The jspi migration removed binaryen from .gitmodules (no asyncify
post-link pass) but left the gitlink behind, so every submodule
command errors with 'no submodule mapping found for binaryen'.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PmR6goSk7JC17h7fkgGvHG
The private repo's content-addressed files route imports
PROJECT_FILE_REVISION_HEADER from @pcbjam/contract, which re-exports
@pcbjam/shared; the previous pin (22217c9) predates that export by one
commit and breaks the closed server's watch build.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PmR6goSk7JC17h7fkgGvHG
Parameterizes the throttle-sweep FPS battery (fpsBattery helper) and adds a
jetson-agx-thor run: asyncify raf 2.1-5.3 / distinct 0.2-0.8 vs JSPI raf
1.9-5.0 / distinct 0.1-0.7 — indistinguishable. Suspension overhead is a
per-event-loop-turn cost; at hundreds of ms of GAL work per frame it stops
discriminating. vme-wren (~27 MB) remains the discriminating size class.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016X9eh1s5sTx1o9Em9KBuwR
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
mimalloc's mi_atomic_yield() is sleep(0) on wasm, reached from malloc's slow
path under cross-thread delayed-free contention. Under JSPI a yield there
suspends the activation INSIDE the allocator — any other activation that runs
next can re-enter mimalloc mid-operation. ms==0 now busy-returns like stock
emscripten. (The removelist half of the staging fix is obsolete here — no
asyncify, no removelist. The mimalloc-storm harness + spec arrive with the
staging merge.)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NDeBaKKhQztd8KiVtHuyXr
Phase 8 in the parent repo. Deleted: asyncify-scheduler.js, apply-asyncify.sh,
apply-finalize.sh, inject-dyncall-shims.sh, asyncify-imports/removelist.txt,
the wasm-opt/finalize stub pair, scripts/binaryen-hoist-pass/ (the fork stays
a dormant submodule; removal is a follow-up), bench/wasm-opt-bench.sh (README
marked historical), wasm/shims/context_sleep.cpp, and the sched-context
harness app + Makefile targets.
PCBJAM_ASYNC_BACKEND is gone: build-wx-wasm.sh hardcodes the jspi stamp
(still force-cleans pre-migration trees), build-kicad-target.sh gives editors
the JSPI link surface and the CLIs nothing (they pin ASYNCIFY=0), the stub
dance is replaced by an unconditional .real-restore, build-wasm-test.sh lost
its whole post-link loop, docker/build.sh's postprocess is the ENV shim only,
and Makefile.wasm links every app JSPI with the scheduler shim as a tracked
prerequisite. pcbjam_async_policy.h keys on __EMSCRIPTEN__.
jspi-scheduler.js: wxWasmMainLoopPump dropped from the wrap census (the
export died with the D5 detach); inert [TRACE] instrumentation removed.
CI: wasm-build.yml rewritten for the single-cache pipeline (one output cache
keyed on compile inputs; post-processed bytes cached after the shim);
opt_level input removed from both callers. wasm-cache-hash.mjs inputs now
cover patch-env-shim.mjs + jspi-scheduler.js + jspi-exports.txt.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NDeBaKKhQztd8KiVtHuyXr
sourcetrail/: Sourcetrail 2021.4.19 indexing pipeline for the wasm port —
compile-db transform (rsp expansion, PCH strip, libc++-11 pinning, path
rewrites), removelist-candidate analysis over the indexed call graph, README
with regen steps and the hard-won tricks (relative --project-file hang,
SDK-header poisoning). Heavy artifacts (.srctrldb, compile dbs, libcxx
headers) stay untracked via the folder's .gitignore.
docs/asyncify-allocator-suspend/plan.md: verified plan for the nanosleep-shim
zero-duration guard (mimalloc mi_atomic_yield=sleep(0) can suspend malloc on
the main thread), the red/green contention test, and the adversarially
verified removelist additions with measured payoff (80.9k -> 60.0k
instrumented functions). Planned, not yet executed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P2NHhbzuEHqP2JmcrSD96D
Converting the symbol-library bridge from one EM_ASYNC_JS into
beginWait -> start -> wxWasmYieldUntil -> take_result made the app hang after
every library request: the KiCad suite went from 7 minutes to 1.2 HOURS, 111
passed with the rest timing out, and the log goes silent right after a lib
request - a parked context nobody resumes. Reverted; the tree keeps the
working EM_ASYNC_JS and no code from the attempt.
