Pipeline: lexer generation (pcbcommon) ordered before the embind
pre-compile — a fresh-build-tree chicken-and-egg any new volume would
hit (generated pcb_lexer.h was included via -I${KICAD_BUILD}/common
before anything generated it). DOM flavor outputs flow through the
host-mounted output/dom/. pcbnew e2e on DOM: editor loads fully (layer
manager, toolbars, WebGL viewport, registry metrics), draw-lines test
passes; the header pixel-reference assertion fails as expected (native
DOM styling vs univ pixels) — port-aware references pending.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
WX_PORT=dom flows host → docker container → build scripts:
build-kicad-target.sh selects wxwidgets-dom + kicad-<app>-dom trees;
docker/build.sh passes the env through, copies outputs to output-dom/
incl. wx-dom.js, and post-processes there; build-wxuniversal-wasm.sh
honors WX_PORT and accepts --no-clean again (the kicad pipeline calls it
with that flag — Phase 1's stricter parser had broken the canvas docker
path). setup-kicad-wasm.sh syncs DOM bundles into tests/apps-dom/kicad
and injects the wx-dom.js script tag into the kicad pages;
playwright-kicad.config.ts serves apps-dom under WX_PORT=dom.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fork delta vs wasm-port: 8 commits, ~107 files — port directory, 35
dispatch headers, build-system entries, 1-line prntbase.cpp condition,
2 generic-notebook bug fixes. No univ/aui/stc changes. Tracker-hook
migration deliberately deferred while the canvas port is alive (the DOM
port doesn't need the hooks; univ's die with univ). Patches in
features/feature/wx-dom-port/ via create-feature-patches.sh.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
tests/apps/Makefile.wasm: pre-js shims are link dependencies (editing
wx.js/wx-dom.js now relinks apps); $^ link rules filter to %.o/%.a so
the new deps stay out of wasm-ld inputs. Seed baseline-screenshots-dom
(251 screenshots) from the green WX_PORT=dom full-suite run (251/0).
Cross-port visual comparison notes in docs/features/wx-dom-port/
(10 bugs found by the protocol, all fixed in wxwidgets).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Build/test plumbing for the second wxWidgets mode: --dom on
build-wxuniversal-wasm.sh (build-wasm/wxwidgets-dom, no universal,
--disable-tooltips, libwx_wasmu_* stubs) and build-wasm-test.sh
(mirrors sources into tests/apps-dom, PORT=dom). Build script now
reconfigures when Makefile.in or autoconf_inc.m4 regenerate (bakefile
chain). Harness: WX_PORT=dom serves apps-dom, namespaces screenshots
into test-results/dom and logs into logs/wxwidgets/dom; compare/update
screenshot scripts take --port dom with baseline-screenshots-dom.
New port-agnostic boot.spec.ts (green on both ports).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The asyncify single-slot work, executed red-green (full ledger:
docs/features/asyncify-arbiter/redgreen.md; decisions record:
docs/features/async/07-decisions-and-outcome.md):
- tests/apps/standalone/asyncify-races/ + tests/asyncify/ + dedicated
playwright config: 8 scenarios reproducing the KiCad asyncify failure
family with the kicad-faithful startup topology (pre-park fiber swap →
park throw through the live trampoline). Built in 3 variants; the
SHIM_DISABLE_TRAMPOLINE_HEAL / SHIM_DISABLE_HANDLESLEEP ablation builds
keep the historical hang and index-out-of-bounds crash reproducible
forever (mutation-style pins for the existing shims).
- scripts/common/shims/handlesleep.js: catch the "unwind" park sentinel
in the wakeUp path — when main's last pre-park suspension was a sleep,
the main-loop park throw escaped through that sleep's promise reaction
as an uncaught rejection (the calculator/gerbview console errors).
- scripts/common/inject-dyncall-shims.sh: SHIM_DISABLE_* ablation knobs.
- Spec tightening (the acceptance bar): 'uncaught exception: unwind'
tolerance DELETED from pcbnew/eeschema specs; load-pcb gained a hard
clean-console gate over 5 asyncify corruption signatures.
- wxwidgets pointer bump: modal LIFO resolvers, pump resolve-on-error,
sync clipboard IsSupported (014f67e6c1).
