open-standards/.github/workflows/wai-byte-equality.yml
Transaction Science e694cb0966
Some checks failed
WAI breaking changes and the conformance runner / WAI breaking · does this change touch WAI (push) Successful in 12s
joule-moq-remote host agent / joule-moq-remote · does this change touch it (push) Successful in 21s
wai-quantum energy-landscape jobs / does this change touch it (push) Successful in 15s
joule-moq-room room service / joule-moq-room · does this change touch it (push) Successful in 23s
wai-app and Create / wai-app · does this change touch it (push) Successful in 23s
jouleclaw public surface must stay documented / Public-surface audit (jouleclaw-rs) (push) Successful in 32s
WAI operation catalogue / does this change touch it (push) Successful in 21s
wai-key-proxy / does this change touch it (push) Successful in 13s
wai-quantum wasmQ / does this change touch it (push) Successful in 20s
wai-ledger service / does this change touch it (push) Successful in 31s
energy meters on real hardware / RAPL and NVML on jetson-hub (push) Successful in 1m5s
wai-ops integer operators / does this change touch it (push) Successful in 30s
joule-code conformance gate / joule-moq-relay, its own energy on a counter (macos) (push) Successful in 1m13s
wai-quantum contract / does this change touch it (push) Successful in 25s
wai-quantum command line / does this change touch it (push) Successful in 24s
wai-det-jpegai WAISM float-level module of record / does this change touch it (push) Successful in 56s
wai formal verification (integer core) / does this change touch it (push) Successful in 1m0s
wai-quantum docs, course and locales / does this change touch it (push) Successful in 12s
wai-audit collector and player integration / does this change touch it (push) Successful in 1m18s
smart-byte conformance gate / does this change touch it (push) Successful in 1m41s
joule-moq-web browser client / joule-moq-web · does this change touch it (push) Successful in 1m18s
wai-int-module WAISM module of record / does this change touch it (push) Successful in 1m8s
wai-quantum wasmQ / the reference modules rebuild to their modules of record · linux (push) Successful in 42s
smart-byte conformance gate / Spec changes require vector-pack updates (push) Has been skipped
wai-fleet service / does this change touch it (push) Successful in 1m11s
wai cross-architecture byte-equality / does this change touch it (push) Successful in 1m13s
wai-quantum resource estimation / does this change touch it (push) Successful in 23s
wai-quantum editors / does this change touch it (push) Successful in 29s
wai-quantum other formats / does this change touch it (push) Successful in 26s
wai-quantum OpenQASM reader / does this change touch it (push) Successful in 24s
wai-app and Create / wai-app · tests, clippy, the line language's verifier (linux) (push) Successful in 1m15s
WAI breaking changes and the conformance runner / WAI breaking · wire, record and API against main · linux (push) Successful in 1m21s
wai formal verification (integer core) / every harness in one shard (push) Successful in 18s
wai-ops integer operators / independent checker (Python) (push) Successful in 31s
joule-moq-remote host agent / joule-moq-remote, every target (macos) (push) Successful in 1m22s
wai-quantum-landscape-fpga reference design / does this change touch it (push) Successful in 34s
wai-quantum variational and algorithm library / does this change touch it (push) Successful in 16s
wai-quantum for JavaScript / does this change touch it (push) Successful in 39s
wai-quantum verdicts and repairs / does this change touch it (push) Successful in 18s
wai-quantum pulse and dynamics / does this change touch it (push) Successful in 29s
wai-quantum views in agent hosts / does this change touch it (push) Successful in 20s
wai-ops integer operators / wasm32 under Node (push) Successful in 47s
wai-sink-android native sink library / does this change touch it (push) Successful in 19s
WAI Studio / does this change touch it (push) Successful in 16s
wai-quantum error correction and mitigation / does this change touch it (push) Successful in 46s
wai-key-proxy / proxy · linux (push) Successful in 1m34s
wai-quantum extension points / does this change touch it (push) Successful in 50s
energy meters on real hardware / sink decode table · per-task energy and the GPU rail on macOS (push) Successful in 2m31s
wai-quantum for Python / does this change touch it (push) Successful in 51s
wai-ops integer operators / wasm32-wasip2 (wasmtime) (push) Successful in 1m7s
wai-sink-swift native sink package / does this change touch it (push) Successful in 28s
wai-quantum studio and desktop / does this change touch it (push) Successful in 49s
wai-quantum service / does this change touch it (push) Successful in 51s
wai-ledger service / service · macos (push) Successful in 1m22s
wai-vector vector graphics / does this change touch it (push) Successful in 27s
wai-ops integer operators / linux-x86_64 (push) Successful in 1m31s
wai-ops integer operators / macos-arm64 (push) Successful in 1m34s
