e2e/CI: dual-engine suites, per-engine screenshots, SwiftShader retired, prod web suite, CI-coverage gate

Squash of experiment/ff-big-modules vs main.

Big-module routing removed: native-EH shrank kicad_editor below
SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on
stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT
crutch are gone — kicad-firefox and kicad-chromium both run the full
suite, with the module compiled the way real users' browsers compile it.

Per-engine screenshots end to end: stableShot/shotPath write
test-results/<engine>/<name>.png; baselines move to
baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots
pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on
<engine>/<name>. Previously Firefox and Chromium renders of one spec
overwrote each other and Firefox renders were never actually gated.
Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium
web-suite shots); manifest generated from the baseline tree.

One merged playwright.config.ts (kicad/asyncify/coroutine/perf as
projects); ~25 dead npm scripts dropped. The web suite is gated in CI for
the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs
triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after
npm ci; last 26 blind-sleep violations fixed.

SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe
(--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag
is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable
without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader
transiently failed the first post-board-load draw and the recovery
cascade ended in a silent permanent Cairo fallback — that engine flip was
the "~1.2% changedRatio both directions" occ-export baseline flake.
Validated 160/160 across two 80-repeat rigs; full analysis in
docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium
baselines shift slightly on llvmpipe — promote once from the first green
run. Deflakes the new coverage exposed: presence baselines settle before
capture; presence fixtures declare current file formats; perf gets its
own outputDir so CI evidence survives; occ-export settles the board paint
before the export dialog; menu-item waits (waitForRenderedByLabel before
clickMenuItem) in 4 specs + the TESTING.md rule.

Web suite runs the PROD build, in parallel: webServer becomes backend
`start` + the standalone's e2e:preview (build-preview.mjs: link-wasm →
stash the public/wasm symlink aside during vite build, build-demo.mjs's
move — then vite preview as the persistent server). The wasm middleware
serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread
worker script's own response must carry COEP or Chrome kills it with
ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time;
VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default
workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed:
shared-page specs become serial groups; locks.spec grabs alice's exact
item via the new kicadCollabTestSelectByUuid hook (cross-tab "first
footprint" order is not a ysync invariant); quit specs poll page.url()
(quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox).
Suite: 51 passed / 12 skipped / 0 failed in 1.6m.

CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be
reachable from the npm scripts the workflows invoke — scraped from
.github/workflows/, resolved through package.json, coverage asked from
playwright --list itself. Rules: uncovered-spec + orphan-project (with a
documented LOCAL_ONLY_PROJECTS allowlist). Gating next to
lint:determinism; 138 spec files / 13 projects accounted for.

Product fixes kept from the investigations (reachable on real GPUs too):
wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always
propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush +
per-instance font texture + first-frame GL-error drain (GAL recovery
recovers instead of falling back to Cairo) + the user-facing eeschema
switch navigates again under __EMSCRIPTEN__ (project-sync's
FaceRegistered gate had rerouted it into the hidden sync player; caught
by the newly-gated web suite).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq

