A review of the group-E fixes found 13 further defects; ten were introduced by
those fixes, two pre-existed and were merely relocated, one is deferred.
Services / transport
E-10 retireWorker synthesized no bg/exit frame, so sharedspice's s_bgRunning
mirror stayed latched true after a mid-run worker death: Run stayed
disabled and the promised fresh-worker restart was unreachable for the
whole session. Retirement now dispatches a synthetic controlled-exit
straight to the installed handler (never through dispatchEvt — a
fabricated frame must not touch the credit ledger). Driving the repro
exposed two further defects, both fixed here: a replacement worker
trapped on pre-init engine reads, and the rerun's cm_input_path/circ hit
that uninitialized engine before KiCad's validate() re-init (the native
flow assumes a crashed engine survives in-process — true for the dll,
false for a dead worker). Reads now answer their empty shapes pre-init,
writes lazy-init, and init is idempotent per worker engine.
E-19 dispatchEvt acked only AFTER handler(evt) returned, and the sharedspice
client deliberately rethrows non-trap errors — so each throw leaked one
unit of the 64-frame credit window until the stream died with a
misattributed "transport exceeded". The ack moves to a finally in both
service copies; the throw still propagates (the trap machinery needs it).
E-20 the oversize-line path promises to transfer the accepted prefix, but
with the window full that flush only DEFERS, and stopEventStream wiped
the deferred queue — losing the diagnostics that explain the failure.
The terminal notice now carries them as pendingEvents; both hosts
deliver them in order, unacked (the fatal frame is outside the credit
protocol).
E-21 the 30s prefetch deadline discarded every model already collected and
reported nothing. A caller-owned progress sink ships the partials and
the omission reaches the export report. (Awaiting the aborted collection
was rejected: an in-flight source fetch is not abortable — E-4's
original disease.) Plus a serving-candidate memo, so a .wrl ref served
by its .step fallback stops re-probing the miss on every export.
Scheduler
E-14 _terminalizeNativeTrap classified by message substring, so any plain JS
error QUOTING 'Aborted(' or 'out of bounds' permanently bricked a
healthy instance. Now structural only: instanceof RuntimeError plus a
duck-typed name check (verified in this build's glue that abort() throws
a genuine RuntimeError both pre- and post-runtime-init). Module.onAbort
now latches the gate — the authoritative notification, previously
ignored.
E-15 the shim half: _pumpResume gates on terminal (catching wakes already
queued at latch time) and resolveWait refuses on terminal WITHOUT
consuming the entry, so a frame stays visibly parked rather than
resuming inside a trapped module.
E-16 the E-5 handler read the realm-global scheduler at dispatch instead of
its installing module's; also frees the per-line buffer on the non-trap
rethrow path.
E-11 get_vec trusted the worker's res.length over the transferred arrays.
Observed death shape: a 4 GiB std::vector threw an unhandled
std::length_error that exited the editor's main loop. Now clamped, with
the buffers freed on every failure path.
Guardrails (replacing two deferred refactors: e2e→production-code injection and
collapsing the four copies of the worker-lifecycle machinery)
E-18 the source contract asserted comment-string counts — rewording failed
CI while moving a guard outside its #ifdef passed. It now parses the
#ifdef regions and asserts on code.
service-stub-parity.ts pins what the four lifecycle copies must share:
credit-window equality parsed from source, the finally-ack, boot
deadlines, terminal-notice consumption. The transport numbers are now
single-sourced from the worker.
CI actually runs the gates: the web/standalone vitest suites (which had
NEVER run in CI), the reducer, the source contract and the parity tool —
with a NON_PLAYWRIGHT_GATES check so deleting a step re-fails the lint.
E-22 the e2e occ stub's 60s boot watchdog, deleted in a66e109, is restored in
the ngspice-stub shape with a wedgeNextBoot() repro hook.
Every behavioral fix has red-then-green evidence (the reds were captured first).
E-17 (a stale RUNNING cross-stamping the next run's generation under E-6's
transport deferral) is DEFERRED with its analysis recorded — a real fix needs
run identity on the bg frames.
Test hygiene: the dwell lint now requires the mandated ": <why>" and all 47 bare
markers carry their reason; three export-report dwells became modal-lease polls;
exact-ledger assertions became relative deltas; the dead data-wx-dom-id branch,
an unused fault hook and unused receipt plumbing are gone; abort scans, wx
dialog drivers, the sim harness and the vitest FakeWorker are each one copy now.
Bumps kicad and wxwidgets to their findings-group-e tips.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
E-5: the module-identity bridge called installingModule._malloc, but this
build exposes _malloc only as a bare glue-closure export (Module._malloc is
absent) — every char/stat event entry threw TypeError, which also starved
the E-6 credit window (thrown dispatches never acked) and wedged the queued
bg-finished frame behind them. The bridge now uses the bare closure exports
(identity is still exact: the EM_JS body IS the installing module's closure;
the __ngspiceOnEvent self-disarm covers supersession).
E-6 (codex reference design corrected — its validation matrix never ran):
a FULL credit window was terminal (stopEventStream at 64 in-flight frames).
Under live e2e that killed a real simulation: bg-thread emissions proxy one
per task, so each line ships as its own frame and a normal transient outruns
a busy main thread. A full window now DEFERS into a bounded FIFO (512 events
/ 4 MiB) drained in order as acks free credit; only true overload or an
invalid ack is terminal. Retention stays bounded (8 MiB in flight + 4 MiB
deferred + 1 MiB open batch). And the service/harness mirror queue now acks
at ENQUEUE — placing a frame in the bounded pre-handler queue is taking
ownership; without that, a stream starting before the C++ handler installs
(the ngspice-probe page) starves the worker window forever.