wai-quantum wasmQ / profile, engines, reference modules, adversarial, vectors · macos (push) Successful in 2m55s
wai-ledger service / service · linux (push) Successful in 2m49s
wai-fleet service / service · macos (push) Successful in 1m2s
wai-audit collector and player integration / collector, player integration and the browser end to end · macos (push) Successful in 2m7s
wai-quantum docs, course and locales / the Studio's words in every locale · linux (push) Successful in 13s
wai cross-architecture byte-equality / macos-arm64 (reference) (push) Successful in 40s
wai-quantum command line / command line · macos (push) Successful in 1m57s
wai-quantum wasmQ / interpreter and vectors under wasm32-wasip2 (wasmtime) (push) Successful in 3m48s
joule-code conformance gate / Build + round-trip C1-C4 vectors (push) Successful in 5m23s
energy meters on real hardware / sink decode table · RAPL on jetson-hub (NVML surveyed) (push) Successful in 4m5s
wai-quantum contract / the policy is linked, and the deprecations page reads (push) Successful in 6s
wai-quantum wasmQ / profile, engines, reference modules, adversarial, vectors · linux (push) Successful in 5m2s
wai-quantum pulse and dynamics / the same bits · macos (push) Successful in 2m7s
wai formal verification (integer core) / kani (stage) (push) Successful in 4m55s
smart-byte conformance gate / Validate conformance JSON files (push) Successful in 8s
wai-quantum editors / editors · macos (push) Successful in 3m5s
smart-byte conformance gate / Second implementation reproduces every vector (push) Successful in 9s
wai-quantum energy-landscape jobs / the same bits · macos (push) Successful in 6m37s
smart-byte conformance gate / VERSIONS.md and README.md must exist (push) Successful in 9s
wai-audit collector and player integration / collector · linux (push) Successful in 2m5s
wai-int-module WAISM module of record / the module of record in Chromium (push) Successful in 18s
wai-quantum docs, course and locales / the course and the command line's languages · linux (push) Successful in 5m7s
wai-quantum service / service energy · linux packages (push) Successful in 2m45s
wai cross-architecture byte-equality / linux-x86_64 (push) Successful in 53s
WAI breaking changes and the conformance runner / WAI breaking and conform · their tests, and the independent verifiers through wai conform · linux (push) Successful in 8m52s
wai formal verification (integer core) / kani (integer-core) (push) Successful in 7m27s
wai-quantum energy-landscape jobs / native against WebAssembly, over many seeds (push) Successful in 9m26s
wai-app and Create / wai-app · Create, browser end-to-end (macos) (push) Successful in 9m29s
wai-quantum energy-landscape jobs / reader, kernels, answers and the stated result · linux, release (push) Successful in 10m15s
wai-fleet service / service · linux (push) Successful in 3m19s
WAI operation catalogue / the catalogue and every door · linux (push) Successful in 10m55s
wai-quantum for Python / python · macos (push) Successful in 4m45s
wai-quantum OpenQASM reader / corpus under wasm32-wasip2 (wasmtime) (push) Successful in 55s
wai-quantum contract / API compatibility · linux (push) Successful in 8m0s
WAI Studio / Studio in Chromium and Firefox · macos (push) Successful in 7m51s
wai-quantum studio and desktop / desktop app · macos (push) Successful in 6m55s
wai-quantum other formats / the set under wasm32-wasip2 (wasmtime) (push) Successful in 3m4s
wai-quantum OpenQASM reader / reader and corpus · linux (push) Successful in 2m21s
wai-sink-android native sink library / every ABI, the AAR, and the instrumented tests in a headless emulator · macos (push) Successful in 8m48s
wai formal verification (integer core) / kani (arith) (push) Successful in 11m38s
wai-quantum other formats / readers, writers and the set · linux (push) Successful in 4m17s
wai-quantum variational and algorithm library / parameters, gradients, optimisers, VQE, QAOA and builders · linux, release (push) Successful in 2m16s
wai-vector vector graphics / macos-arm64 (push) Successful in 2m49s
wai-quantum verdicts and repairs / the intent set and the verdict under wasm32-wasip2 (wasmtime) (push) Successful in 1m44s
wai-quantum editors / editors · linux (push) Successful in 6m6s
wai-quantum contract / schemas and the committed tool lists · linux (push) Successful in 11m2s
wai-quantum service / service · macos (push) Successful in 5m56s
wai-quantum pulse and dynamics / reader, simulator and pulse jobs · linux, release (push) Successful in 2m45s
wai-int-module WAISM module of record / linux · both checkers, wasmtime (Cranelift, Winch), Wasmi and Node (push) Successful in 8m56s
wai-quantum pulse and dynamics / reader, simulator and pulse jobs under wasm32-wasip2 (wasmtime) (push) Successful in 2m2s
wai-quantum verdicts and repairs / verdicts, repairs and the intent set · linux (push) Successful in 3m44s
wai-quantum for JavaScript / javascript · linux (push) Successful in 4m41s
WAI Studio / the catalogue's actions, the unit tests and the core · linux (push) Successful in 1m30s