Recorded for the next attempt:
- The PRECONDITION it found (four lines, quoted in the doc, deliberately NOT
landed): wxWasmYieldUntil must not park a context whose wait is already
resolved, because resolveWait deletes the entry. The in-place form had no
such window; every bridge in this pattern re-opens it. It was not the hang,
but it has only ever existed in the broken build, so it lands WITH the
conversion it protects, not before.
- Three hypotheses in evidence order, cheapest first: the resolve never
reaching the registry (this is the first KiCad-side caller of the wait
registry - check Module["_wxWasmSchedResolveContextWait"] exists in the
KiCad link); the lib request running inside the chooser's modal where the
dispatch context is ALREADY parked on a nested wait (a context holds one
park); or the Asyncify instrumentation closure shifting when the
EM_ASYNC_JS left that translation unit.
- Process: gate a bridge conversion on a SINGLE spec, never the full suite.
This cost a 1.2-hour run to learn one bit.
Tree is unchanged from the last verified landing state (STAR_DISPATCH=0,
kicad 139 passed / 1 pre-existing occ-probe) - docs only.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LBjomQfKyRa3jBdeAKpmTw
The increment doc 22 ordered last round. Every entry that can reach a tool
coroutine - the four DOM callbacks and the mailbox tick - now goes through the
scheduler, so a coroutine is never entered by a star transfer from one path and
a direct symmetric swap from another. At D-on all four canvas-tool specs are
GREEN and the KiCad suite is 136/3 (was 135/5 before the sleep work, with the
tools red throughout).
context_sleep's wake learned the mirror lesson: now that the mailbox runs ON a
context, it must NOT call drain_all from there (drain refuses re-entry, and
should) - it marks ready and lets the outer drain_all perform the entry, with
an armed pump as a backstop.
Recorded honestly, not papered over: two of the three remaining D-on failures
are the timer-park and quasimodal-strand levers, each failing ONE assertion -
"scheduler shim observed the concurrent-park window", expected >0, got 0 -
while fired/done/parked/errors all pass. That counter needs TWO concurrent
in-place Asyncify parks, and the lever stages "timer park x MAIN-LOOP YIELD
PARK"; D5 removed the main loop's Asyncify park, so the overlap cannot occur.
Re-pinning those levers to the post-migration invariant is a Phase F decision
alongside fiber-resume-park's red->green flip, NOT an assert to relax now.
Landing state: STAR_DISPATCH=0, kicad 139 passed / 1 (pre-existing occ-probe).
Next: Phase E - the K1-K7 bridges are the only in-place parks left under a
context, and are exactly what the current()!=0 fallback still tolerates.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LBjomQfKyRa3jBdeAKpmTw
wasm/shims/context_sleep.cpp: a main-thread nanosleep whose frame stands on a
scheduler context that OWNS the stack arms a mailbox wake and yield_parks that
context instead of suspending the stack in place. It lives in the sleep
primitive rather than in tool_manager.cpp on purpose - KiCad and the wx core
stay untouched (CLAUDE.md's fork rule) and the whole K7 class moves at once,
not just TOOL_MANAGER::RunSynchronousAction's spin loop.
MEASURED AT D-ON, and it is NOT what unblocks Phase D. The four canvas-tool
specs still fail, but the trace now names a different cause: the fatal swap is
old=<libcontext ROOT> new=<tool coroutine> with mouseEventHandlerFunc above it
- a DOM mouse handler entering wasm DIRECTLY on the main stack, bypassing the
tick. So one coroutine is entered two ways: by the tick through the dispatch
context as a STAR TRANSFER, and by DOM handlers as a DIRECT SYMMETRIC SWAP. A
capture written by one path cannot be rewound by the other -> index out of
bounds in doRewind. That is section 7 rule 5 (partial migration is worse than
none) in its purest measured form, and it is why the harness stays green: its
coroutines are only ever entered from one place.