Test updates: worker-batch reducer — new "a full credit window defers and
drains in order, never terminal" case pinning the regression; the storm case
now proves the deferred caps are the terminal edge. board-ready.ts gains the
owner-free openBoardProgrammatically (codex helper the ported occ-export
spec needs; the barrier-based waitForUiBoardReady was NOT taken).
occ-export.spec: domId is optional on this line's registry (coordinate
fallback is the supported path).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Adapted from codex/asyncify-execution-owner-core 3753320. Service side mirrors
the occ-service shape (E-1 watchdogs, E-2 fail-all + boot-death fix — onerror
now rejects the in-flight boot waiter instead of stranding it, E-3
onmessageerror terminal, Blob URL revoked, per-generation evtQueue cleared on
retirement).
E-6 transport bounds (worker hunks re-applied inside the emscripten-6
em-pthread else-branch — the codex file predates that split, so this is a
re-application, not a cherry-pick):
- batch cut at 512 lines / 1 MiB exact JSON-UTF-8 bytes, measured before a
line is retained; a single line > 1 MiB flushes the accepted prefix then
stops the event stream terminally (never retained);
- posting gated by a 64-frame / 8 MiB unacked credit window; each frame
carries { eventSequence, eventBytes } and is released only by an exact
{ sequence, bytes } ack; any mismatched ack is terminal;
- the service mirrors the same 64-frame / 8 MiB bound on its pre-handler
queue, acks after handing a frame to __ngspiceOnEvent, and retires the
generation on invalid credit; { fatal } frames retire the worker.
Tests: ngspice-service.test.ts (11, ported) — watchdogs, crash/bootError/
decode-fault settlement + recovery, out-of-order ids, sync postMessage throw,
stale-generation event drops, fatal-frame retirement. tests/tools/
ngspice-worker-batch-unit.ts (node:vm over the production worker source;
`npm run ngspice:worker-batch`) — bounded ordered chunks, byte-pressure
flush, 100k-chunk credit storm, over-limit line, exact ack lease. e2e harness
twin updated to speak the ack protocol (adds __ngspiceServiceTestHooks).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Adapted from codex/asyncify-execution-owner-core 3753320 (scheduler-free on
that branch already; one comment line re-worded for the JSPI line):
- E-1: generation-slotted WorkerSlot with a 2-min boot watchdog (armed before
resolveWasmBase, so a hung delivery lookup expires too) and a 30-min
per-request response watchdog; retireWorker() is the single idempotent
funnel (fail that generation's pendings, terminate, revoke the worker Blob
URL, clear timers/listeners).
- E-2: worker.onerror is wired for the worker's whole life and settles every
in-flight STEP/export request; a synchronous postMessage throw settles its
request without leaking the pending id; late frames from a retired
generation are inert.
- E-3: worker.onmessageerror retires the generation like error does.
- E-4: collectBoardModelFiles is a pure source/IDB/network path (no editor
MEMFS round-trip) taking an AbortSignal checked at every loop head;
prefetchBoardModels races it against a 30 s deadline — timeout is non-fatal
(export proceeds without models) and late results are inert.
Tests: occ-service.test.ts (7, ported) — boot/response watchdog expiry,
crash-settles-all, bootError retry, decode-fault retirement (invokes the real
onmessageerror transition, per J-4), hung-prefetch export; models-bridge.test.ts
+3 — abort inertness, zero FS access on the collect path. e2e harness twin
updated to the same generation shape (adds __occServiceTestHooks/failDecode).
False-green audited: 19 cases fail with the fixes reverted.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
CI firefox read 0.228 header diff against the Chromium reference while its own
dark and light renders were pixel-identical — pure Firefox-vs-Chromium AA, not
a theme leak. Add the firefox light render as wizard-04-finish-headless-firefox.png
and pick the reference by browserName.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LvvARNMxQrq17vogn3REMU
A bare expect(false).toBe(true) after the 90s poll hid WHAT diverged in the
drift-trio S4 same-item races (CI 2026-08-24 ×2, local under 15-worker
contention 2/21). The failure now reports each tab that differs from A and
the first differing model line, so the next divergence self-documents
(memory: s4-value-race-divergence — never re-quarantine without the diff).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AH2iPekUGsEAYnMUD5BmAi
emcc only runs Binaryen at link -O2+ (link.py: should_run_binaryen_optimizer
returns OPT_LEVEL >= 2) and we link at -O1, so the shipped module had never seen
wasm-opt at all — it kept its entire 19.56 MB name section, ~20% of the editor
(-sJSPI sets ASYNCIFY=2, which suppresses wasm-ld's --strip-debug, leaving
wasm-opt as the only thing that would drop it). Step 8.2 runs it post-link and
in-container, so CI's cached compile phase covers it and the host post-process
stays pure-host.
Default -O2, picked by measuring every level on the same module: -O0 already
captures 27% of the raw win (it is mostly the name section), -O2 costs 23 s and
gives the best frame rate, and -O3/-O4/-Os/-Oz cost 48-132 s for at most 1.5%
more brotli — -O4 is not even smaller than -O3. Targets that already link -O2/-Oz
(occ_service, kicad_tools) are skipped by testing for the target_features
section, which emcc strips whenever it ran the optimizer itself, so there is no
hard-coded target list to drift. Feature flags come from the module's own
target_features section and so cannot diverge from the link.
The perf specs reported requestAnimationFrame ticks as "FPS". That is not a frame
rate: rAF fires on the compositor's schedule whether or not the GAL redrew, and
it read 120/s on a board where the renderer completed zero frames in six seconds.
measureInteractionFps now counts completed GAL frames — runs of draws to the
default framebuffer, exactly one per frame in every AA mode — and drives a pure
middle-drag pan after a zoom-to-fit. Mixing wheel zoom into the drive made the
result depend on where the wheel left the view: +-20% across identical repeats,
against +-2% for pan alone. The report gains a GAL fps column with a regression
flag on the 1x number; rAF is kept so historical runs stay comparable.