wai-quantum energy-landscape jobs / the module under wasm32-wasip2 (wasmtime) (push) Successful in 16m38s
joule-code conformance gate / joule-moq-relay, every target (push) Failing after 5m40s
wai-quantum extension points / the registry under wasm32-wasip2 (wasmtime) (push) Failing after 1m32s
wai-quantum resource estimation / estimation and calibration · linux, release (push) Successful in 9m28s
wai-sink-swift native sink package / libwai slices, desktop tests, simulator tests · macos (push) Successful in 11m46s
wai-quantum views in agent hosts / views · the page and its engine (push) Successful in 4m41s
wai-vector vector graphics / wasm32-wasip2 (wasmtime) (push) Successful in 1m11s
joule-code conformance gate / joule-moq-ingest, every target (push) Successful in 7m3s
joule-moq-remote host agent / joule-moq-remote, every target (linux-arm64) (push) Successful in 1m51s
wai-quantum error correction and mitigation / codes, decoders, texts and mitigation · linux, release (push) Successful in 3m47s
wai-vector vector graphics / wasm32 under Node (push) Successful in 39s
joule-moq-web browser client / joule-moq-web · browser end-to-end (macos) (push) Successful in 17m57s
wai formal verification (integer core) / kani (plan) (push) Successful in 18m32s
wai-vector vector graphics / fuzz, peak allocation, comparison count (push) Successful in 19s
wai-quantum extension points / the registry and the example · linux (push) Failing after 4m38s
wai-vector vector graphics / linux-x86_64 (push) Successful in 3m4s
wai-ops integer operators / linux-aarch64 (push) Successful in 2m3s
wai-det-jpegai WAISM float-level module of record / linux · both checkers, F1, F2 on wasmtime (Cranelift, Winch, Pulley) and Wasmi, the planted faults (push) Successful in 20m30s
wai-quantum views in agent hosts / views · a scripted host (push) Successful in 7m14s
wai-quantum service / service · linux (push) Failing after 5m10s
wai cross-architecture byte-equality / linux-aarch64 (push) Successful in 52s
wai-vector vector graphics / wasm32 in three browser engines (push) Successful in 2m57s
wai-vector vector graphics / independent checker (Python) (push) Successful in 3m15s
wai-quantum-landscape-fpga reference design / the board's code on the board's processor's architecture · linux-arm64, release (push) Successful in 1m32s
wai-quantum for Python / python · linux (push) Successful in 8m18s
repository checks / lockfiles, advisories, tests in CI, workflow hygiene, spec and section references, public text (push) Successful in 10m50s
wai formal verification (integer core) / kani (sound) (push) Successful in 24m18s
wai-int-module WAISM module of record / linux-arm64 · both checkers, wasmtime (Cranelift, Winch), Wasmi and Node (push) Successful in 6m33s
wai release binaries build on Windows / the seven release binaries · windows (push) Failing after 25m52s
wai-quantum variational and algorithm library / parameters, gradients, optimisers, VQE, QAOA and builders · linux-arm64, release (push) Successful in 2m56s
joule-moq-room room service / joule-moq-room, every target (push) Successful in 9m40s
wai-quantum pulse and dynamics / the same bits · linux-arm64 (push) Successful in 2m55s
wai-quantum studio and desktop / studio · browser (push) Successful in 11m54s
wai cross-architecture byte-equality / windows-x86_64 (push) Successful in 2m40s
wai-quantum extension points / the registry and the example · linux-arm64 (push) Failing after 2m58s
wai-sink-android native sink library / the JNI layer and the platform-free half on the host JVM · linux-arm64 (push) Successful in 4m25s
wai-quantum studio and desktop / desktop app · linux (push) Successful in 14m4s
wai formal verification (integer core) / kani (isqrt512) (push) Successful in 30m23s
wai-quantum energy-landscape jobs / the same bits · linux-arm64 (push) Successful in 14m8s
WAI Studio / Studio in WebKit · linux (push) Successful in 17m32s
wai-vector vector graphics / linux-aarch64 (push) Successful in 7m20s
wai-quantum command line / command line · linux (push) Successful in 30m0s
wai formal verification (integer core) / kani (stage-long) (push) Successful in 35m31s
wai-quantum-landscape-fpga reference design / the member against the model of the logic, and the logic against the model · linux, release (push) Successful in 27m59s
wai-quantum docs, course and locales / every command and snippet in the docs · linux (push) Successful in 38m42s
wai-det-jpegai WAISM float-level module of record / linux-arm64 · both checkers, F1, F2 on wasmtime (Cranelift, Winch, Pulley) and Wasmi, the planted faults (push) Successful in 24m21s
wai-quantum studio and desktop / desktop app · windows (push) Successful in 14m27s
wai formal verification (integer core) / kani (rans64-t2) (push) Successful in 13m24s
ci: a push to main takes a concurrency group of its own, so no later push cancels its run
Every workflow grouped its runs per ref with cancel-in-progress: true, and
15 of them called that safe on main because main advanced by merge commits.
