ready.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "load-pcb-microwave-01-dialog-open.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "load-pcb-microwave.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "load-pcb-pic_programmer-00-pcbnew-ready.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "load-pcb-pic_programmer-01-dialog-open.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "load-pcb-pic_programmer.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "logerror-01-initial.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "logerror-02-before-error.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "logerror-03-single-error-dialog.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "logerror-04-multiple-errors-dialog.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "logerror-05-mixed-levels.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "maximize-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "maximize-02-fullscreen.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "maximize-03-canvas.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "menu-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "menu-02-menubar.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "menu-03-file-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "menu-04-edit-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "menu-05-all-menus.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "mobile-chrome.png",
"engine": "chromium"
},
{
"name": "mobile-pan-00-base.png",
"engine": "chromium"
},
{
"name": "mobile-pan-01-panned.png",
"engine": "chromium"
},
{
"name": "mobile-pan-02-restored.png",
"engine": "chromium"
},
{
"name": "mobile-pinch-00-base.png",
"engine": "chromium"
},
{
"name": "mobile-pinch-01-zoomed.png",
"engine": "chromium"
},
{
"name": "mobile-pinch-02-restored.png",
"engine": "chromium"
},
{
"name": "mobile-toggle-rehidden.png",
"engine": "chromium"
},
{
"name": "mobile-toggle-shown.png",
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "modal-01-border.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "modal-02-before-drag.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "modal-03-after-drag.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "modal-04-before-resize.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "modal-05-after-resize.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "notebook-01-scrolled.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "notebook-02-scrolled-again.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "occ-export-dialog.png",
"engine": "chromium"
},
{
"name": "occ-export-done.png",
"engine": "chromium"
},
{
"name": "occ-probe-done.png",
"engine": "chromium"
},
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
{
"name": "ownerdrawn-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "ownerdrawn-02-layer.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "ownerdrawn-03-font.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "ownerdrawn-04-icon.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "ownerdrawn-05-log.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pcbnew-context-menu.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pcbnew-context-submenu.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pcbnew-dark-mode-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
{
"name": "pcbnew-draw-lines-00-before-tool-click.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pcbnew-draw-lines-01-after-click.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pcbnew-draw-lines-02-after-drawing.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pcbnew-loaded-css.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pcbnew-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pcbnew-move-00-before.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "pcbnew-move-01-after.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
{
"name": "pcbnew-sidebar-scrollbar-dragged.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pcbnew-sidebar-scrollbar.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pickers-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pickers-02-colors.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pickers-03-font.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pickers-04-preview.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pl_editor-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pl_editor-02-no-wizard.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pl_editor-03-file-menu.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pl_editor-04-save-as-dialog.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pl_editor-04b-after-enter.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pl_editor-context-menu.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pl_editor-load-rendered.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pl_editor-scrollbar-dragged.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "pl_editor-scrollbar.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "popup-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "popup-02-status.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "popup-03-palette-dismissed.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "popup-03-palette-open.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "popup-04-color.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "popup-05-positioning.