Final state: asyncify suite 7/7, wx e2e 291/292 (1 skip), KiCad e2e 40
passed / 2 skipped with ZERO corruption signatures in any log across all
six apps.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two research dossiers under docs/features:
- async/: Asyncify currData contention — the clipboard crash and post-idle
tool hang as one disease (single global suspension slot, three producers),
asyncify internals walkthroughs, prior art, arbiter designs A/B.
- wasm-exceptions/: -fexceptions vs -fwasm-exceptions — measured invoke-driven
share of the asyncify tax on pcbnew (59% raw / 64% gzip; download 64.5->36 MB
if migrated), KiCad dialog-in-catch audit (85 direct / 93 review of 636,
catch_audit.py included), binaryen toolchain status (partial EH support since
v125; TryTable unsupported), and a catch-arm-hoisting pre-pass design that
would obsolete the KiCad-side refactor. Cross-linked with the parked
KICAD_WASM_EH end-to-end experiment (docs/wasm-exceptions-experiment.md).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- WasmTool: hard cutover to startKicadCollab (gates on
kicadCollabSnapshotItems; scalar reconciler no longer runs in the app)
- file-seed (ysync 0005 tie-in): empty rooms are seeded from the opened
file via fileToDoc/docToY (meta+layout+items — file recoverable from the
Y.Doc alone); populated rooms still adopt (seed-once unchanged)
- binding: gate UP applies until seed() — the provider's initial state
sync otherwise streams the full doc into an editor that already holds
the file (trapped eeschema's paste path in the real app)
- pcbnew blobForItem: footprints Format a uuid-corrected copy directly —
SaveSelection's copy regenerated mandatory-field uuids (FOOTPRINT copy
ctor assigns into fresh fields), breaking wire identity (rebuild)
- roundtrip.spec on the v2 items wire; new passing pcbnew
footprint-containment round trip (the 0004 containment win); full
fixture stays fixme on the tracked envelope-parse limit
- real-app two-tab emit verified for eeschema + pcbnew (see 0008 doc)
Full kicad suite: 45 passed, 3 skipped (known), 0 failed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add kicadCollabSnapshotItems / kicadCollabApplyItems / window.kicadCollab.onItems
to pl_editor, eeschema, pcbnew — per-item native-blob payloads
({added/changed:[{sexpr,parent}], removed:[uuid]}) alongside the untouched scalar
wire (legacy collab specs stay green).
- pl_editor: itemBlob per item; apply = SetPageLayout(append) + replace-by-uuid
(pointer-snapshot safe); bare payloads get the kicad_wks envelope. Snapshot +
apply + emit verified headless.
- eeschema: clipboard Format per item (symbols carry their lib_symbols); apply
mirrors the native paste — LoadContent into a throwaway sheet → detach →
replace-by-uuid → symbol lib relink (blob's lib_symbols first, live screen's
second) → SCH_COMMIT, in the CallAfter+COROUTINE context. Snapshot + apply
verified headless (a "lost" lone junction turned out to be correct connection
cleanup — test uses text).
- pcbnew: blobForItem per ROOT item with child→footprint lifting in flushDiff;
apply = makeFromBlob + commit replace-by-uuid on the fiber; bare non-footprint
payloads get wrapInBoardEnvelope (live board layer table). Snapshot + footprint
replace/add WITH children (the 0004 containment gap, closed) + removal verified
headless. Track/via/zone/text blob-apply hits the documented asyncify-fragile
envelope parse (reconfirmed empirically — a verbatim SaveSelection segment
envelope dies silently in the commit) and stays on the legacy scalar apply;
tracked in ysync 0008 status.
- eeschema/pcbnew scheduleFlush now runs flushDiff inside a COROUTINE: the
per-item Format in the v2 emit needs the fiber stack (0007 lesson). Their emit
remains unverifiable headless (both legacy two-tab tests are test.skip:
"open=false → SCH_COMMIT no-ops" / "harness can't PAINT") — verify in the real
app at Stage D; pl_editor's emit IS verified.
- tests/kicad/items-bridge.spec.ts: per-tool suite (snapshot uuids → local-edit
emit (where drivable) → apply changed/added/removed via save-readback → no
apply echo). 3/3 pass; roundtrip + collab suites unaffected.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The pcbjam repo's own git-feature skills + scripts/git-workflow predated the
web/pcbjam-shared submodule and only covered root+kicad+wxwidgets. Add the MIT
contract repo as a fourth member everywhere:
- repos.sh: pcbjam_shared key (path web/pcbjam-shared, main `main`) + layout doc.