It no longer does: work lands as direct commits, several an hour, and
Forgejo hands runners the oldest queued job first, so each cancelled run
was replaced at the back of the queue. With the linux-arm64 queue hours
deep, a workflow main touched more often than that never reached the
front. In the day to 2026-11-10 01:11Z, 578 of the 810 runs main's pushes
started were cancelled; after 13:58Z on 11-09 no run of joule-moq-room or
joule-moq-remote on main finished (31 of joule-moq-room's next 32 were
cancelled, the 32nd waiting), which has held the wai 0.6.0 release check
since 00:49Z.

Every workflow now groups by
  ${{ github.workflow }}-${{ github.event_name == 'push' && github.sha || github.ref }}
with cancel-in-progress: true. A pull request's newer push still
supersedes its older run; a push to main (or a tag) has a group of its own
that no later push shares; a scheduled run keeps its branch's group, which
no push shares now, so only the next scheduled run supersedes it.
wai-breaking, wai-key-keychain, wai-key-windows and
wai-release-windows-build already grouped a push by its commit; the last
gains the explicit cancel-in-progress: true.

Not cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}. Forgejo
16.0.6 does evaluate the expression (EvaluateWorkflowConcurrency in the
runner's jobparser), but false with a group is Forgejo's QueueBehind mode
(ConfigureActionRunConcurrency, services/actions/workflows.go): a run is
not offered to a runner while another run of its group is running or an
older one is pending (getConcurrencyCondition, models/actions/task.go),
and nothing cancels a waiting one. That is GitHub's false without its
one-pending limit: main's runs of a workflow would go one at a time, each
through the whole label queue, and every push would add one more.

What bounds the queue: main holds one run per push that touches a
workflow, and Forgejo cancels a job left waiting 24 hours
(cancel_abandoned_jobs, ABANDONED_JOB_TIMEOUT; the forge keeps the
default), so a runner host that is down cannot keep more than a day's
pushes queued. Estimated from each workflow's last finished main run, a
day of main's pushes with every run kept needs about 950 linux-arm64 and
2700 macos slot-minutes, against at least 6 linux-arm64 slots (8640 a
day) and 3 macos (4320).

scripts/check-workflow-concurrency.sh now requires this block: it still
refuses a missing block and a group that does not vary by ref, and now
refuses a group per ref alone and cancel-in-progress other than true. Its
self-test first checks that the untouched tree passes, then that each
broken form fails with its own error.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-11-09 20:46:02 -05:00

411 lines
23 KiB
YAML