So the next increment is the DOM event entries (mouse/key/wheel/resize must
hand their events to the dispatch context as the tick does), not another park
site. It subsumes the one-root work too: with no dispatch on the main stack,
resolve_root_identity() always answers "the running context".
Landing state: STAR_DISPATCH=0, kicad 139 passed / 1 (pre-existing occ-probe)
= baseline, with the sleep shim in and inert.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LBjomQfKyRa3jBdeAKpmTw
Traced, not inferred: TOOL_MANAGER::RunSynchronousAction spins
`while(synchronousControl == STS_RUNNING) { wxYield(); wxMilliSleep(1); }`
(tool_manager.cpp:370-371), and on wasm wxMilliSleep -> nanosleep ->
__wasm_main_thread_yield_ms is an ASYNCIFY PARK OF THE STACK IT STANDS ON —
in a loop, inside a tool body, with a nested wxYield() dispatch running on
top of the parked stack.
The case closes on its callers: they are exactly the tools whose specs died
at D-on — edit_tool_move_fct (move-with-m, presence-locks move),
sch_drawing_tools (draw wires), the drawing/edit tools behind draw-lines.
The pcbnew spec comments already described the symptom from outside ("the
asyncified pointer-move handler") without naming the park; this is it.
So doc 21's K7 "anywhere" class has ONE caller that blocks Phase D, and it
moves first: the wait must yield the owning context instead of sleeping in
place, with the atomic's transition marking it ready. Recorded with the fix
shape, the upstream-divergence question it raises, and the gate that matters
(the KiCad suite's four canvas-tool specs — the harness has no
RunSynchronousAction and stays green either way).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LBjomQfKyRa3jBdeAKpmTw
Phase B increment recorded in §10. The wx battery is GREEN at D-on (395/1,
the 1 pre-existing) with dispatch contexts, context waits and star transfers
all live - the first clean battery of the migration. Gaps 1 and 2 from the
D-on probe are closed (terminal coroutine finish; wake/refusal semantics),
and a third containment was found and added: an exception escaping a handler
propagates out through drain()'s fiber swap and would otherwise leave the
registry mid-transition, dead-pumping every later wait. Shim carries the new
abandon call; .ci-cache-epoch -> 12.
THE BOUNDARY: on the full KiCad suite D-on loses four canvas-tool specs
(draw-wires, draw-lines, move-with-m, presence-locks move) to `index out of
bounds` in doRewind - the blue screen itself. Real tool coroutines park IN
PLACE inside their bodies, and a star transfer over an already-parked stack
rewinds state the fiber layer cannot see; the harness's coroutines yield
cleanly, so it goes green while KiCad does not (the doc-19 lesson again:
the harness models the shape, not the parks). So D cannot carry KiCad until
the tool-body park sites are contexts too - C+E completion, which §5 already
ordered before the flip. This measurement makes that ordering non-negotiable.
Landing state verified: STAR_DISPATCH=0, kicad 139 passed / 1 (pre-existing
occ-probe glb) = the Phase A baseline exactly, wx battery 395/1.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LBjomQfKyRa3jBdeAKpmTw
The D-on probe (wxWASM_STAR_DISPATCH=1 on top of D5) is measured and
recorded in doc 22 §10: the week's 'environmental cliff' was adopted-fiber-
stack alignment (trap 1, fixed in wx - retires the D5 entry's findings 2-3
with a mechanism); with it fixed D5+D reached 388 passed; the remaining
reds decompose into three named Phase B gaps (wake ordering on a running
context, finished-coroutine transfer livelock, invocation-aware root
routing), each with a deterministic repro suite. D parks again behind the
switch until Phase B owns coroutine lifetimes.
app-quit.spec.ts + a wx_test_quit hook in minimal_test drive a real
File->Quit-shaped exit through D5's detached teardown (loop exits on its
context, S6 latch 'clean', OnExit + wxUninitialize there) - green in the
D-on probe, closing the D5 teardown-gate open item.