- assert-clean.sh / for-each-repo.sh: count-agnostic wording.
- git-feature-start/commit/sync/finish SKILL.md: pcbjam-shared in the branch /
commit (kicad, wxwidgets, pcbjam-shared, then root) / rebase (root first) /
finish orders; root pointer-bump staging + diff now include web/pcbjam-shared,
and the root stage step excludes it like the other submodule trees.
Mirrors the root pcbjam-private skills, which already covered shared.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- standalone wasm/collab/types.ts: the CollabItem/CollabDelta data shapes now
come from MIT @pcbjam/shared (collab-wire zod schemas) via re-export; the
RUNTIME adapter interface (CollabBridge) stays here. emptyDelta/isEmptyDelta
alias the shared helpers — reconciler untouched.
- tests/collab/browser-entry.ts: fix the import path stale since the repo
restructure (web/apps/frontend → web/standalone) — the collab bundle had been
silently unbuildable; its opts type now tracks startCollab's real signature.
- tests/kicad/*-collab.spec.ts: update KicadCollab.start calls from the
pre-ysync-0004 { channel, settleMs } API to { provider: { kind:
"broadcastchannel", settleMs }, room } — the drift the broken bundle hid.
Verified: pl_editor 2/2, eeschema 2 passed, pcbnew 7 passed (skips pre-existing);
shared 47 unit tests; both typechecks clean.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Pointer bump for the MIT sexpr/kicad-doc converter + schema + round-trip suite;
record the vitest devDep in the web workspace lockfile.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Block A save embind exports + the round-trip integration harness.
Save exports (wasm/bindings, GPL):
- pcbnew: kicadSaveBoard(path) via PCB_IO_KICAD_SEXPR::SaveBoard(GetBoard()).
- eeschema: kicadSaveSchematic(path) via SCH_IO_KICAD_SEXPR::SaveSchematicFile.
Saves GetCurrentSheet().Last() (NOT Schematic().Root()): the wasm open-flow
nests the opened doc under an auto-created project root, so Root()'s screen
holds only a child-sheet symbol, not the loaded items.
(pl_editor already had kicadSaveDrawingSheet.)
Round-trip harness (tests/kicad/roundtrip.spec.ts):
- load fixture → save (ORIG) → kicadCollabSnapshot → reload (fresh wasm) →
open empty → kicadCollabApply → save (REGEN) → assert sexprDiff(ORIG,REGEN).equal.
- Two separate pages (extract closed before rebuild opens) so the process-global
wasm heap frees between boots (pcbnew ~190MB). Open both sides from the same
filename (eeschema embeds it as the root Sheetfile property). pcbnew boots the
seeded pcbnew-collab.html (plain pcbnew.html's first-run wizard blocks boot).
- pl_editor + eeschema round trips PASS (lossless). pcbnew is test.fixme with
tracked apply-coverage findings (footprints/zones not reconstructed; segment
width + via size lost; fp_text→gr_text) — bridge gaps for follow-up, not test bugs.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add the shared, uuid-keyed structural s-expr comparator (README §B) used by both
the live drift check (0003) and the round-trip tests (0004), plus a vitest setup.
- wasm/collab/sexpr-diff.ts: zero-dependency module (importable from standalone
app code and the Playwright specs). Tokenizes s-expr text into nested lists,
indexes item nodes by their direct (uuid "X") child, compares item sets order-
insensitively as multisets of normalized children, emitting
{ equal, added, removed, changed } with per-property a/b. ignoreTokens drops
volatile tokens. False-positive-free only for same-serializer inputs.
- wasm/collab/sexpr-diff.test.ts: 15 cases (parse, equality, changes, ignoreTokens).
- vitest config + test scripts (node env, isolated from the app build).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The automatic recovery uses an in-place location.replace reload, which does NOT
reliably release the OOM'd wasm heap in every browser (notably Firefox), so the
respawn can land right back in an out-of-memory state.
Add a user-gesture escape hatch to the terminal MemoryExhaustedDialog: an "Open
in a new tab" button that opens the editor in a fresh top-level browsing context
(clean retry chain) and closes the current tab. Because it runs from a real
click, window.open is not popup-blocked; window.close() may still be refused on
the original user-opened tab, which is fine — the new tab is the survivor. Copy
tells the user this sometimes helps where the in-place reload didn't.
- oom-watch.ts: export respawnInNewTab(win) — fresh-tab respawn with ?oomRetry
cleared.