CI has no GPU, so its number is a software-rasteriser redraw rate — a regression
signal, not a user-facing frame rate. Method and measurements in the bench report.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016X9eh1s5sTx1o9Em9KBuwR
tests/wizard-04-finish-headless.png is a spec-owned reference image the
dark-mode spec diffs against in-process (toolbar must ignore the
browser's dark theme) — it was never a promotable baseline, but the
baselines-to-R2 migration (6dfa6eb) swept it out of git with the rest of
the PNGs, and every e2e run since fails with ENOENT at
screenshot-compare.ts:56. Restored verbatim from 6dfa6eb^ and documented
as deliberately committed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The runtime is JSPI-only; this removes everything that still pretended
otherwise. Three exhaustive sweeps (C++/JS+build+CI/tests+docs) drove
the inventory; every deletion verified by grep closure + full gates.
Broken-right-now fixes:
- deploy-staging.yml passed the retired opt_level input — the workflow
could not even start. Removed.
- env.sh carried dead exports with a live -sASYNCIFY=1 inside
(WASM_LDFLAGS/PTHREAD_LDFLAGS, zero consumers). Removed; the
WASM_LEGACY_EXCEPTIONS rationale rewritten to the real reason.
- docker/build.sh exported PCBJAM_ASYNC_BACKEND (read nowhere). Gone.
Dead weight removed:
- binaryen submodule (nothing builds or invokes it), wasm-opt-bench
workflow + scripts/bench/, get-wasm-opt.sh, diagnostics.js (242 lines
of Asyncify-API-only code), the KICAD_PIPELINE background-postprocess
scaffolding (existed to parallelize the deleted wasm-opt phase; the
postprocess is a seconds-long node script and now runs inline),
build-monitor's dead asyncify rows, sched-context orphan build
output, dead .gitignore entries, the .jspi-assets spike dir (the two
wf-result research JSONs moved to docs/features/async/migration-evidence/).
- bindings: fiber_park.h + its 12 embind registrations (broken-if-
called under JSPI), the kicadOpenFileStart/OPEN_JOB starter route,
main_stack_runner.h + 5 includes, the always-null context-sleep weak
hook in nanosleep_yield.c.
- shim: the backend field (installed-flag idempotency instead),
noteContextWait (dead both sides), the __wxAsyncifyDump alias (+ the
WasmTool fallback and string-dump normalize branch).
- web: the emscripten-6-ignored mainScriptUrlOrBlob option in boot.ts
(gerber-demo keeps it: it loads the deployed CDN release, which
predates emscripten 6 — noted inline).
Conditionals: all 'backend === jspi' checks reduced to scheduler-
presence checks; races_quiescent re-keyed from Asyncify.state (vacuous)
to real backlog quiescence (resumeReady/mutatorQueue — NOT _windowLive,
which is the probing activation's own window by definition).
Renames (identifiers only, no file renames): ASYNC_LINK_FLAGS→
JSPI_LINK_FLAGS and Makefile ASYNC_LDFLAGS→JSPI_LDFLAGS,
kicadCollabFiberBusy→kicadCollabBusy (embind + web + tests),
collab_common.h fiber*→apply*/coroutine naming, asyncifySignatures→
wasmTrapSignatures (lists byte-identical).
Tests: the two remaining vacuous [wx-asyncify]/fiber-resume-refused
asserts re-keyed to live JSPI beacons; eeschema-load's failure message
no longer sends the developer to a deleted script; wait-beacons' dead
families/parser deleted; lane-0 legacy-glue guards removed (lane 0 is
unconstructible); the embind test.fail re-gated with the JSPI reason
(plain embind invokers cannot suspend — verified still failing);
lint-determinism now scans tests/jspi (166 files clean);
eeschema-collab local-move gated to chromium (~50% flaky on FF even
solo; pcbnew twin covers both engines).
Docs: DEBUG.md rewritten as the JSPI debugging guide; build.md
describes the single-phase build; docs/features/async/README.md
banner-marked historical and repointed at the NEW
23-jspi-runtime.md (current architecture: export census, turnstile,
libcontext ownership + refusal contract, embind call shapes, the
em-pthread service-wrapper trick, exception policy, known gaps).
Gates on the cleaned tree: test:e2e 725 passed / 0 failed (after the
quiescence-probe fix; the 3 other reds were verified contention flakes
solo-green or the documented FF gate), web 76/0, jspi 18/18 both
engines, vitest 295/295 + 17/17, all lints green, live-app census
clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016X9eh1s5sTx1o9Em9KBuwR
Live-app fix (Place Footprints / routing dead in Chrome): submodule
bumps carry the coroutine ownership fix (kicad 012d95ecb4) and the
handler-exception survival fix (wxwidgets 1b5f0e31f4).
Emscripten-6 fallout:
- occ/ngspice worker wrappers: mainScriptUrlOrBlob was removed
upstream; pthread children re-run the wrapper blob, so an em-pthread
realm now importScripts the glue and gets out of the way (before:
recursive service boots, pool never fills, silent 180s boot hangs —
every occ spec and ngspice bg_run).
- Makefile.wasm: -sASYNCIFY frankenlinks on the no-wx coroutine repro
targets ported to -sJSPI (the JSPI-only libcontext crashed at first
yield under them); mainloop/gl repro pages drive their tick through a
promising export (emscripten_set_main_loop callbacks cannot suspend);
retired inject-dyncall-shims lines removed (targets were unbuildable
since Phase 8); $stringToNewUTF8 force-included (the EM_ASM value
bridge aborted the runtime on the first decoded exception).
- fiber-park levers: neither embind shape can drive suspending levers
(plain throws on strict-JSPI Firefox; emscripten::async() re-executes
its invoker on settle) — kept sync for manual Chromium probing, spec
coverage moved to the jspi-coroutine harness (18 cases).
Suite work:
- Playwright 1.61.1 -> 1.62.1 (Firefox 153: JSPI on by default).
- fiber-resume-park.spec retired -> coroutine-lifecycle.spec: census
gate over boot / board load / chooser open / cancel (deterministically
red on the pre-fix build).
- Blind asyncify-era pins re-keyed: quasimodal-strand + wait-beacons
beacon regexes, footprint-chooser-close liveness -> wx parking-timer
heartbeat (scheduler counters idle flat on Firefox).