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "popup-06-log.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "print-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "print-02-preview-panel.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "print-03-browser-print-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "print-04-preview-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "print-05-page-setup-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "print-06-print-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "print-07-callbacks.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "print-08-all-buttons.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "printpreview-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "printpreview-02-preview-area.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "printpreview-03-preview-frame.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "printpreview-04-page-setup.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "printpreview-05-settings.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "probe-00-after-wizard.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "probe-01-file-menu-open.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "probe-02-after-open-click.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "probe-03-late.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "project-sync-00-board-open.png",
"engine": "chromium"
},
{
"name": "project-sync-01-update-dialog.png",
"engine": "chromium"
},
{
"name": "project-sync-02-resync-first-R777.png",
"engine": "chromium"
},
{
"name": "project-sync-03-resync-second-R888.png",
"engine": "chromium"
},
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
{
"name": "propgrid-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "propgrid-02-categories.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "propgrid-03-selected.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "propgrid-04-manager.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "radiogroups-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "radiogroups-02-selections.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "regions.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "scrollbar-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "scrollbar-02-dragged.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "searchctrl-01-visible.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "searchctrl-02-typing.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "searchctrl-03-after-enter.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "selectheight-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "selectheight-02-result.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "specialized-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "specialized-02-treebook-initial.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "specialized-03-treebook-subpage.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "specialized-04-treebook-expand.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "specialized-05-bitmapcombo-dropdown.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "specialized-06-bitmapcombo-select.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "specialized-07-rearrange-list.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "specialized-08-get-order.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "spinctrl-01-visible.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "spinctrl-02-after-click.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "stc-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "stc-02-python-default.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "stc-03-drc-mode.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "stc-04-plain-mode.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "stc-05-insert-sample.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "stc-06-cleared.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "stc-07-line-numbers-toggle.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "stc-08-folded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "stc-09-typed.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "stc-10a-python.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "stc-10b-drc.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "stc-10c-python-again.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "symbol_editor-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "symbol_editor-02-metrics.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "textdecor.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "threadpool-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "timer-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "timer-02-started.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "timer-03-ticked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "timer-04-stopped.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "timer-05-fast-started.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "timer-06-fast-running.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "timer-07-fast-stopped.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "timer-08-reset.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "toolbar-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "toolbar-02-buttons.