- MemoryExhaustedDialog: "Open in a new tab" (primary) + "Reload this tab"
(secondary), with honest copy about Firefox + unsaved edits.
- WasmTool: wire onOpenNewTab to respawnInNewTab.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Automatically recover from an out-of-memory by respawning a fresh browsing
context (fresh wasm heap) and discarding the dead one — the manual "duplicate
tab, close the OOM'd one" fix, automated and capped at 2 retries so a too-small
machine stops instead of looping.
- recovery/oom-watch.ts: the recover() state machine with the retry counter in
the URL (?oomRetry=N, cap 2). Mode A soft detection — Module.onAbort plus
window error/unhandledrejection matched against an OOM signature, behind an
idempotent latch. Mode B hard-kill detection — a per-tab localStorage
heartbeat sentinel; a stale sentinel on the next load continues the retry
chain (heartbeat keys avoid mistaking a second live tab for a crash). Healthy
session (alive HEALTHY_RESET_MS) resets the chain. Default strategy is
location.replace; window.open+close is behind ?oomStrategy=newtab. Platform
access goes through injectable win/storage seams for testability.
- recovery/MemoryExhaustedDialog.tsx: terminal "out of memory" UI, reusing
0001's BlockingDialog; copy is honest about lost unsaved edits.
- wasm/boot.ts: forward emscripten Module.onAbort to the watcher (optional
BootOptions.onAbort).
- WasmTool: install the watcher, gate boot on its proceed flag (skip when the
chain is already exhausted), and render the terminal dialog.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a pre-boot "potato check" for the standalone editor: probe the device
against what the KiCad WASM tools actually require and warn (or block, with an
override) before the expensive WASM asset fetch.
- preflight/capabilities.ts: pure, unit-testable probe → CapabilityReport with
stable codes. Fatal: no SharedArrayBuffer/cross-origin-isolation, no
WebAssembly, no Workers+Atomics, no WebGL2. Warn: low deviceMemory, low JS
heap limit (Chromium), few cores, mobile, small screen. Thresholds in one
tunable const block.
- preflight/BlockingDialog.tsx: reusable modal blocking overlay (no dismiss
except via explicit action) — also for feature 0002's terminal dialog.
- preflight/PreflightGate.tsx: runs the probe once on mount; fatal → blocking
dialog with "Try anyway" (short-circuits the asset fetch); warnings →
dismissible advisory banner with localStorage "don't show again" keyed to the
exact warning codes.
- ToolPage: wrap WasmTool in PreflightGate so the gate runs before boot.
Optional micro-bench (phase 3) deferred per the spec.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ci-ubicloud.yml (cherry-picked from the ubicloud-ci experiment branch, green
on run 27375556034: 291 wx + 38 kicad tests in 1h52m) now triggers on main
pushes and replaces ci-full-build.yml and ci.yml. The Hetzner flow needed
create/delete-VM jobs, a PAT + hcloud token, and a shared one-slot
concurrency lock; the Ubicloud runner is a plain runs-on label and finished
faster end-to-end. wasm-opt-bench.yml stays (dispatch-only Hetzner bench
harness, unrelated to CI).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
build-gal-webgl-test.sh compiles kicad headers that need boost from
build-wasm/sysroot/include — a host dir on dev machines but inside the
kicad-build-cache docker volume on CI (run 27370316200: ptr_vector.hpp not
found). The Makefile uses only $(SYSROOT)/include, so headers suffice.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
First wx-suite run (27359020746): 262/262 non-GL tests passed; all 29
gal-webgl failures were galTest.isReady() timeouts because
gal_webgl_test.{js,wasm} are gitignored artifacts built by their own script
(build-gal-webgl-test.sh), which the workflow never ran — the page loaded
but the wasm 404'd. The script self-sources the local emsdk like
build-wasm-test.sh. Those 29 timeouts x3 attempts were also ~1.4h of the
1.7h e2e wall-clock.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Host-side wxUniversal build + wx test apps (build-wxuniversal-wasm.sh /
build-wasm-test.sh, mirroring the dispatch-only ci.yml) followed by
`npm run test` before the KiCad e2e. The wx suite runs on bundled headless
Chromium; several specs use WebGL (gal-webgl.spec.ts), so the chromium
project gains a CI-gated --enable-unsafe-swiftshader launch arg — without
it newer headless Chromium refuses software WebGL on the GPU-less runner.