- occ/ngspice test providers: 60s boot timeout + worker error
surfacing (a worker death used to be a silent 180s timeout).
- Harness pages: stale 9.99 config dir -> 10.0 (library_manager wxCHECK
noise, chooser had no libraries).
- gal-webgl harness: missing artifacts rebuilt (boost/glm extracted to
the host sysroot), PgmOrNull stub added for the rebased GAL.
- jspi-scheduler: clean-shutdown console line restored (app-quit
contract), quarantine never yanks SP from a live window.
Gates: test:e2e 699 passed / 0 failed (wx-chromium, kicad-firefox,
kicad-chromium, jspi-firefox, coroutine-firefox); web ff/cr/mobile 71
passed; lint:ci-coverage 166, lint:determinism 163, screenshots
manifest 492 current, corpus 7/7, tools:contract green. Offline
screenshot baselines show expected mass drift from the engine bump —
re-baseline (screenshots:noise -> promote) is a follow-up.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016X9eh1s5sTx1o9Em9KBuwR
Doc 20 D-1, second slice: the injector now injects asyncify-scheduler.js
unconditionally — the WX_SCHEDULER=0 legacy opt-out, the
SHIM_DISABLE_HANDLESLEEP ablation skip, and the
SHIM_DISABLE_TRAMPOLINE_HEAL ablation skip are gone, and
scripts/common/shims/handlesleep.js is deleted (the scheduler subsumed
its capture/restore, fiber guard, and trampoline-heal duties in S2).
Comment-only scrubs point the remaining references at the scheduler
shim. .ci-cache-epoch bumped (shim/injector behavior changed).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
Doc 17 step S0 scaffolding: WX_SCHEDULER=1 injector path with an
observation-only asyncify-scheduler.js skeleton (legacy shim stays
authoritative until S2), guard-beacon extraction with occurrence
recovery for rate-limited beacons, and the fixme'd N2 ordering spec
(add-then-move probe; un-fixme at S1).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TfxKn5utcntBSnxz4ZnYKs
drift-trio-scenarios S4/S5 flaked 4× across this week's CI runs (three wasm
instances + full-suite load on the runner) while passing 77/78 locally under
stress — and the one local miss carried zero guard beacons, i.e. the same
under-load convergence shortfall, not a code path. Same treatment as the
follow spec: condition-based polls keep their shape, windows grow to what a
starved box actually needs (inline S4/S5 polls 20s→60s, waitAllContain
30s→90s, settleConverged 30s→90s). Local convergence stays ~1s.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019SE4o46Lnq3hF574FFq8x4
runOnFiber now runs bodies strictly one-at-a-time through a park-safe FIFO
(collab_common.h): the per-body fire-and-forget coroutine interleaved under
load — an asyncify park inside commit.Push let the event loop start the next
body, so a local commit and a remote apply ran interleaved on shared state
(s_applyingRemote is one global), silently losing applies on the actively-
editing receiver (fuzz finding #10a; B now fuzzes clean; 39-test suite green).
kicadCollabFiberBusy embind probe (merged + standalone registrations): a
bare-embind-stack scratch save during a parked fiber mis-dispatches (table
index OOB) — trio.ts modelText/drift and production drift-detect now defer
while fiber work is in flight (#10b hardening; the trap's root cause is still
open and needs a symbolized stack — fuzz stays fixme'd).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01G5cAM9M6q34n5X4dbrfVvi
drift-trio-scenarios.spec.ts: disjoint ping-pong, same-item interleave (no
settle between bursts), conflict pairs (move-vs-delete / value-vs-value /
move-vs-move — winner is CRDT policy, asserted only as convergence + drift
silence), undo storm, 12-edit burst churn, late-joiner adopt, and Ctrl+S mid
peer burst; per-tool adapters, marker-waits before every sweep (finding #7).
S4 exposed finding #9: mutation hooks resolved item pointers at call time and
committed later on the fiber — a remote remove in between frees the pointer
(doApplyItems) and the commit resurrects the deleted item. Phase-B mutation
hooks now re-resolve by uuid ON the fiber; a vanished item makes the mutation
lose silently. applyDeltaToY's concurrent update/delete verified coherent
(full resurrect or full remove) — no shared change needed. trio.ts: TabSet
oracles + exported startV2 for duo/late-join composition.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01G5cAM9M6q34n5X4dbrfVvi
17 kicadCollabTest* action primitives (eeschema: wire/junction/no-connect/
label/symbol/move/mirror/duplicate; pcbnew: track/via/text/zone/flip/
footprint-field/lock/move/duplicate), each a real SCH_/BOARD_COMMIT on the
fiber so the listener → flushDiff emit runs as for UI edits; tool-unique
names, merged-image safe. Fixes surfaced by the catalogs (0008 §10 #4–#8):
eeschema adds SetParent before staging (Push silently skips listener
notifications for unparented items), and pcbnew blobForItem now Formats
non-footprints with the FILE writer + wrapInBoardEnvelope — SaveSelection's
transfer copy cleared the locked flag, so (locked yes) never reached the doc.
drift-trio.spec.ts gains full A/B-alternating catalogs with per-step landed
gate + oracle sweep. Bumps kicad for the Duplicate child-uuid re-roll.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01G5cAM9M6q34n5X4dbrfVvi
Squash of experiment/ff-big-modules vs main.
Big-module routing removed: native-EH shrank kicad_editor below
SpiderMonkey's x86-64 code budget (runs 29355049705/29356152413 green on
stock Firefox), so BIG_MODULE_SPECS routing and the baseline-only-JIT
crutch are gone — kicad-firefox and kicad-chromium both run the full
suite, with the module compiled the way real users' browsers compile it.
Per-engine screenshots end to end: stableShot/shotPath write
test-results/<engine>/<name>.png; baselines move to
baseline-screenshots/{chromium,firefox}/ and the whole tools/screenshots
pipeline (compare/promote/manifest/spec-map/changelog/Discord) keys on
<engine>/<name>. Previously Firefox and Chromium renders of one spec
overwrote each other and Firefox renders were never actually gated.