name: wai cross-architecture byte-equality
# The `replicate` determinism contract, checked on real silicon. WAI claims its
# integer-exact decode paths (rANS entropy coder, integer fixed-point synthesis)
# produce BYTE-IDENTICAL output on every machine. This matrix proves it: every
# OS/arch leg of the matrix below decodes the same fixtures and asserts the same golden
# bytes — rANS streams produced by CompressAI's upstream encoder, and an
# int_synth output hash produced on the reference machine (macOS arm64). One
# diverging byte anywhere fails the suite.
#
# Coverage is still-image AND temporal, classical AND learned: the rANS coder,
# integer fixed-point synthesis, the JPEG-AI entropy stage, the inter-frame video
# decoder (`wai.video.int_motion`), and the neural-keyframe variant
# (`wai.video.int_hyper`) whose I-frame comes out of a learned conditional codec,
# and the MLIC++-topology codec (`wai.neural.int_mlicpp`), whose entropy network runs
# before every checkerboard pass and must give every platform the encoder's scale
# bucket for every element. The learned integer audio and Gaussian-splat attribute codecs
# (`wai.audio.int_codec`, `wai.splat.int_codec`) decode their committed WIA1 and WIS1
# payloads to their PCM and attribute digests. The binary payloads SPEC §5 pins (WIV1,
# WIH1, WIA1, WIS1) are parsed
# with the wire parser a sink runs and decoded to the same digests, and the staged-delivery
# residual layers (WIR1) of the staged corpus are applied stage by stage to pinned digests.
# Video is the strictest leg — inter-frame decode is a
# feedback loop, so one divergent half-pel rounding does not stay one byte, it
# predicts the next frame and compounds for the rest of the clip.
#
# The conformance crate (`wai/byte-exact-conformance`) #[path]-includes the
# pure-integer sources from wai-rs — no system codec libraries, no ONNX runtime
# — so every leg runs in about a minute.
# A pull request's newer push supersedes its older run; a push to main does not.
#
# On a pull request only the newest commit's verdict is wanted. With no
# concurrency group, a session pushing four versions of a branch queued four
# complete sets of runs and every one of them had to drain: on 2026-10-05 a PR's
# 21 contexts settled none in 70 minutes behind a backlog of 25 runs, and the
# queue grew faster than it cleared. So a pull request's older runs are
# cancelled.
#
# Verified on this forge (Forgejo 16.0.5) in `ci-test` before being added here:
# the key parses, and a newer push moves an older run to `cancelled` even while
# it is still queued, which is the case that matters.
#
# A push to main takes a group of its own, its commit, so no later push cancels
# its run. Work lands as direct commits to main, several an hour, and Forgejo
# hands runners the oldest queued job first, so a run the next push cancelled
# was replaced at the back of the queue: while the linux-arm64 queue was hours
# deep, a workflow touched more often than that never reached the front. In the
# day to 2026-11-10 01:11Z, 578 of the 810 runs main's pushes started were
# cancelled; joule-moq-room's last run on main to finish ended at 13:58Z on
# 11-09, and 31 of the 32 after it were cancelled. `cancel-in-progress: false`
# would not help: Forgejo then holds each run of a group until the one before it
# has finished, and keeps every one waiting. Forgejo cancels a job left waiting
# 24 hours.
concurrency:
group: ${{ github.workflow }}-${{ github.event_name == 'push' && github.sha || github.ref }}
cancel-in-progress: true
on:
pull_request:
paths:
- 'wai/byte-exact-conformance/**'
- 'wai/wai-rs/src/neural/int_synth.rs'
- 'wai/wai-rs/src/neural/jpegai_mans.rs'
# Every source the crate #[path]-includes must be able to fire this
# workflow, or a change can break cross-arch equality and merge green.
- 'wai/wai-rs/src/int_synth.rs'
- 'wai/wai-rs/src/int_entropy.rs'
- 'wai/wai-rs/src/jpegai_mans.rs'
- 'wai/wai-rs/src/rans.rs'
- 'wai/wai-rs/src/int_video.rs'
- 'wai/wai-rs/src/int_transform.rs'
- 'wai/wai-rs/src/int_hyper_synth.rs'
- 'wai/wai-rs/src/int_mlicpp.rs'
- 'wai/wai-rs/src/int_audio.rs'
- 'wai/wai-rs/src/int_splat.rs'
- 'wai/wai-rs/src/int_payload/wire.rs'
- 'wai/wai-rs/src/staged/predict.rs'
- 'wai/wai-rs/src/staged/wir1.rs'
- 'wai/wai-rs/src/staged/wir1_arith.rs'
- 'wai/staged-conformance/**'
# The fast kernels int_synth, int_transform and int_hyper_synth dispatch to
# by default: the goldens here decode through them.
- 'wai/wai-rs/src/int_kernels.rs'
# The integer operator set and the bridge to it: the goldens decode int_synth,
# the audio and splat codecs, the conditional codec (WIH1, the WIV1 keyframe)
# and the MLIC++-topology codec through their lowerings too.
- 'wai/wai-rs/src/int_ops.rs'
- 'wai/wai-ops/**'
- 'wai/wai-rs/tests/vectors/int_video/**'
- 'wai/wai-rs/tests/vectors/int_hyper/**'
- 'wai/wai-rs/tests/vectors/int_mlicpp/**'
- 'wai/wai-rs/tests/vectors/int_audio/**'
- 'wai/wai-rs/tests/vectors/int_splat/**'
- '.github/workflows/wai-byte-equality.yml'
push:
branches:
- main
paths:
- 'wai/byte-exact-conformance/**'
- 'wai/wai-rs/src/neural/int_synth.rs'
- 'wai/wai-rs/src/neural/jpegai_mans.rs'
# Every source the crate #[path]-includes must be able to fire this
# workflow, or a change can break cross-arch equality and merge green.
- 'wai/wai-rs/src/int_synth.rs'
- 'wai/wai-rs/src/int_entropy.rs'
- 'wai/wai-rs/src/jpegai_mans.rs'
- 'wai/wai-rs/src/rans.rs'
- 'wai/wai-rs/src/int_video.rs'
- 'wai/wai-rs/src/int_transform.rs'
- 'wai/wai-rs/src/int_hyper_synth.rs'
- 'wai/wai-rs/src/int_mlicpp.rs'
- 'wai/wai-rs/src/int_audio.rs'
- 'wai/wai-rs/src/int_splat.rs'
- 'wai/wai-rs/src/int_payload/wire.rs'
- 'wai/wai-rs/src/staged/predict.rs'
- 'wai/wai-rs/src/staged/wir1.rs'
- 'wai/wai-rs/src/staged/wir1_arith.rs'
- 'wai/staged-conformance/**'
# The fast kernels int_synth, int_transform and int_hyper_synth dispatch to
# by default: the goldens here decode through them.
- 'wai/wai-rs/src/int_kernels.rs'