First time this suite runs on a working CI runner; treat as experimental.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Same rationale as the Hetzner workflow change in 3168eff: headless Firefox
cannot create any GL context on the GPU-less runner.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Run 27330725978 died at container create: "range of CPUs is from 0.01 to
30.00, as there are only 30 CPUs available" with KICAD_DOCKER_CPUS=30 on the
30-vcpu standard-30. The Hetzner box (fresh docker-ce) accepts cpus == nproc;
Ubicloud's preinstalled daemon doesn't. A 29/30 CFS quota is noise.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Same recipe as ci-full-build.yml (all 6 tools, Binaryen 130 from source,
pipelined wasm-opt, KiCad e2e) but on ubicloud-standard-30: a plain runs-on
label, so no create/delete-Hetzner-VM jobs and no shared one-slot concurrency
lock. Push-triggered ONLY on this branch (plus workflow_dispatch) so it never
competes with the Hetzner main CI.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The xvfb/WebGL fix (f37ddc4) took CI from 22 to 11 failures (identical on
Hetzner 27343415331 and Ubicloud 27343416511) — everything left is the
pcbnew family, still dying at wasm compile: pcbnew's ~190M debug module
exceeds SpiderMonkey's per-process code budget on x86-64 even with the
baseline-only JIT (the same module compiles on arm64, whose denser code
fits — which is why the local arm64 repro passed). No pref raises that
limit; V8 handles the module fine.
New CI-only 'chromium-ci' project (bundled Chromium, headless, SwiftShader
via --enable-unsafe-swiftshader) carries pcbnew/pcbnew-collab/load-pcb/
load-pcb-probe; the firefox project ignores them on CI. Workflows call the
new test:kicad:ci script which runs both projects. Local test:kicad is
unchanged (firefox everywhere, system Chrome for headed runs).
Validated under x86 emulation (linux/amd64 playwright image on Rosetta):
the pcbnew wizard test passes on chromium-ci in 23s against the fresh
v130-postprocessed 192M module.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
build-kicad-target.sh swaps emsdk's wasm-opt/wasm-emscripten-finalize for
no-op stubs (real binaries preserved as *.real) so emcc skips in-link
asyncify — intended for the CONTAINER emsdk, but a host-mode run left the
host tools/emsdk stubbed (since Jun 9). The stub fakes --version and exits 0,
so every local build's host-side finalize/asyncify/-O2 "succeeded" while
doing nothing: output wasm shipped non-asyncified and aborts at boot with
"asyncify_stop_unwind is not a function" (and stayed 122M vs the correct
187M). get-wasm-opt.sh and apply-finalize.sh now prefer the *.real binary
whenever it exists, making the resolution immune to a stubbed emsdk.
Found while validating the CI e2e fix: pcbnew.wasm built this morning could
not boot in any browser. Local artifacts built since Jun 9 may need their
postprocess re-run (apply-finalize.sh + apply-asyncify.sh).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
CI runs (Hetzner 27329612719, Ubicloud 27330989479) failed 22/41 identically,
in two buckets, both firefox-environment issues (the suite is firefox-only:
test:kicad -> --project=firefox):
1. No WebGL: headless Firefox can't create ANY GL context on a GPU-less VM
(blocklist bypass still dies with FEATURE_FAILURE_WEBGL_EXHAUSTED_DRIVERS),
so the GAL canvas failed and an error dialog blocked every UI flow. Fix:
run headed under xvfb (GLX + Mesa llvmpipe software GL) with
webgl.force-enabled — workflows now wrap the e2e step in xvfb-run.
2. pcbnew wasm OOM: the ~190M module OOMs Firefox's optimizing wasm JIT at
compile ("InternalError: out of memory") and the app never boots. Fix:
javascript.options.wasm_optimizingjit=false (baseline-only compile).
Both prefs are CI-gated in playwright-kicad.config.ts; local runs keep stock
behavior. Boot got slower (baseline JIT), so the wizard helpers now wait for
the registry to have actual UI ENTRIES (the registry OBJECT exists pre-boot)
instead of clicking into a fixed 10x500ms window.
Validated in a local docker repro of the CI env (playwright:v1.57.0 image,
no GPU): pl_editor canvas, gerbview wizard+canvas, and pcbnew wizard (on a
properly asyncified 187M module) all pass; before the fix they reproduced the
exact CI signatures.