Landing state (D off): coroutine + coroutine-pthread + app-quit green;
nested red at case 3 in the pre-existing doc-19 wake-window family
(cases 1-2, yesterday's blockers, now pass); races wakeup_during_transition
stays layout-sensitive. Both are the class the C+B+E flip removes.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LBjomQfKyRa3jBdeAKpmTw
D5 (main loop on a context, DoRun returns) is implemented and measured:
387 passed with D5 on, identical failure set with D5 off. The §10 D5 entry
records the working-tree bisection of the remaining reds, none of them D5:
1. the WIP jump_fcontext->fiber_transfer libcontext change is not
behavior-preserving even dormant (kills nested baseline_fiber_alone;
passes with the file reverted to Phase A) - fix before the flip, and it
corrects this doc's earlier attribution of the WIP-tip failure to D;
2. the Phase A scoreboard is stale for coroutine-nested TODAY: at the
exact Phase A state it dies at fiber_yield_across_modal_close -
environment-sensitive, re-baseline before attributing;
3. races wakeup_during_transition flips with wx binary layout alone.
All three are the hot-main-swap-out wake-window class that C+B+E remove
structurally - expect those harnesses green AT the flip, not before.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LBjomQfKyRa3jBdeAKpmTw
main -> wxEntry -> OnRun -> MainLoop -> DoRun, and wx/app.h:102 says plainly
that when OnRun returns the program starts shutting down. So DoRun returning
propagates into wx teardown - which collides with the standing rule to keep
fixes inside the wasm port. Records the three options in evaluation order
(wasm-port OnRun override first, __WXWASM__-gated core change second, parking
DoRun rejected with its reason) and the teardown surface the gate must cover.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EarUW9DS1c1sSW4ZNrkGQS
The per-frame park is how a synchronous C++ frame waits; a frame that must not
park can only RETURN. So D5 = loop body on a context + DoRun returns + main()
returns without tearing wx down (EXIT_RUNTIME=0). Step 3 is the real cost and
needs its own gate - teardown, ~wxTopLevelWindowWasm, the S6 shutdown latch
and the v0.1.28 quit-notify fix all assume DoRun returning means the app ends.
Checked rather than assumed that no smaller fix exists: the scheduler stack IS
the main stack, so while it is parked in the rAF yield any pump entry re-enters
a logically suspended stack, and the loop parks every frame.
Recommends doing D5 first and alone (wx-only, cheap to iterate) before
re-enabling the already-written D/C/B/E on top.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EarUW9DS1c1sSW4ZNrkGQS
NOT GREEN. resolveWait marks a context-parked waiter ready and arms a pump
instead of resolving a promise nobody awaits.
The doc records the measured correction: DoRun parks the MAIN stack every
frame in wxWasmYieldToBrowser, which doc 21 called safe-by-construction only
because dispatch also ran there. With the scheduler swapping contexts from the
tick, those interleave over one currData - overlapped-wake, the exact class
this work exists to remove. So the main loop must become a context (D5) and
the bridges (E) join the same flip: D5+D+C+B+E land together.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EarUW9DS1c1sSW4ZNrkGQS
The scheduler fiber and libcontext's root would both adopt the main stack -
two emscripten_fiber_t describing one stack, mutual corruption on first entry.
Harmless in Phase A (KiCad never called drain), fatal at the flip. So dispatch
must move onto a context FIRST, the libcontext root then adopts the running
context rather than the main stack, and only then can jump_fcontext become a
star transfer.
Also records that the synchronous-Call question is answered and pinned.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EarUW9DS1c1sSW4ZNrkGQS
star_transfer_call_is_synchronous stages exactly the TOOL_MANAGER shape -
Call() then read Running() on the next line - and pins that a parked caller
resumed by the scheduler is indistinguishable from a synchronous return.
star_transfer_chain pins a three-deep nested call (root -> A -> B and back) in
strict order, with a parked star context proved undisturbed throughout.
A rule learned the hard way here: a fiber entry must NEVER return (emscripten
ends the program if it does), so these bodies park forever once their script
is done, like libcontext's trampoline loops. Getting it wrong wedges the
battery with no output at all.
The fiberRunning assertion flips to the stronger Phase B invariant: once the
pump is quiescent NO fiber is on the CPU, the scheduler is.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EarUW9DS1c1sSW4ZNrkGQS
The synchronous TOOL_MANAGER contract vs the star: Call() must run to first
yield before returning, drain() refuses re-entry, and dispatch is not yet a
context - the same knot from the other side, and why B+C+D are one commit.