# The integer operator set and the bridge to it: the goldens decode int_synth,
# the audio and splat codecs, the conditional codec (WIH1, the WIV1 keyframe)
# and the MLIC++-topology codec through their lowerings too.
- 'wai/wai-rs/src/int_ops.rs'
- 'wai/wai-ops/**'
- 'wai/wai-rs/tests/vectors/int_video/**'
- 'wai/wai-rs/tests/vectors/int_hyper/**'
- 'wai/wai-rs/tests/vectors/int_mlicpp/**'
- 'wai/wai-rs/tests/vectors/int_audio/**'
- 'wai/wai-rs/tests/vectors/int_splat/**'
- '.github/workflows/wai-byte-equality.yml'
jobs:
# Does this pull request's own diff touch the paths below? This forge matches a
# workflow's `paths:` against the push, so a merge of main into a branch fires the
# workflow whatever the pull request changes; every job here waits on this answer.
# A push to main, a schedule, or anything the script cannot determine answers
# `true` (scripts/pr-touches-paths.py).
relevant:
name: does this change touch it
runs-on: gate
outputs:
run: ${{ steps.gate.outputs.run }}
steps:
- uses: actions/checkout@v5
with:
fetch-depth: 0
- id: gate
run: |
set -euo pipefail
r=$(python3 scripts/pr-touches-paths.py .github/workflows/wai-byte-equality.yml \
"${{ github.event.pull_request.base.sha }}" \
"${{ github.event.pull_request.head.sha }}")
echo "this change touches wai-byte-equality: $r"
echo "run=$r" >> "$GITHUB_OUTPUT"
byte-equality:
needs: relevant
if: needs.relevant.outputs.run == 'true'
name: ${{ matrix.label }}
runs-on: ${{ matrix.os }}
strategy:
fail-fast: false
matrix:
# Self-hosted fleet labels, not hosted images. macos-x86_64 is
# dropped: no Intel-Mac runner exists in the fleet, and by the same
# argument the cargo-classes matrix already makes, it adds no arch
# coverage — x86_64 is held by linux-x86_64 and windows-x86_64, arm64
# by linux-aarch64 and the macos-arm64 reference.
# windows-x86_64 was removed on 2026-09-30 while its runner was offline, and
# is back now that the runner is (a `windows` job passed in run 19310). Not a
# required status, as wai-quantum-webgpu.yml's member-windows is not: like every
# leg here it runs only when the `relevant` gate above answers true. A job on an
# offline runner stays pending and never settles, so if the runner goes offline
# again the leg comes out of this list until it is back.
# The two disk steps below come in a bash form for the unix legs and a pwsh form
# for Windows, whose `bash` is the WSL stub.
include:
- { os: linux, label: linux-x86_64 }
- { os: linux-arm64, label: linux-aarch64 }
- { os: macos, label: macos-arm64 (reference) }
- { os: windows, label: windows-x86_64 }
env:
# The floor of the disk check below, in GiB. The leg writes under 1 GiB (a
# shallow checkout of about 0.3 GiB plus its pack, and a release target under
# 0.1 GiB, measured 2026-11-03), and a cold toolchain install about 1 GiB more
# under HOME; the rest is room for the jobs that share the disk. Measured with the
# step's own loop on 2026-11-04, every unix label had 50 GiB or more free under all
# four directories (the Windows host was not measured), so the floor fails a leg
# only on a host within a few GiB of full. The macOS runner's HOME and TMPDIR sit
# on its internal disk, which has gone from 7 GiB free to 118 MiB within hours, and
# at that level its jobs failed writing under HOME.
DISK_FLOOR_GIB: '4'
steps:
# Infrastructure first, before the checkout writes anything. Main's run 17676
# failed on both arm64 legs with "No space left on device", linux-aarch64 in
# `git fetch` and macOS in the toolchain's temporary files: the hosts' disks were
# full and no fixture was decoded, yet the matrix read as a byte-equality failure.
# This step measures the free space under every directory the leg writes to (the
# workspace, the runner's temporary directory, TMPDIR, and HOME, where the
# toolchain lives), and writes and syncs 32 MiB once on each filesystem, which a
# disk out of inodes or remounted read-only fails though df shows room. Below the
# floor, or on a write refused for space, an I/O error or a read-only filesystem,
# it fails, and its name and its error say infrastructure.
#
# On linux-aarch64 the four directories are on the runner VM's data disk, so df
# reads the guest's free space, not the host volume's. The disk is an image file
# on that volume, and on 2026-11-04 all but 0.7 of its 200 GiB were already
# allocated there, so a full host volume could take back at most that much; the
# write probe catches the rest.
#
# The forge still reports the leg's own status context as failed (`wai
# cross-architecture byte-equality / <label>`), as for a divergence; the class is
# in the failed step's name and its error. A job of its own per label would put it
# in the context, but the `linux` label has two hosts, so that job could measure
# another host's disk than the leg's.
- name: "infrastructure: this runner's disk has room for the leg"
id: disk
if: matrix.os != 'windows'
shell: bash
run: |
set -euo pipefail
short=0; seen=""
for d in "${GITHUB_WORKSPACE:-$PWD}" "${RUNNER_TEMP:-/tmp}" "${TMPDIR:-/tmp}" "$HOME"; do
p=$d; while [ ! -e "$p" ]; do p=$(dirname "$p"); done
set -- $(df -Pk "$p" | awk 'NR == 2 {print $4, $6}')
echo "$d: $(($1 / 1048576)) GiB free on $2"
if [ "$1" -lt $((DISK_FLOOR_GIB * 1048576)) ]; then
echo "::error::infrastructure: ${{ matrix.label }}: $d is on $2, which has $(($1 / 1024)) MiB free, below the floor of $DISK_FLOOR_GIB GiB. The runner host's disk is full: no fixture was decoded, so this is not a byte-equality failure."