Seeded from CI run 29421380806 (92 new firefox baselines, +24 chromium
web-suite shots); manifest generated from the baseline tree.
One merged playwright.config.ts (kicad/asyncify/coroutine/perf as
projects); ~25 dead npm scripts dropped. The web suite is gated in CI for
the first time ever (4 rotted specs fixed, 5 broken lib-bridge specs
triaged as fixme in docs/features/web-e2e-rot/); cheap lint step after
npm ci; last 26 blind-sleep violations fixed.
SwiftShader retired: CI Chromium renders WebGL on ANGLE → Mesa llvmpipe
(--use-gl=angle --use-angle=gl --ignore-gpu-blocklist; the blocklist flag
is mandatory — llvmpipe is blocklisted and WebGL is silently unavailable
without it) in BOTH configs. Under WORKERS=4 congestion SwiftShader
transiently failed the first post-board-load draw and the recovery
cascade ended in a silent permanent Cairo fallback — that engine flip was
the "~1.2% changedRatio both directions" occ-export baseline flake.
Validated 160/160 across two 80-repeat rigs; full analysis in
docs/features/wx-parity-bugs/occ-export-context-eviction.md. Chromium
baselines shift slightly on llvmpipe — promote once from the first green
run. Deflakes the new coverage exposed: presence baselines settle before
capture; presence fixtures declare current file formats; perf gets its
own outputDir so CI evidence survives; occ-export settles the board paint
before the export dialog; menu-item waits (waitForRenderedByLabel before
clickMenuItem) in 4 specs + the TESTING.md rule.
Web suite runs the PROD build, in parallel: webServer becomes backend
`start` + the standalone's e2e:preview (build-preview.mjs: link-wasm →
stash the public/wasm symlink aside during vite build, build-demo.mjs's
move — then vite preview as the persistent server). The wasm middleware
serves /wasm/* in preview and emits COOP/COEP/CORP itself (a pthread
worker script's own response must carry COEP or Chrome kills it with
ERR_BLOCKED_BY_RESPONSE). VITE_* flags bake at build time;
VITE_ALLOW_USER_OVERRIDE joins turbo globalEnv. fullyParallel + default
workers: 5.2m → 1.4m. Determinism fixes the parallel run exposed:
shared-page specs become serial groups; locks.spec grabs alice's exact
item via the new kicadCollabTestSelectByUuid hook (cross-tab "first
footprint" order is not a ysync invariant); quit specs poll page.url()
(quit supersedes its own navigation — NS_BINDING_ABORTED on Firefox).
Suite: 51 passed / 12 skipped / 0 failed in 1.6m.
CI-coverage gate (lint:ci-coverage): every tests/**/*.spec.ts must be
reachable from the npm scripts the workflows invoke — scraped from
.github/workflows/, resolved through package.json, coverage asked from
playwright --list itself. Rules: uncovered-spec + orphan-project (with a
documented LOCAL_ONLY_PROJECTS allowlist). Gating next to
lint:determinism; 138 spec files / 13 projects accounted for.
Product fixes kept from the investigations (reachable on real GPUs too):
wx 7799fd1be5 — paint flags clear before dispatch + Invalidate always
propagates; kicad 3dcfea5e45 — SwiftShader pass-boundary flush +
per-instance font texture + first-frame GL-error drain (GAL recovery
recovers instead of falling back to Cairo) + the user-facing eeschema
switch navigates again under __EMSCRIPTEN__ (project-sync's
FaceRegistered gate had rerouted it into the hidden sync player; caught
by the newly-gated web suite).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018eUxiPApHgGiu9NFyQfhAq
The occ_service export worker has its own MEMFS — the editor's lazily-fetched
lib models were invisible there, so every export was a bare board (54
"Could not add 3D model" warnings on pic_programmer, 2 STEP products).
- models-bridge: collectBoardModelFiles(boardText) — scan refs, ensure via
the 0004 sparse source (IDB/R2, wrl->step fallback), read staged bytes
back, dedupe by real staged path.
- occ-service.ts: attach the collected models to every export request
(best-effort — prefetch failure still exports, misses reported by the
exporter); transfer the body buffers.
- occ-worker.js (shared app/harness): pass req.models through to occExport.
- occ_service_main.cpp: occExport(board, params, models) stages each entry
under PCBJAM_3D::MODELS_MEMFS_ROOT (path-sanitized) for the exporter's
staged-model probe (kicad 83645275ac), removed again after the export.
- tests: harness occ stub mirrors the prefetch against the page kicadLibs
provider + captures report/productCount; new occ-export-models.spec.ts
guards the delivery (green companion pins preconditions; guard asserts 0
missing lib models + component PRODUCTs). pic_programmer: 17/17 staged,
87 products @ 13.3 MB (was 2 @ 402 KB). models-bridge unit tests 13/13.
Known remainder (0007 step 4): project-local ${KIPRJMOD} refs still drop.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017UjpnviP3ZDxTM1Ap63Sqv
Make the Playwright e2e + kicad suites deterministic so screenshot flake stops
tracing to timing races.
- Blind page.waitForTimeout -> condition waits (expect.poll, web-first
assertions, waitUntil) + readiness helpers (waitForWxApp, waitForCanvasApp).
Remaining sleeps are documented interaction dwells (annotated).
- Defensive "if element exists" branches -> loud asserts; label-fallback chains
-> normalized clickMenuItemByText. First-run wizard for/if loops removed by
seeding calculator/gerbview/pcbnew HTMLs.
- Screenshots: new stableShot(page, name) settles the render in-page (canvas
hash over rAF) then writes a raw PNG to test-results/ for the existing offline
gate (tools/screenshots vs baseline-screenshots). Replaces toHaveScreenshot,
which did inline compare + its own baselines and had decoupled the specs from
the real gate. scale:'css' pinned.
- retries: 0 in both configs.