Known-fail left on CI: zoom-cursor — the zoom-in anchors at ~ -P in headed
Firefox (screen-vs-client coordinate mix-up in the wheel->GAL path, headed
mode only); marked test.fixme(CI) pending a wx-wasm-layer investigation.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
pcbnew_embind.cpp pulls kicad_clipboard.h -> pcb_io_kicad_sexpr_parser.h ->
pcb_lexer.h, a header GENERATED during make (make_lexer custom command on the
pcbcommon target). The embind compile runs right after CMake configure, so on
a fresh build dir the header doesn't exist yet — the ephemeral Hetzner main CI
(run 27327303032) died on it in minutes. Local/incremental dirs already had the
header, which is why 9dbfddc's -I${KICAD_BUILD}/common alone seemed sufficient.
Pre-build pcbcommon before compiling the bindings: make resolves the real
dependency graph (no duplicated lexer-generator args), the app target needs
pcbcommon anyway so total build time is unchanged, and incrementally it's a
no-op. Verified by deleting pcb_lexer.h/pcb_keywords.cpp/pcbnew_embind.o inside
the docker build-cache volume and rebuilding: pre-build regenerates the header
before the embind compile succeeds.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- ci-build-slowness-findings.md: the full investigation log — measured
phase breakdowns of the 4h05m baseline, the v121 lock convoy, the
badly-built official Binaryen Linux tarballs (4-13x slow asyncify),
allocator/THP/arm64 dead ends, and the validated 1h14m41s result.
- wasm-exceptions-experiment.md: -fwasm-exceptions would shrink the
asyncify set at the root (pcbnew 338 MB -> 92 MB raw) but is blocked
by an LLVM br_table codegen bug in emscripten 4.0.2; full resume
recipe + plumbing patch included.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Per run, a fresh ccx53 VM is created (Cyclenerd/hcloud-github-runner),
builds + tests run on it, and it is always torn down — no persistent
infra, no lingering cost. Secrets: HCLOUD_TOKEN, HETZNER_RUNNER_PAT.
- ci-full-build.yml: the main-branch CI and the only push-triggered
workflow. Full build of all 6 tools (pipelined wasm-opt, self-built
Binaryen 130, ~1h15m) + KiCad e2e; docs-only pushes excluded via
paths-ignore.
- ci.yml (eeschema-only smoke) and wasm-opt-bench.yml (config sweeps):
workflow_dispatch-only, since every push-run costs a paid VM.
- Shared one-Hetzner-slot concurrency group across the workflows.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Tooling that produced the perf findings (committed for rerunnability):
- scripts/bench/wasm-opt-bench.sh + o2-config-sweep.sh + sweep.conf:
replay asyncify/-O2 over a cached fixture under allocator/THP/core
matrices on the Hetzner runner.
- scripts/bench/setup-vm.sh + cloud-init/ + vm-build.sh: local QEMU
(HVF) Ubuntu guest with Docker CE to verify Linux builds without
burning paid runners.
- .gitignore: bench fixtures/results and the VM image stay local.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Three orchestration fixes, validated end-to-end on the Hetzner ccx53
(run 27280051992, 1h14m41s vs 4h05m baseline, e2e identical):
- docker-compose CPU/memory caps were hardcoded to dev-Mac defaults
(10 CPUs / 32G); now env-tunable via KICAD_DOCKER_CPUS/KICAD_DOCKER_MEM
(CI exports nproc/110G — the 32-core runner was compiling on 10 cores).
- docker/build.sh: split build_app into compile_app (container) +
postprocess_app (host-side dyncall shims + finalize + asyncify + -O2)
and added KICAD_PIPELINE=1 mode that overlaps each tool's host-side
wasm-opt with the next tool's container compile (max 2 concurrent
postprocesses — pcbnew -O2 peaks ~34G RSS). Also: comma-separated app
lists for cheap pipeline repros.
- get-wasm-opt.sh: Binaryen default 121 -> 130 (fixes the v121 -O2 lock
convoy, ~9x) and BINARYEN_BUILD_FROM_SOURCE=1 support: the official
x86_64-linux release tarballs (Alpine/musl, no LTO, assertions on) run
asyncify 4x slower than a stock gcc -O3+LTO build with sha256-identical
output; self-build takes ~5 min and is instantly repaid.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Restores the immediate-mode GL shim that the WASM runtime loads; it had
been dropped from the tree while the CI work was in flight.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