short=1; continue
fi
case " $seen " in *" $2 "*) continue ;; esac
seen="$seen $2"; f="$p/.wai-disk-probe.$$"
if ! err=$(dd if=/dev/zero of="$f" bs=1048576 count=32 conv=fsync 2>&1); then
case "$err" in
*"No space left"*|*"Input/output error"*|*"Read-only file system"*|*"quota"*)
echo "::error::infrastructure: ${{ matrix.label }}: a 32 MiB write under $p, on $2, failed: $(printf '%s' "$err" | head -n 1). The runner host's disk cannot take the leg's writes: no fixture was decoded, so this is not a byte-equality failure."
short=1 ;;
*) echo "::warning::a 32 MiB write under $p failed for another reason: $(printf '%s' "$err" | head -n 1)" ;;
esac
fi
rm -f "$f"
done
exit $short
# The same check on the Windows leg, in pwsh: its `bash` is the WSL stub, which
# would measure another system's disk. The directories are the workspace, the
# runner's temporary directory, TEMP, and the profile directory, where rustup keeps
# the toolchain. The free space is what each one's volume leaves the runner's user,
# quotas included, and the probe writes 32 MiB through to each volume once. The
# write errors named infrastructure are the Win32 codes for a full disk (39, 112),
# a write-protected one (19), an I/O device error (1117) and a quota (1295).
- name: "infrastructure: this runner's disk has room for the leg (Windows)"
id: disk_windows
if: matrix.os == 'windows'
shell: pwsh
run: |
Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass -Force
$floor = [int64]$env:DISK_FLOOR_GIB * 1GB
$short = 0; $seen = @()
$tmp = [System.IO.Path]::GetTempPath()
$dirs = @(
$(if ($env:GITHUB_WORKSPACE) { $env:GITHUB_WORKSPACE } else { $PWD.Path }),
$(if ($env:RUNNER_TEMP) { $env:RUNNER_TEMP } else { $tmp }),
$(if ($env:TEMP) { $env:TEMP } else { $tmp }),
$HOME
)
foreach ($d in $dirs) {
$p = [System.IO.Path]::GetFullPath($d)
while (-not (Test-Path -LiteralPath $p)) { $p = Split-Path -Parent $p }
$vol = [System.IO.Path]::GetPathRoot($p)
$free = [System.IO.DriveInfo]::new($vol).AvailableFreeSpace
Write-Output "${d}: $([math]::Floor($free / 1GB)) GiB free on $vol"
if ($free -lt $floor) {
Write-Output "::error::infrastructure: ${{ matrix.label }}: $d is on $vol, which has $([math]::Floor($free / 1MB)) MiB free, below the floor of $env:DISK_FLOOR_GIB GiB. The runner host's disk is full: no fixture was decoded, so this is not a byte-equality failure."
$short = 1; continue
}
if ($seen -contains $vol) { continue }
$seen += $vol; $f = Join-Path $p ".wai-disk-probe.$PID"
try {
$s = [System.IO.FileStream]::new($f, [System.IO.FileMode]::Create, [System.IO.FileAccess]::Write, [System.IO.FileShare]::None, 1048576, [System.IO.FileOptions]::WriteThrough)
try {
$buf = [byte[]]::new(1048576)
for ($i = 0; $i -lt 32; $i++) { $s.Write($buf, 0, $buf.Length) }
$s.Flush($true)
} finally { $s.Dispose() }
} catch {
$e = $_.Exception; while ($e.InnerException) { $e = $e.InnerException }
$msg = (($e.Message -split "`r?`n")[0]).TrimEnd('.', ' ')
if (($e.HResult -band 0xFFFF) -in 19, 39, 112, 1117, 1295) {
Write-Output "::error::infrastructure: ${{ matrix.label }}: a 32 MiB write under $p, on $vol, failed: $msg. The runner host's disk cannot take the leg's writes: no fixture was decoded, so this is not a byte-equality failure."