CALL_CONTEXT::Continue is a hand-rolled scheduler that should disappear rather
than be ported. Phase A's grace-ring-over-capacity: 33 says B must fix
coroutine LIFETIME, not just topology. Records the upstream-divergence
decision B has to take first.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EarUW9DS1c1sSW4ZNrkGQS
Bumps wxwidgets (fiber lane in sched_context.h) and kicad (libcontext's wasm
backend becomes an adapter over it), so there is now ONE party performing
every emscripten_fiber_swap and recording who is on the CPU. Behaviour is
preserved throughout - libcontext still decides, the registry observes and
beacons any disagreement - which is the de-risking step doc 20's D2 never had.
Harness + spec gain three fiber-lane scenarios: a fresh fiber enters at its
entry and a swap suspends the swapper (fiber_roundtrip), releasing a suspended
fiber is legal while a stale id refuses instead of use-after-free
(fiber_release_suspended), and symmetric swaps leave a parked star context
undisturbed (fiber_and_star_coexist). The spec asserts the lane's counters,
including the tripwire that must stay zero: fiberNonEnterableSwaps.
Doc 22 gains the Phase A work log: the gate (kicad suite 139/1, the 1 being the
pre-existing occ-probe glb matrix; wx 346/1/3; batteries 48/48; zero tripwires
anywhere), the four bugs the tripwires caught and the single rule behind three
of them, the grace ring and the 33-coroutine measurement Phase B needs from it,
and the process traps this run paid for.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EarUW9DS1c1sSW4ZNrkGQS
Doc 20 gets a status table instead of a narrative: D-1/D0/D1 done, D2
reverted, D3's goal met but its MEANS skipped (waits still park in
place — only the stack they park on changed), D4/D5/D6 not started. Plus
an explicit "what was skipped and is still owed" list, so nothing quietly
reads as delivered: dispatch contexts, waits-as-yields, bridges,
real-park-site buffer sizing, the both-EH CI matrix.
Doc 22 is the handoff plan a fresh session can implement from:
- the bug it targets is the BLUE SCREEN (a context recovered twice or by
the wrong fiber), not doc 19's hang, which is fixed and was only one
ingredient;
- the diagnosis that matters: bookkeeping is already centralised
(currData is single-writer, the shim sees every swap) but the DECISION
is not — libcontext decides swaps and never tells anyone, so three
layers each invented heuristics. Centralising means moving the
decision;
- why D2/D3 knotted, with the measurements, so nobody retries that order;
- phases A-F with estimates (~4-6 wk): A absorbs libcontext behaviour-
preservingly and is landable ALONE (the de-risking step D2 never had),
B+C+D must land as one commit, E bridges, F deletes the guards only
once they are provably silent;
- gates including the crash's own repro, since the batteries never
reproduced it and therefore cannot certify it;
- the traps this run paid for: 16-byte fiber stack alignment, the
per-fiber stack-limits gotcha, host-side asyncify buffers, the
stack-ownership rule, partial migration, and the gitignored-artifact
triage trick;
- measured build/test cycle costs, so the estimates are grounded.
README: index rows for 21 and 22, doc 19 marked FIXED, and the stale
2026-06 "single decisive next step" folded away behind the current one.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
Doc 19: status DIAGNOSED+PINNED RED -> FIXED, with the mechanism and the
explicit note that waits still park in place; the cure was to stop
parking on a COROUTINE stack, not to stop parking.
Doc 20: D3 work log (layering table, gate numbers) and a status that no
longer claims a phase queue. The bug that motivated the whole core
rewrite is closed WITHOUT contexts, so D2/D4/D5/D6 have lost their
forcing function and must be re-justified rather than continued by
default — including the honest option of deleting D1's unused context
layer.
Trap recorded: detection must capture the main stack's bounds at
top-level DoRun, not query them live — finishContextSwitch resets the
limits to the incoming fiber's, so a live query reports "main stack"
from everywhere and detects nothing.