- Guard: tests/tools/lint-determinism.ts (npm run lint:determinism) bans blind
sleeps / toHaveScreenshot / inline retries / swallowed catches in specs;
documented exceptions carry a marker. Rules in tests/TESTING.md.
Assertions, coverage, and renders unchanged (semantic-equivalence reviewed;
captures pixel-identical modulo inherent timer/timestamp/3d-raytrace variance).
Both suites green at retries:0 (e2e 340, kicad 92); ~35-61% faster.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BVX1pHMvRPYHdp6ZfEawrk
The 3d-webgl merge (kicad eb13ff3bdc: the viewer now defaults to the real
OpenGL renderer via wasm/gl1, and occ-split moves STEP parsing into the
occ_service worker) made the raytracer-era orchestration on this branch moot —
main's chromium-ci phase is green at 15-way parallelism (28666407570 /
28698861536). Drop what no longer earns its complexity, keep the diagnostics,
fix main's live flake, and make the deadlock spec test what it was written for.
- REVERT the chromium-ci-3d serial project, the two-phase test:kicad:ci, the
SwiftShader GPU-process flags, and the resize-drag/models skips: config and
package.json are byte-for-byte back to main's shape. The raytracer contention
they guarded is no longer on the CI path.
- FIX main's live flake: run 28698861536 is green only via retry
(3d-viewer.spec:26 flaky) and 28666407570's deadlock red sampled an ALL-ZERO
pixel signature — the viewer's first frame lags the canvas's creation on
software WebGL under parallel load, and sampling too early reads an all-black
backbuffer. New waitForThreeDRender() gates render assertions on actual
pixels (1s-interval full-frame CPU reads) instead of fixed sleeps, used by
3d-viewer.spec:26 and the models render tail.
- KEEP the storm-proofed samplers (one full-frame getImageData on a
willReadFrequently canvas replacing 256 per-pixel GPU round-trips per sample
— the "GPU stall due to ReadPixels" trigger) and the logThreeDDiag
instrumentation: engine-independent, and they de-risk every remaining
software-GL pixel read.
- models spec: bridge assertions stay front-loaded (the protocol regression
signal is independent of the render); the occ_service parse verdict is now
POLLED — it lands async relative to the bridge ensures, so asserting it
immediately raced the worker; the render tail runs again everywhere. (The
pre-webgl raytracer+models renderer-death documented in a17f3be does not
affect the OpenGL default path — the raytracer-toggle+models combination
remains untested product surface, tracked outside this branch.)
- deadlock spec: the deadlock it guards is raytracer-specific and the viewer
now defaults to OpenGL — on the GL engine it either passes vacuously (fast
renders make every liveness assertion trivial, 28698861536) or fails on the
black first frame (28666407570). It now flips the engine via the "Use
raytracing" toolbar toggle (loud assert if the toggle moved) and
cross-checks engagement by requiring the canvas pixels to CHANGE after the
flip with no input in between (the raytraced frame is lit differently; a GL
re-render reproduces identical pixels; heap growth is unusable — mimalloc
satisfies the raytracer from freed arena pages). That guard immediately
caught a REAL defect: on the webgl-era wasm build the toggle is INERT (the
click lands and "Reload time" updates, but the canvas never changes —
suspects: DoRePaint's silent catch(runtime_error) freezing the canvas after
a raytracer Redraw throw, or ToggleRaytracing writing m_boardAdapter.m_Cfg
while RenderEngineChanged() reads GetAppSettings<…>(), possibly different
instances in the merged bundle). The spec is therefore test.skip-annotated
as a KNOWN ISSUE with the full engine-force machinery in place — unskipping
it self-validates the product fix. The CI-skip also stays (raytracer
liveness needs real-GPU pacing; the Worker-boot deadlock mechanism is
covered on CI by the standalone wx harnesses).
- 180s viewer-open waits kept as pure CI headroom (never slow a passing run).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The headed Mesa-llvmpipe switch (ce44636) was built on a misdiagnosis. With
--use-gl=angle --use-angle=gl under Xvfb, Chromium gets NO WebGL AT ALL on the
GPU-less CI VM: runs 28652367347 + 28664038296 both log
`glcanvas count before opening 3D viewer: 0` (even the pcbnew board's own GAL
canvas fails to create) and every failure is the FIRST viewer-open wait timing
out. "Timeouts instead of crashes" meant no GL context existed to crash, not
that llvmpipe was stable-but-slow. The Mesa/GLX path works for the Firefox
project, but not for Chromium on this image.
Headless SwiftShader is the environment where GL provably works — the
serialized run 28649537489 opened the viewer and rendered/sampled 56 colours
(viewer:26/130/209 all passed single-tab). Its remaining failures split into
two families, each addressed on its own evidence:
1) SwiftShader GPU-process kills (deadlock spec's lost canvas/black frame):
- chromium-ci-3d back to headless + --enable-unsafe-swiftshader, plus
--disable-gpu-watchdog and --disable-gpu-process-crash-limit: heavy churn
stalls the software-GL GPU process until the watchdog shoots it
(CONTEXT_LOST -> removed GL canvas -> "Target crashed"); with the watchdog
off a slow op completes, and without the crash limit a lost GPU process
keeps restarting for later tests. Same args locally (no-ops on a healthy
real GPU).
- Storm-proof every canvas sampler (3d-viewer, models, deadlock specs): one
full-frame getImageData on a willReadFrequently (CPU-backed) 2D canvas +
JS grid sampling, replacing 256 per-pixel getImageData GPU round-trips per
sample — the "GPU stall due to ReadPixels" trigger, worst in the deadlock
spec's 1.5s-interval settle polling. Sampled pixels and sig order are
unchanged.
- edge-resize spec: keep the open + 5-resize-handles assertions on CI, then
test.skip the drag — each resize step re-raytraces synchronously on the
wasm main thread and the 12-step drag blocked mouse.move past the 240s
budget even single-tab. Real-GPU-paced; still runs locally (deadlock
spec's CI-skip from d9900fe unchanged, same rationale).