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "toolbar-03-new-clicked.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "toolbar-04-zoom.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "toolbar-05-toggle-on.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "toolbar-06-toggle-off.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "toolbar-07-statusbar.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "toolbar-08-all-buttons.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "tree-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "tree-02-hierarchy.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "tree-03-selected.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "tree-04-expanded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "tree-05-collapsed.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "tree-06-added.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "tree-07-deleted.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "validators-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "validators-02-text.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "validators-03-integer.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "validators-04-float.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "validators-05-netname.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "validators-06-button.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wasmedge-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wasmedge-02-filesystem.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wasmedge-03-threading.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wasmedge-04-fonts.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wasmedge-05-clipboard.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wasmedge-06-memory.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wasmedge-07-runall.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wasmedge-08-log.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "web-calculator.png",
"engine": "chromium"
},
{
"name": "web-chrome-hidden.png",
"engine": "chromium"
},
{
"name": "web-chrome-restored.png",
"engine": "chromium"
},
{
"name": "web-eeschema.png",
"engine": "chromium"
},
{
"name": "web-footprint_editor.png",
"engine": "chromium"
},
{
"name": "web-gerbview.png",
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"name": "web-pcbnew.png",
"engine": "chromium"
},
{
"name": "web-pl_editor.png",
"engine": "chromium"
},
{
"name": "web-read-only-viewer.png",
"engine": "chromium"
},
{
"name": "web-switch-missing-pcb-created.png",
"engine": "chromium"
},
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
{
"name": "web-switch-pcb-to-sch.png",
"engine": "chromium"
},
{
"name": "web-switch-sch-to-pcb.png",
"engine": "chromium"
},
{
"name": "web-symbol_editor.png",
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wizard-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wizard-02-launch.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wizard-03-next-page.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wizard-04-back-page.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wizard-05-cancel.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wxgrid-01-tab.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wxgrid-02-selection.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wxgrid-03-editing.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wxgrid-controls.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "wxgrid-dedicated-page.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "xface-preferences.png",
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "xml-01-loaded.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "xml-02-input.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "xml-03-parse.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "xml-04-traverse.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "xml-05-create.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
},
{
"name": "xml-06-results.png",
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
"engine": "chromium"
},
{
"name": "3d-viewer-00-board-loaded.png",
"engine": "firefox"
},
{
"name": "3d-viewer-models-pic_programmer-render.png",
"engine": "firefox"
},
{
"name": "3d-viewer-models-pic_programmer.png",
"engine": "firefox"
},
{
"name": "3d-viewer-pic_programmer-render.png",
"engine": "firefox"
},
{
"name": "3d-viewer-pic_programmer.png",
"engine": "firefox"
},
{
"name": "3d-viewer-titlebar.png",
"engine": "firefox"
},
{
"name": "appearance-00-layers.png",
"engine": "firefox"
},
{
"name": "appearance-01-objects.png",
"engine": "firefox"
},
{
"name": "appearance-02-nets.png",
"engine": "firefox"
},
{
"name": "appearance-03-layers-again.png",
"engine": "firefox"
},
{
"name": "appearance-10-layers-scrolled.png",
"engine": "firefox"
},
{
"name": "appearance-11-layers-scrolled-back.png",
"engine": "firefox"
},
{
"name": "appearance-20-objects-scrolled.png",
"engine": "firefox"
},
{
"name": "calculator-before-switch.png",
"engine": "firefox"
},
{
"name": "calculator-color-code.png",
"engine": "firefox"
},
{
"name": "calculator-loaded.png",
"engine": "firefox"
},
{
"name": "coroutine-01-loaded.png",
"engine": "firefox"
},
{
"name": "coroutine-02-summary.png",
"engine": "firefox"
},
{
"name": "coroutine-nested-01-loaded.png",
"engine": "firefox"
},
{
"name": "coroutine-nested-02-summary.png",
"engine": "firefox"