docs/features/web-e2e-rot/01-editor-lib-bridge-flows.md

@@ -0,0 +1,57 @@
+# Web-suite rot: footprint/symbol editor library-bridge flows (5 specs)
+
+Surfaced 2026-07-15 by wiring the web suite into CI for the first time
+(branch `experiment/ff-big-modules`, runs 29414003275 / 29415027412). These
+specs were green when written and have silently rotted since — the web suite
+never ran in CI, so nothing caught it.
+
+## Affected specs (marked `test.fixme` — flip back when fixed)
+
+| spec | first-broken symptom (local, web-chromium) | CI symptom |
+|---|---|---|
+| `web/footprint-browse-remote.spec.ts` | boots, lib tree lists libs, but expanding `Resistor_SMD` never produces child rows (FootprintEnumerate yields nothing) | `#canvas` never visible in 180 s |
+| `web/footprint-write-remote.spec.ts` | boots; New Footprint + Ctrl+S never lands an item in the backend (`/api/scopes/default/libs/<lib>/items` stays empty 30 s) | same boot timeout |
+| `web/footprint-write-spike.spec.ts` | boots (route fixed); `window.__pcbjamSaved` never captures a body after New Footprint + save (`?fpwrite=1` spike provider silent) | same boot timeout |
+| `web/symbol-write-remote.spec.ts` | same family as footprint-write-remote, symbol domain | same boot timeout |
+| `web/symbol-write-spike.spec.ts` | same family as footprint-write-spike (`?libwrite=1`) | same boot timeout |
+
+## Evidence pointing at one root cause
+
+All five live in the same domain: `window.kicadLibs.request(...)` traffic from
+the footprint/symbol EDITOR tools (enumerate / save). Probing a booted
+`/default/projects/demo/-/footprint_editor` locally:
+
+- the lib tree lists the origin libs (Capacitor_SMD, Diode_SMD, LED_SMD,
+  My Symbols, Resistor_SMD) — the *list* path works (pre-sync/IDB);
+- `__libsCalls` (a wrapper capturing every `kicadLibs.request`) records ZERO
+  calls during boot and ZERO on expanding a lib — the per-lib
+  enumerate/get/save traffic never happens;
+- one `[pageerror] __name is not defined` fires at boot — esbuild's
+  keep-names helper missing in whatever context executes a bundled callback;
+  prime suspect for the provider dying silently on first use.
+
+By contrast the SCHEMATIC editor's bridge works (eeschema-fp-selector records
+`index` calls), and eeschema/pcbnew tool pages boot and pass their specs — the
+rot is specific to the fp/sym editor lib flows, not the bridge as a whole.
+
+Separately fixed while triaging (NOT part of this bug): dead `/p/<slug>/<tool>/`
+routes in 3 of these specs (router grammar is `/:scope/projects/:name/-/:tool`
+since the scope/kind/name change), the tool-switch overlay race + `/p/` URL
+asserts, and read-only-editor's cross-tab SelectFirst-order assumption.
+
+## Why fixme and not expected-fail
+
+The ysync convention (expected-fail repros) is right for fast unit-level
+repros. These five die in 180 s boot timeouts on CI — as `test.fail()` they
+would burn ~30 min of CI per run across two engines for zero extra signal.
+`test.fixme` keeps them visible in every report as skipped-with-reason;
+remove the marker (and delete this table row) when the bridge flow is fixed.
+
+## CI-only sibling (kept RUNNING, conditional expected-fail)
+
+`web/eeschema-fp-selector.spec.ts` completes its flow but records a
+`[pageerror] index out of bounds` wasm trap on CI (BOTH engines,
+llvmpipe/SwiftShader) that does not reproduce on a Mac with real GL — see
+`test.fail(!!process.env.CI, …)` in the spec. Likely the same software-GL
+render-path family as the presence ghost-wipe (SwiftShader pass-boundary
+flush, webgl_gal.cpp). Needs its own investigation on a CI-like rig.