2) KNOWN PRODUCT BUG uncovered in the models spec (NOT SwiftShader, NOT CI):
raytracing a scene WITH component models kills the Chromium renderer
process outright ~6s after scene build — silent process death with a FLAT
wasm heap (531-637 MB watched at 2s intervals; not OOM), no console error,
no wasm abort, no crash report. Deterministic on a real GPU (5/5),
fixture-independent (700 KB USB-C and 61 KB USON-8 STEP crash identically);
the same death hits CI SwiftShader at ~66s. Board-only raytraces complete
and stay alive, so the bug is specific to the model path. The spec's one
historical green ended at raytrace-age ~6s — inside the death window by
luck; every CI run of it ever died. Fix here: front-load the bridge
assertions (ensure requests fire during scene BUILD, before the raytrace),
so protocol regressions still fail the suite everywhere, and test.skip the
render tail unconditionally with the bug documented at the skip site. The
raytrace->blit pipeline stays gated on CI by 3d-viewer.spec's board-only
render. Bug tracked for a dedicated fix session.
Correct the now-wrong llvmpipe comments (config, open-wait rationale, deadlock
skip reason). Expected CI matrix: viewer open/render/stacking/titlebar +
models-bridge on SwiftShader serial; edge-resize drag + camera-move deadlock
on real GPUs locally; the Worker-boot deadlock mechanism stays covered on CI
by the standalone wx harnesses.
Locally validated on a real GPU (--project=chromium): 5 passed + 1 skipped
(the documented models render tail).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Follow-up to the headed-llvmpipe switch. That run (28652367347) stopped CRASHING —
every failure became a plain timeout, confirming llvmpipe gives a stable WebGL context
(no more CONTEXT_LOST / "Target crashed") but software raytracing is just slow:
- The first board raytrace takes ~60s on llvmpipe, right at openThreeDViewer's 60s
wait → raise it to 180s (shared helper + the models spec's inline copy). Real GPU
returns in ~2s, so it is only CI headroom.
- The deadlock spec asserts the wasm main thread stays responsive within ~15s DURING a
raytrace — a premise only true on a fast (real) GPU. Under slow software rendering a
legitimately-slow raytrace is indistinguishable from a deadlock, so skip it on CI
(test.skip on process.env.CI). The Worker-boot deadlock MECHANISM stays covered on CI
by the standalone coroutine-pthread-ondemand / raytrace-threads wx harnesses; the spec
still runs locally on a real GPU.
- Gate the headed + llvmpipe launch flags on CI so chromium-ci-3d behaves like a normal
bundled-Chromium project locally.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The three heavy 3D-viewer specs (3d-viewer, 3d-viewer-deadlock, 3d-viewer-models)
crashed on CI (run 28604015154): "Target crashed", "browser has been closed", and a
black frozen canvas. Each boots the 3D-enabled pcbnew build, which pre-warms
~hardwareConcurrency*2+8 (~68 on the 30-core VM) Web Workers AND runs a multi-threaded
CPU raytracer over SwiftShader software-WebGL. Under fullyParallel the three spec FILES
run in separate browser processes concurrently, and the simultaneous raytrace tabs
exhaust the shared GPU process's ~16 live-WebGL-context limit / renderer memory. The
pool pre-warm (the deadlock fix, 7630c7e) is correct and present in the tested binary;
per-file isolation prevented in-process Worker accumulation but NOT this cross-process
concurrency — the gap these specs' first CI exposure hit.
Fix (test-orchestration only; no wasm/submodule changes):
- Route the three specs to a dedicated `chromium-ci-3d` Playwright project (own
outputDir=pw-artifacts/kicad-3d so its failure traces survive the later perf run's
wipe; the existing tests/pw-artifacts/** upload already captures it).
- test:kicad:ci runs it as a SECOND `playwright test` pass with --workers=1, so at most
one raytrace tab is alive at a time (the environment in which a single 3D load is
reliably green — docs/features/3d-raytracer/README.md §7). An exit-code-aggregating
sequencer runs both phases and fails if either fails (no silent skip of the 3D pass).
- Cap phase-1 CI workers to 12 as headroom under the WebGL-context cap.
- Log a resource-diagnostic snapshot (hardwareConcurrency, pthread pool, GL-canvas
count, wasm/JS heap) before each 3D interaction so any recurrence is attributable.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Move OpenCASCADE out of the merged editor image into occ_service: a separate
emscripten module (-sASYNCIFY=0, MODULARIZE, in-container -Oz finalize, 2N+8
pre-warmed pthread pool) booted lazily in a dedicated Web Worker on the first
STEP export or STEP/IGES model parse. kicad_editor.wasm ~190 MB -> 130 MB;
sessions that never touch OCC never fetch its 57 MB. STEP export works in the
browser for the first time: the unchanged desktop dialog runs EXPORTER_STEP,
whose wasm shadow suspends into globalThis.occService and the export bytes go
straight to a browser download (never entering the editor heap). STEP/IGES 3D
models parse in the worker via the oce shadow (S3D WriteCache/ReadCache wire).
- wasm/occ-service/: service CMake target (hooked from the kicad fork's
top-level CMakeLists, wasm/editor pattern), embind entry
(occExport/occLoadModel), wxConfig pre-js.
- wasm/stubs/{exporter_step,oce_plugin}_stub.cpp: EM_ASYNC_JS worker bridges
(callee-shadowing; no caller #ifdefs).
- web/standalone: provider installed whenever the kicad_editor bundle boots
(cross-face safe); ONE shared worker-boot source occ-worker.js (vite ?raw;
the e2e stub reads the same file) — blob worker with locateFile absolutized
against the glue URL; export download-name guard.
- deps: OCC builds with RapidJSON so its glTF/GLB writer exists — pinned to
the vcpkg master snapshot 2025-02-26 (24b5e7a8b27f), the same code official
KiCad consumes via vcpkg.json's opencascade[rapidjson]; rapidjson's latest
tag (v1.1.0, 2016) is ill-formed under modern clang.