},
{
"name": "eefpsel-01-boot.png",
"engine": "firefox"
},
{
"name": "eefpsel-02-chooser.png",
"engine": "firefox"
},
{
"name": "eefpsel-03-selected.png",
"engine": "firefox"
},
{
"name": "eeschema-crosshair-00-small.png",
"engine": "firefox"
},
{
"name": "eeschema-crosshair-01-full.png",
"engine": "firefox"
},
{
"name": "eeschema-crosshair-02-45.png",
"engine": "firefox"
},
{
"name": "eeschema-crosshair-03-small-again.png",
"engine": "firefox"
},
{
"name": "eeschema-draw-wires-00-before-tool-click.png",
"engine": "firefox"
},
{
"name": "eeschema-draw-wires-01-after-click.png",
"engine": "firefox"
},
{
"name": "eeschema-draw-wires-02-after-drawing.png",
"engine": "firefox"
},
{
"name": "eeschema-load-rendered.png",
"engine": "firefox"
},
{
"name": "eeschema-loaded-css.png",
"engine": "firefox"
},
{
"name": "eeschema-loaded.png",
"engine": "firefox"
},
{
"name": "eeschema-sim-frame.png",
"engine": "firefox"
},
{
"name": "eeschema-sim-plot.png",
"engine": "firefox"
},
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
{
"name": "eeschema-url-regex.png",
"engine": "firefox"
},
{
"name": "frame-runtime-eeschema.html.png",
"engine": "firefox"
},
{
"name": "frame-runtime-footprint_editor.html.png",
"engine": "firefox"
},
{
"name": "frame-runtime-pcbnew.html.png",
"engine": "firefox"
},
{
"name": "frame-runtime-symbol_editor.html.png",
"engine": "firefox"
},
{
"name": "gerbview-01-loaded.png",
"engine": "firefox"
},
{
"name": "gerbview-02-metrics.png",
"engine": "firefox"
},
{
"name": "gerbview-print-00-loaded.png",
"engine": "firefox"
},
{
"name": "gerbview-print-01-dialog.png",
"engine": "firefox"
},
{
"name": "gerbview-print-02-reopened.png",
"engine": "firefox"
},
{
"name": "import-settings-modal-stack.png",
"engine": "firefox"
},
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
{
"name": "load-pcb-microwave-00-pcbnew-ready.png",
"engine": "firefox"
},
{
"name": "load-pcb-microwave-01-dialog-open.png",
"engine": "firefox"
},
{
"name": "load-pcb-microwave.png",
"engine": "firefox"
},
{
"name": "load-pcb-pic_programmer-00-pcbnew-ready.png",
"engine": "firefox"
},
{
"name": "load-pcb-pic_programmer-01-dialog-open.png",
"engine": "firefox"
},
{
"name": "load-pcb-pic_programmer.png",
"engine": "firefox"
},
{
"name": "occ-export-dialog.png",
"engine": "firefox"
},
{
"name": "occ-export-done.png",
"engine": "firefox"
},
{
"name": "occ-probe-done.png",
"engine": "firefox"
},
{
"name": "pcbnew-context-menu.png",
"engine": "firefox"
},
{
"name": "pcbnew-context-submenu.png",
"engine": "firefox"
},
{
"name": "pcbnew-dark-mode-loaded.png",
"engine": "firefox"
},
{
"name": "pcbnew-draw-lines-00-before-tool-click.png",
"engine": "firefox"
},
{
"name": "pcbnew-draw-lines-01-after-click.png",
"engine": "firefox"
},
{
"name": "pcbnew-draw-lines-02-after-drawing.png",
"engine": "firefox"
},
{
"name": "pcbnew-loaded-css.png",
"engine": "firefox"
},
{
"name": "pcbnew-loaded.png",
"engine": "firefox"
},
{
"name": "pcbnew-move-00-before.png",
"engine": "firefox"
},
{
"name": "pcbnew-move-01-after.png",
"engine": "firefox"
},
{
"name": "pcbnew-sidebar-scrollbar-dragged.png",
"engine": "firefox"
},
{
"name": "pcbnew-sidebar-scrollbar.png",
"engine": "firefox"
},
{
"name": "pl_editor-01-loaded.png",
"engine": "firefox"
},
{
"name": "pl_editor-02-no-wizard.png",
"engine": "firefox"
},
{
"name": "pl_editor-03-file-menu.png",
"engine": "firefox"
},
{
"name": "pl_editor-04-save-as-dialog.png",
"engine": "firefox"
},
{
"name": "pl_editor-04b-after-enter.png",
"engine": "firefox"
},
{
"name": "pl_editor-context-menu.png",
"engine": "firefox"
},
{
"name": "pl_editor-load-rendered.png",
"engine": "firefox"
},
{
"name": "pl_editor-scrollbar-dragged.png",
"engine": "firefox"
},
{
"name": "pl_editor-scrollbar.png",
"engine": "firefox"
},
{
"name": "probe-00-after-wizard.png",
"engine": "firefox"
},
{
"name": "probe-01-file-menu-open.png",
"engine": "firefox"
},
{
"name": "probe-02-after-open-click.png",
"engine": "firefox"
},
{
"name": "probe-03-late.png",
"engine": "firefox"
},
{
"name": "project-sync-00-board-open.png",
"engine": "firefox"
},
{
"name": "project-sync-01-update-dialog.png",
"engine": "firefox"
},
{
"name": "project-sync-02-resync-first-R777.png",
"engine": "firefox"
},
{
"name": "project-sync-03-resync-second-R888.png",
"engine": "firefox"
},
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
{
"name": "symbol_editor-01-loaded.png",
"engine": "firefox"
},
{
"name": "symbol_editor-02-metrics.png",
"engine": "firefox"
},
{
"name": "web-calculator.png",
"engine": "firefox"
},
{
"name": "web-chrome-hidden.png",
"engine": "firefox"
},
{
"name": "web-chrome-restored.png",
"engine": "firefox"
},
{
"name": "web-eeschema.png",
"engine": "firefox"
},
{
"name": "web-footprint_editor.png",
"engine": "firefox"
},
{
"name": "web-gerbview.png",
"engine": "firefox"
},
{
"name": "web-pcbnew.png",
"engine": "firefox"
},
{
"name": "web-pl_editor.png",
"engine": "firefox"
},
{
"name": "web-read-only-viewer.png",
"engine": "firefox"
},
{
"name": "web-switch-missing-pcb-created.png",
"engine": "firefox"
},