Gate: kicad 139 passed / 1 failed (pre-existing occ-probe glb); wx +
asyncify + coroutine 394 passed / 1 failed (pre-existing
environment-sensitive modal:125). coroutine-nested — the battery D2
regressed — is green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
THE BUG IS FIXED. tests/kicad/quasimodal-strand.spec.ts flips from a
test.fail() pin to a plain green regression test: 3/3 runs
closed=true dialogs=0 refused-resumes=0 (was closed=false dialogs=1
refused-resumes=1 on every run).
Mechanism: a quasi-modal's nested event loop parked on the TOOL
COROUTINE's stack, which suspends the fiber's body where the fiber layer
cannot see it — so the stale-fiber guard quarantined the fiber and then
refused its own resume, the dispatch guard was never released, and every
click after that was deferred forever. Bouncing the nested loop onto the
main stack leaves the coroutine suspended the legitimate way (a recorded
fiber swap), so nothing is quarantined and nothing is refused.
Layering, so this is not a pile of WASM ifdefs in KiCad:
- wx (3d37db3bf1) owns the POLICY and the hook; it must not know what a
coroutine is.
- wasm/bindings/main_stack_runner.h is the only place that may know both
sides: it finds the frame's TOOL_MANAGER and bounces via RunMainStack.
Header-only and self-installing, so no build-script change; included by
every editor's binding TU.
- KiCad gets ONE ifdef-free method (2c777efede), needed only because
TOOL_STATE is opaque outside TOOL_MANAGER. libcontext and dialog_shim
are untouched — an earlier draft edited both and was reverted.
This also reframes the remaining plan: the doc-19 class is closed WITHOUT
migrating tool coroutines onto scheduler contexts. Note it does not make
the wait a context yield — waits still park in place, just never on a
coroutine stack.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
Bumps wxwidgets fc762bed (yield_park refuses a yield from a foreign
stack).
D3's phase text assumed the wait's caller runs on the context it yields.
The doc-19 path does not: the dialog is opened from a KiCad tool
coroutine — a libcontext fiber above whatever dispatched it — so a yield
would save the tool fiber's stack into the host context's fiber struct.
Guarded now, and the rule is explicit: a wait yields the context that
OWNS its stack.
That makes D3 three things, not one: tool coroutines become scheduler
contexts, waits yield the owning context, and dispatch moves onto a
context (the old D2). They are atomic — D2 alone regressed the nested
battery, and D3 alone has no context to yield. §5's architecture already
listed "tool contexts" among the three kinds the scheduler drains; only
the phase text was narrower.
Also pinned green: fiber-on-context nesting works. An earlier buggy
scenario trapped and briefly suggested otherwise, which would have sent
the plan somewhere wrong.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
Two scenarios, deliberately separate because they answer different
questions that D3 depends on:
- fiber_nests_in_context: can a libcontext-style fiber run nested inside
a scheduler context at all? YES. Worth pinning explicitly — an earlier
version of this scenario trapped `unreachable` and briefly looked like
proof that nesting was impossible, which would have been a serious
wrong turn for the plan. The real cause was the test's own bug: the
host-side emscripten_fiber_t was zero-initialised, so the swap unwound
into a null asyncify buffer. Nesting works; the scenario now proves it.
- foreign_stack_refused: a yield from that nested fiber is REFUSED, and
the host context and scheduler both survive it.
The spec asserts foreignStackRefusals == 1, so the refusal path cannot
silently stop being exercised.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
Bumps wxwidgets d220aae5 (D2a: sched_context.h moved into wx's port,
header-only so evtloop.cpp can see it) and 5ee60a81 (D1 fix: 16-align
context stacks — EM_ASM's arg buffer lives on the running stack and the
glue asserts buf % 16 == 0, so misaligned contexts trapped in
readEmAsmArgs; std::vector<char> only gives malloc's 8-byte alignment).
The dispatch switch itself is NOT landed. Running the tick's
ProcessEvents on a context took the battery from 363 green to 388/7, six
of them the coroutine-nested harness wedging at
fiber_create_run_destroy_inside_modal via aliased-wake-live ->
fiber-resume-refused — doc 19's mechanism. A quasi-modal opened from a
tick handler suspends the dispatch context INSIDE the still-in-place
wait, putting one more Asyncify layer under every libcontext fiber.