- tests: occ-export dialog e2e (lazy-fetch boundary + STEP download bytes),
occ-probe incl. a 9-format matrix (step/stpz/brep/xao/ply/stl/glb/u3d/pdf),
3d-viewer-models hard-asserts the worker parse; occ provider stub installed
ambiently by the kicad fixtures.
Validated against desktop kicad-cli 10.0.4: geometric exact equality (bbox
delta 0 um, volume delta 0.0000%) for STEP/GLB/STL/BREP/STPZ across three
boards and option sweeps — with desktop OCC 7.9 vs wasm OCC 7.8; PLY/XAO/PDF
structurally equal; U3D same-size (quantizer float LSBs differ). Full kicad
e2e green on Firefox and Chromium; standalone verified end to end (lazy fetch
only on the Export click; export.step 60,628 B ISO-10303-21; loadModel 700 KB
STEP -> 569 KB scenegraph cache).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
New tooling in tests/tools/screenshots/ (TypeScript via tsx):
- compare.ts: one pixelmatch engine (AA-excluded), connected-component
"where to look" boxes, old|new+boxes|heatmap triptych, per-engine floors.
- promote.ts: churn-free updater — overwrite a baseline only when decoded
pixels differ beyond the floor, copying CI bytes verbatim (no re-encode
churn); pulls a CI run via `gh run download` or a local --from dir.
- post-discord.ts: always-on CI-on-main report (SHA + e2e status + the
track-only runtime-perf table), then screenshot triptychs, batched +
size-capped + flood-collapsed + 429-aware.
- perf-report.ts: perf table with Δ vs the previous main run (via gh).
- changelog.ts: no-build git-history baseline differ (Discord trigger B).
- noise.ts / gen-manifest.ts: calibration + manifest generation.
CI wiring:
- wasm-build.yml: post-test step runs the gate + report on the already-
produced test-results (no extra build); report-only (continue-on-error),
posts only on push to main, inert without DISCORD_WEBHOOK_URL.
- ci-ubicloud.yml: secrets: inherit (pass the webhook through).
- screenshot-changelog.yml: ~30s no-build changelog on baseline changes.
screenshot-manifest.json: canonical 354-name set + best-effort engine tags
(313 chromium-swiftshader / 41 firefox-llvmpipe).
Normalize scale:'device'->'css' across 18 spec files (no-op at CI DSF=1)
so committed baselines are uniformly css-scaled.
Design: CI's Linux render is the single source of truth; no pinned
container (accept rare env drift -> re-promote); dev commits via promote.
Replaces the byte-cmp compare-screenshots.sh + file-size-proxy
update-baseline-screenshots.sh (kept for now until the first re-baseline).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Measures the current build's cold load, open+render, and pan/zoom FPS
(1x/4x/6x CPU-throttle sweep) for both editors, writing
tests/test-results/perf-*.json. Track-only: asserts only that the app booted
and the document opened — no perf thresholds, so it never gates CI.
Runs on a new Chromium `perf` Playwright project (CDP throttling; pcbnew needs
V8 anyway) via `npm run test:perf`, and is wired into wasm-build.yml as a
continue-on-error step so the numbers are captured/uploaded without flaking the
gate. Reuses the existing fixtures, ready-signal, fs-inject, and board-ready
helpers; FPS counter is cancel-able so a throttle sweep on one page doesn't
accumulate rAF loops.
.gitignore: ignore /benchmark-builds/ (disposable prebuilt-WASM + standalone
harness bundle used to reproduce the native-vs-JS-EH comparison locally).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Builds KiCad's 3D viewer for the browser (opt in with BUILD_3D_VIEWER=ON) and
wires up the legacy-GL emulation it needs.
- build-kicad-target.sh / docker/build.sh: BUILD_3D_VIEWER →
-DKICAD_BUILD_3D_VIEWER_WASM=ON + -sLEGACY_GL_EMULATION + the GL js-library.
- wasm/shims/gl_immediate_shim.js: display-list emulation (record/replay board
layers), fixed-function + GLU stubs, throw-guards for unsupported pnames, and
per-context GLImmediate init for the viewer's SECOND WebGL context.
- tests/kicad/3d-viewer.spec.ts + utils/pcbnew-ready.ts: open View → 3D Viewer.
- bumps the kicad submodule (EMSCRIPTEN-guarded GL changes).
WIP: the viewer window + UI render and geometry draws (glError=0x0), but the
board is not yet visibly rendered — GLImmediate's FFP shader program isn't
linked on the viewer's 2nd WebGL context. Full status + next steps in
features/feat/add-3d-view/notes.md.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds an end-to-end test that drives File→Open in pcbnew, injects
the .kicad_pcb and .kicad_pro files into MEMFS at the dialog's
default starting directory, drives the menu + filename text input
+ Enter accept path, and screenshots the loaded board. Parametrized
for both kicad/demos/microwave (RF polygon footprints) and
kicad/demos/pic_programmer (full multi-IC layout).
Without the rtree fix bumped in via the kicad submodule, the load
would abort on every PCB at rtree.h:1771 Classify; the test asserts
no [RTREE-DIAG] line and no Aborted(. The post-load clipboard
RuntimeError in __asyncjs__js_clipboardHasText is a separate,
pre-existing wasm-port limitation that we explicitly do not regress
on here.
- tests/kicad/load-pcb.spec.ts: serial-mode parametrized spec
- tests/kicad/load-pcb-probe.spec.ts: one-shot diagnostic probe
for inspecting wxFileDialog state on the canvas
- tests/kicad/utils/fs-inject.ts: FS.writeFile bridge from Node fs
- tests/kicad/utils/board-ready.ts: poll-for-no-dialogs readiness
- tests/baseline-screenshots/load-pcb-*.png: 6 baselines covering
both demos at pcbnew-ready / dialog-open / loaded states
- features/.../rtree-debug-findings.md: full diagnosis trail with
an upstream-reportable summary the maintainer can lift verbatim
- kicad submodule bumped to 07d8130d44 (shape_poly_set rtree fix)
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