e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate Squash of experiment/ff-big-modules vs main. Big-module routing removed: native-EH shrank kicad_editor below SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT crutch are gone — kicad-firefox and kicad-chromium both run the full suite, with the module compiled the way real users' browsers compile it. Per-engine screenshots end to end: stableShot/shotPath write test-results/<engine>/<name>.png; baselines move to baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on <engine>/<name>. Previously Firefox and Chromium renders of one spec overwrote each other and Firefox renders were never actually gated. Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium web-suite shots); manifest generated from the baseline tree. One merged playwright.config.ts (kicad/asyncify/coroutine/perf as projects); ~25 dead npm scripts dropped. The web suite is gated in CI for the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after npm ci; last 26 blind-sleep violations fixed. SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe (--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader transiently failed the first post-board-load draw and the recovery cascade ended in a silent permanent Cairo fallback — that engine flip was the "~1.2% changedRatio both directions" occ-export baseline flake. Validated 160/160 across two 80-repeat rigs; full analysis in docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium baselines shift slightly on llvmpipe — promote once from the first green run. Deflakes the new coverage exposed: presence baselines settle before capture; presence fixtures declare current file formats; perf gets its own outputDir so CI evidence survives; occ-export settles the board paint before the export dialog; menu-item waits (waitForRenderedByLabel before clickMenuItem) in 4 specs + the TESTING.md rule. Web suite runs the PROD build, in parallel: webServer becomes backend `start` + the standalone's e2e:preview (build-preview.mjs: link-wasm → stash the public/wasm symlink aside during vite build, build-demo.mjs's move — then vite preview as the persistent server). The wasm middleware serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread worker script's own response must carry COEP or Chrome kills it with ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time; VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed: shared-page specs become serial groups; locks.spec grabs alice's exact item via the new kicadCollabTestSelectByUuid hook (cross-tab "first footprint" order is not a ysync invariant); quit specs poll page.url() (quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox). Suite: 51 passed / 12 skipped / 0 failed in 1.6m. CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be reachable from the npm scripts the workflows invoke — scraped from .github/workflows/, resolved through package.json, coverage asked from playwright --list itself. Rules: uncovered-spec + orphan-project (with a documented LOCAL_ONLY_PROJECTS allowlist). Gating next to lint:determinism; 138 spec files / 13 projects accounted for. Product fixes kept from the investigations (reachable on real GPUs too): wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush + per-instance font texture + first-frame GL-error drain (GAL recovery recovers instead of falling back to Cairo) + the user-facing eeschema switch navigates again under __EMSCRIPTEN__ (project-sync's FaceRegistered gate had rerouted it into the hidden sync player; caught by the newly-gated web suite). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
2026-07-17 12:10:40 +02:00
{
"name": "web-switch-pcb-to-sch.png",
"engine": "firefox"
},
{
"name": "web-switch-sch-to-pcb.png",
"engine": "firefox"
},
{
"name": "web-symbol_editor.png",
"engine": "firefox"
},
{
"name": "xface-preferences.png",
"engine": "firefox"
feat(tests): screenshot regression + Discord review, perf-tracked New tooling in tests/tools/screenshots/ (TypeScript via tsx): - compare.ts: one pixelmatch engine (AA-excluded), connected-component "where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors. - promote.ts: churn-free updater — overwrite a baseline only when decoded pixels differ beyond the floor, copying CI bytes verbatim (no re-encode churn); pulls a CI run via `gh run download` or a local --from dir. - post-discord.ts: always-on CI-on-main report (SHA + e2e status + the track-only runtime-perf table), then screenshot triptychs, batched + size-capped + flood-collapsed + 429-aware. - perf-report.ts: perf table with Δ vs the previous main run (via gh). - changelog.ts: no-build git-history baseline differ (Discord trigger B). - noise.ts / gen-manifest.ts: calibration + manifest generation. CI wiring: - wasm-build.yml: post-test step runs the gate + report on the already- produced test-results (no extra build); report-only (continue-on-error), posts only on push to main, inert without DISCORD_WEBHOOK_URL. - ci-ubicloud.yml: secrets: inherit (pass the webhook through). - screenshot-changelog.yml: ~30s no-build changelog on baseline changes. screenshot-manifest.json: canonical 354-name set + best-effort engine tags (313 chromium-swiftshader / 41 firefox-llvmpipe). Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1) so committed baselines are uniformly css-scaled. Design: CI's Linux render is the single source of truth; no pinned container (accept rare env drift -> re-promote); dev commits via promote. Replaces the byte-cmp compare-screenshots.sh + file-size-proxy update-baseline-screenshots.sh (kept for now until the first re-baseline). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 18:01:15 +02:00
}
]
}