docs/features/wx-parity-bugs/occ-export-context-eviction.md

@@ -0,0 +1,119 @@
+# occ-export flake: SwiftShader congestion → GAL recovery cascade (fixed by GL-stack swap)
+
+Status 2026-07-16: fully root-caused with page-level canvas/context telemetry +
+temp C++ instrumentation in a local docker rig. **Final remedy: CI Chromium
+now runs WebGL on ANGLE → Mesa llvmpipe instead of SwiftShader**
+(`tests/playwright.config.ts`, `CHROMIUM_CI_ARGS`) — the same software-GL
+stack the Firefox projects use, where none of this ever happened. Four small
+product fixes found along the way were kept (below); everything else was
+deliberately **not** hardened — the trigger cannot occur outside congested
+SwiftShader.
+
+## Symptom
+
+`chromium/occ-export-{dialog,done}.png` bistable on CI only: the board behind
+the modal STEP-export dialog flips between runs (~1.2% changedRatio both
+directions). Firefox (llvmpipe) never affected.
+
+## The causal chain (every step observed in the rig)
+
+1. **Transient SwiftShader failure at the first post-board-load draw**
+   (~10-15% of instances at `REPEAT=40-80 WORKERS=4`): `DoRePaint threw:
+   Requested render buffer size is not supported`, plus LRU context evictions
+   the moment a 3rd context appears (21/80 runs; the usual victim is a leaked
+   hidden 881×159 startup GAL canvas, so the `CONTEXT_LOST_WEBGL` line alone
+   was harmless). No flag stops it — `--max-active-webgl-contexts=64`,
+   `--force-gpu-mem-available-mb=4096`, `--disable-low-end-device-mode`,
+   `--disable-gpu-driver-bug-workarounds` all verified present and useless.
+
+2. **`recoverFromGalError` attempt 1** (`common/draw_panel_gal.cpp:205`)
+   destroys + recreates the WebGL GAL (board canvas `glcanvas-2` →
+   `glcanvas-3`).
+
+3. **Attempt 1 then always failed** on stale cross-context GL state:
+   - static `g_fontTexture` bound on the new context → `INVALID_OPERATION:
+     bindTexture: object does not belong to this context` → `checkGlError`
+     throw (**fixed — kept**: per-instance `m_fontTexture`,
+     `common/gal/webgl/webgl_gal.{cpp,h}`);
+   - sticky GL error flags left by the old GAL's teardown (its `LockCtx`
+     targets a lost context, so the deletes land on the current one) →
+     the fresh GAL's first `checkGlError` ("generating vertices buffer:
+     invalid enum") misattributed the leftover and threw (**fixed — kept**:
+     drain `glGetError()` before first-frame `init()` in
+     `WEBGL_GAL::BeginDrawing`). With both fixes, reinit-on-WebGL succeeds
+     (validated: 0 Cairo fallbacks / 60 after, vs 100% of recoveries before).
+
+4. **Attempt 2 = silent permanent Cairo fallback** (`GAL_FALLBACK ==
+   GAL_TYPE_CAIRO` on emscripten, dialog suppressed). Cairo renders the board
+   through the wx software path — shots LOOK painted but are a different
+   engine → subtle whole-board diffs vs the WebGL baseline. **This engine
+   flip IS the original "~1.2% changedRatio both directions" flake.**
+   `FULLSCREEN_QUAD` (singleton VAOs) is the same static-GL-state disease and
+   produced `bindVertexArray: object does not belong` + one wasm abort in
+   rare runs — NOT fixed (unreachable once recovery never triggers; noted
+   here in case in-app recovery robustness is ever wanted).
+
+5. **wx paint-flag races turned it into a stuck-blank board** (both
+   **fixed — kept**, `wxwidgets/src/wasm/window.cpp`):
+   - `Invalidate` early-out skipped the upward walk while the sweep can
+     strand a window "flags set, ancestors clear" → every later `Refresh()`
+     swallowed. Fix: always propagate to the top.
+   - `PaintSelf`/`PaintChildren` cleared the dirty flags AFTER dispatching
+     paint, so an `Invalidate` issued reentrantly from inside a paint handler
+     (canvas recreated mid-frame, board-load refresh mid-sweep) was clobbered
+     with nothing re-issuing it. Fix: `DoPaint` snapshots + clears both flags
+     BEFORE dispatching. (Pre-fix rigs showed stuck half-painted toolbars —
+     this is a real production-reachable bug, not a CI artifact.)
+
+   One rare mode ("grid-mode": board data loaded, UI painted, canvas alive,
+   items never drawn, zoom-to-fit doesn't help; VIEW had all 464 items and
+   the same Clear/DisplayBoard ordering as healthy runs) was still being
+   diagnosed when the GL-stack swap made it — and everything above —
+   unobservable: **80/80 runs clean on llvmpipe**, all dialog shots pixel-
+   identical (single 55.65% ratio value across all 80; SwiftShader scattered
+   over 55.5–56.8%).
+
+## The fix that ships
+
+`CHROMIUM_CI_ARGS` = `--use-gl=angle --use-angle=gl --ignore-gpu-blocklist`
+(CI-gated). Requirements, all already present on CI: an X display
+(`xvfb-run -a` wraps the whole e2e step) and Mesa (installed by
+`playwright install --with-deps`). `--ignore-gpu-blocklist` is mandatory —
+llvmpipe is on Chromium's software-GL blocklist and WebGL silently reports
+unavailable without it (the exact analog of the Firefox projects'
+`webgl.force-enabled`). Headless works; headed is NOT needed.
+Renderer-string ground truth (Chrome falls back silently — always verify):
+`gl.getExtension('WEBGL_debug_renderer_info')` must report
+`ANGLE (Mesa, llvmpipe …)`, not SwiftShader.
+
+Consequence: chromium renders shift slightly → the chromium baseline set
+must be re-promoted once from the first green CI run
+(`cd tests && npm run screenshots:promote -- --run <id>`).
+
+## Test-side hardening (kept, independent)
+
+Popup menu items register in the element registry progressively while the
+menu paints; a coarse ">3 menuitems" gate + single-shot `clickMenuItem` races
+(4–8/20 under contention, pre-existing). Rule (tests/TESTING.md): wait for the
+specific item via `waitForRenderedByLabel(page, label, { elementType:
+'menuitem' })` before every `clickMenuItem`. Applied to `occ-export`,
+`occ-export-models`, `3d-viewer-models`, `load-pcb`; `pl_editor` already did
+this; `clickMenuItemByText` waits internally.
+
+## Open leads (not blocking)
+
+- The leaked hidden 881×159 startup GAL canvas: destroying it would free a
+  context + memory.
+- If in-app GL-error recovery ever matters on real GPUs: fix the
+  `FULLSCREEN_QUAD` singleton VAOs (per-instance like the font texture) and
+  audit remaining cross-context statics.
+
+## Repro rig (for revalidation)
+
+Docker `mcr.microsoft.com/playwright:v1.57.0-jammy`, worktree bind-mount,
+container-local node_modules volume, `CI=1`, `xvfb-run -a npx playwright test
+--project=kicad-chromium --repeat-each=N --workers=4` with a temp spec that
+loads `pic_programmer`, screenshots board-load / settled / export-dialog /
+post-poke, and logs the WebGL renderer string + canvas/context telemetry per
+instance. Canaries: `CONTEXT_LOST_WEBGL`, `does not belong to this context`,
+board-strip pixel ratio of the shots.