Pooling contexts (8 burned in 30 ms) and falling back to entry-stack
dispatch both failed to avoid it, because the layer exists as soon as
the context is suspended.
That is doc 20's own risk 2 arriving on schedule, so the plan is
corrected rather than the symptom patched: D3 (waits become context
yields) must come first, after which the dispatch context is released at
its tick boundary instead of suspended and the failure class is
structurally absent. Verified the revert: the nested battery is green
again at this baseline (5 passed / 1 failed, the 1 being the
environment-sensitive modal:125 that also fails without any of this).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
Doc 20: D1 entry (star topology as the load-bearing decision, the
9-scenario gate, measured memory) and the §3 asset table flipped —
registry/park/resume/drain are no longer "stubs that throw".
Doc 21: what D1 built, plus the sizing instruction for D4 — take each
bridge's deep-park high-water from its own beacon; do NOT carry the
harness's ~34 B/frame figure (synthetic frames, three locals each = a
floor) and do not inherit libcontext's 512 K by default either.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
Doc 20 §6 D1, second slice: the dedicated test app D1 calls for, plus
the doc 20 §7 risk-1 gate.
tests/apps/standalone/sched-context/ drives the primitives with no wx
linked (a failure can only be the contexts layer) and assertions ON —
unlike the races harness, an emscripten "cannot start an async
operation" state here IS the bug, not a tolerated state. Nine
scenarios, each pinning one invariant; the load-bearing ones:
- parked_does_not_block: a parked context does not stop others running.
This is doc 19's freeze made unrepresentable — there a parked activity
held a global interlock and the UI died; here "parked" simply means
"not runnable", and three workers run past it before it resumes.
- one_transition_in_flight: a context calling drain() gets a no-op, with
a second ready context queued so a buggy nested drain would actually
run something and be caught.
- async_wake: a real macrotask hop (setTimeout → mark_ready → drain),
the shape every production bridge has.
- deep_park_sizing: parks 64 live frames deep to measure what a park
actually costs.
The memory gate asserts the ceiling (peak live contexts, peak bytes),
that nothing leaked (live=0, bytes=0, created==finished), that nothing
was left mid-transition, and that refusals occurred (zero would mean the
illegal-operation scenarios stopped provoking).
Measured here: ~34 B/frame, 2200 B for a 64-frame park, 1 MB peak for 4
concurrent contexts. Recorded with the caveat that the harness's frames
carry three locals each, so this is a FLOOR — real bridges save far more
per frame (libcontext runs 512 K after a 64 K buffer silently
overflowed). The apparatus and its units are validated; the sizing
DECISION needs deep-park numbers from real bridges at D3/D4.
playwright.config: asyncify-firefox now matches every spec in ./asyncify
instead of only asyncify-races* — a new harness there is covered by
construction rather than by remembering to widen the pattern.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
Doc 20 §6 D1, first slice: wasm/sched/context.{h,cpp} implements
create / yield_park / mark_ready / drain with the registry as truth.
The shape that matters — a STAR, not libcontext's symmetric swap:
contexts only ever swap OUT to the scheduler, and only the scheduler
swaps IN. So "is this target safe to enter?" stops being a guess
(libcontext's swap_suspended / parked / hot-main refusals) and becomes a
lookup: the registry says Parked/Ready and holds the buffer. Doc 19's
refused-resume is unrepresentable here because resume is not a decision
made at the swap site.
Enforced, not hoped for:
- at most one transition in flight (drain() refuses re-entry, so a
context calling drain() cannot turn the star into a cycle);
- mark_ready() never resumes inline — it queues, and drain() resumes
from a clean stack (doc 13 §1.4's deferred-wake law, per context);
- yield_park() off a context is REFUSED, which is the "nothing parks in
place" rule made mechanical;
- destroy() on a non-Finished context is refused (freeing a parked
stack strands whatever is on it);
- FIFO ready queue (no starvation);
- main-thread-only, per doc 21 §2's pthread finding.
Memory is accounted from the start (doc 20 risk 1): live/peak contexts,
bytes/peak bytes, and per-context asyncify high-water measured from
asyncify_data.stack_ptr — the same quantity the shim reports as `rem=`,
from the other end. Sizes deliberately start at 128 KB C stack + 128 KB
asyncify buffer rather than inheriting libcontext's 512 K, so the number
gets derived from evidence; a >75% buffer use beacons BUFFER-PRESSURE
because that overflow is silent corruption, not a crash.
No production path runs on this yet — dispatch moves at D2.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
Doc 19: status DIAGNOSED -> DIAGNOSED + PINNED RED; the repro section
leads with the automated deterministic spec (the manual Leonardo flow
stays as the field repro).
Doc 20: D0 work-log entry; status D-1 + D0 done; next = D1.
Doc 21: red-spec section updated to "landed", with the staging test's
anti-vacuity role.
Carried finding: the strand reproduces on a 2-OBJECT fixture schematic,
so warm-load byte volume (the 68/1 dice-loader) is not an ingredient —
two concurrent parks suffice. That is why D3 closes this class and D5 is
not required for it.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
Doc 20 D0 second deliverable (doc 21 §4). tests/kicad/quasimodal-strand.spec.ts
reproduces the Symbol Properties hang on demand, in two parts:
- "staging" (GREEN): double-click the fixture symbol → Symbol Properties
opens → arm the parking timer (wasm/bindings/timer_park.h) WHILE the
dialog is up. Because the opener's fiber park is open-ended, a timer
firing now necessarily parks on top of it — the overlap is structural,
not a won race. Asserts the dialog opened, the timer fired, its OK
button is hittable, and the shim beaconed concurrent contexts. Keeps
the red pin from rotting into vacuity, and fails loudly on its own.
- "doc-19 red" (test.fail()): clicks OK and asserts the desired end state
— dialog closes, zero fiber-resume-refused beacons, wait books balanced
(no unresolved nested/modal wait). Goes green at D3, when Playwright
will report "expected to fail but passed" and the marker comes off.
Verified 6/6 consecutive full-file runs, identical outcome each time:
closed=false dialogs=1 refused-resumes=1 — the doc-19 mechanism exactly
(quarantined fiber's legitimate resume refused, dialog never closes).
Deliberately NOT asserted in staging: the timer park COMPLETING (whether
a park survives the aliasing is the disease under test) and sawParked (a
100ms sampler can miss a short park) — both are reported, not gated.
Retires tests/kicad/dialog-deadlock-probe.spec.ts: the 8/4 throwaway probe
that established the mechanism. This spec supersedes it and, unlike it, is
deterministic (the probe's 3 blind waitForTimeouts were the only
determinism-lint violations in the tree; the guard is now clean).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
Doc 20 D0 first deliverable: every Asyncify park site post-D-1,
classified by whose stack it suspends (tool fiber / entry stack / main
loop) with a routing decision per site — 14 production sites (8 wx, 9
KiCad/bridge counting quartets/pairs) + 3 deliberate test levers.
W1 (wxWasmYieldUntilJs) and W3 (popup) route to D3 context yields;
the clipboard/font/lib/3D/occ/ngspice/nanosleep bridges route to D4;
W2 (per-frame yield) stays safe-by-construction unless D5 is taken.
Also settles doc 20 risk 4 (pthreads): all parks are main-thread only;
the lib bridge's pthread path is blocking-proxy, not Asyncify, and is
out of migration scope. Defines what D4's "no handleSleep on a fiber
stack" assertion must cover.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
Gate results (single battery, scheduler-only): wx app battery +
asyncify + coroutine 363 passed / 3 skipped / 0 failed; full kicad
suite 138 passed / 30 skipped / 1 failed — the one failure is the
pre-existing local occ-probe glb case (predates D-1, unrelated).
Post-gate scrubs: vestigial 'startModal' removed from ASYNCIFY_IMPORTS
(tests/apps/Makefile.wasm) and env.startModal from
scripts/common/asyncify-imports.txt (the import no longer exists in any
wasm; both lists are boundary supersets so behavior is identical).
Stale comment pointers to the deleted legacy modal machinery updated.
Doc 20: D-1 work log added, status flipped to IN PROGRESS.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