$short = 1
} else {
Write-Output "::warning::a 32 MiB write under $p failed for another reason: $msg"
}
} finally {
Remove-Item -LiteralPath $f -Force -ErrorAction SilentlyContinue
}
}
exit $short
- name: Checkout
uses: actions/checkout@v4
# unix: bootstrap rustup (docker image ships no Rust; macOS host has it but
# launchd's PATH omits ~/.cargo/bin). Windows is a host executor with rustup
# already present, and must use pwsh — its `bash` is the WSL stub.
- name: Install Rust 1.99.1 (unix)
if: matrix.os != 'windows'
shell: bash
run: |
export PATH="$HOME/.cargo/bin:$PATH"
command -v rustup >/dev/null 2>&1 || \
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y --profile minimal --default-toolchain 1.99.1
rustup toolchain install 1.99.1 --profile minimal
rustup default 1.99.1
echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
- name: Install Rust 1.99.1 (Windows)
if: matrix.os == 'windows'
shell: pwsh
run: |
rustup toolchain install 1.99.1 --profile minimal
rustup default 1.99.1
- name: Decode fixtures, assert golden bytes
working-directory: wai/byte-exact-conformance
run: cargo test --release -- --nocapture
# A disk can also fill while the leg runs, from the jobs that share it (three
# conformance jobs at once filled the arm64 VM's data disk mid-link). When a step
# after the check failed, the same measurement and write are made again: below the
# floor, or on a write refused for space, the failure is named infrastructure; with
# room, it stands as it is.
- name: "infrastructure: was the disk full when the leg failed?"
if: failure() && steps.disk.outcome == 'success'
shell: bash
run: |
set -euo pipefail
short=0; seen=""
for d in "${GITHUB_WORKSPACE:-$PWD}" "${RUNNER_TEMP:-/tmp}" "${TMPDIR:-/tmp}" "$HOME"; do
p=$d; while [ ! -e "$p" ]; do p=$(dirname "$p"); done
set -- $(df -Pk "$p" | awk 'NR == 2 {print $4, $6}')
echo "$d: $(($1 / 1048576)) GiB free on $2"
if [ "$1" -lt $((DISK_FLOOR_GIB * 1048576)) ]; then
echo "::error::infrastructure: ${{ matrix.label }}: $d is on $2, which has $(($1 / 1024)) MiB free, below the floor of $DISK_FLOOR_GIB GiB. The disk filled while the leg ran: read the failure above as the runner host's disk, not as a byte-equality failure."
short=1; continue
fi
case " $seen " in *" $2 "*) continue ;; esac
seen="$seen $2"; f="$p/.wai-disk-probe.$$"
if ! err=$(dd if=/dev/zero of="$f" bs=1048576 count=32 conv=fsync 2>&1); then
case "$err" in
*"No space left"*|*"Input/output error"*|*"Read-only file system"*|*"quota"*)
echo "::error::infrastructure: ${{ matrix.label }}: a 32 MiB write under $p, on $2, failed: $(printf '%s' "$err" | head -n 1). The disk filled while the leg ran: read the failure above as the runner host's disk, not as a byte-equality failure."
short=1 ;;
*) echo "::warning::a 32 MiB write under $p failed for another reason: $(printf '%s' "$err" | head -n 1)" ;;
esac
fi
rm -f "$f"
done
[ "$short" = 1 ] || echo "Every disk this leg writes to has room: the failure above is not the disk."
exit $short
# The same measurement and write on the Windows leg, in pwsh, as its first check.
- name: "infrastructure: was the disk full when the leg failed? (Windows)"
if: failure() && matrix.os == 'windows' && steps.disk_windows.outcome == 'success'
shell: pwsh
run: |
Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass -Force
$floor = [int64]$env:DISK_FLOOR_GIB * 1GB
$short = 0; $seen = @()
$tmp = [System.IO.Path]::GetTempPath()
$dirs = @(
$(if ($env:GITHUB_WORKSPACE) { $env:GITHUB_WORKSPACE } else { $PWD.Path }),
$(if ($env:RUNNER_TEMP) { $env:RUNNER_TEMP } else { $tmp }),
$(if ($env:TEMP) { $env:TEMP } else { $tmp }),
$HOME
)
foreach ($d in $dirs) {
$p = [System.IO.Path]::GetFullPath($d)
while (-not (Test-Path -LiteralPath $p)) { $p = Split-Path -Parent $p }
$vol = [System.IO.Path]::GetPathRoot($p)
$free = [System.IO.DriveInfo]::new($vol).AvailableFreeSpace
Write-Output "${d}: $([math]::Floor($free / 1GB)) GiB free on $vol"
if ($free -lt $floor) {
Write-Output "::error::infrastructure: ${{ matrix.label }}: $d is on $vol, which has $([math]::Floor($free / 1MB)) MiB free, below the floor of $env:DISK_FLOOR_GIB GiB. The disk filled while the leg ran: read the failure above as the runner host's disk, not as a byte-equality failure."
$short = 1; continue
}
if ($seen -contains $vol) { continue }
$seen += $vol; $f = Join-Path $p ".wai-disk-probe.$PID"
try {
$s = [System.IO.FileStream]::new($f, [System.IO.FileMode]::Create, [System.IO.FileAccess]::Write, [System.IO.FileShare]::None, 1048576, [System.IO.FileOptions]::WriteThrough)
try {
$buf = [byte[]]::new(1048576)
for ($i = 0; $i -lt 32; $i++) { $s.Write($buf, 0, $buf.Length) }
$s.Flush($true)
} finally { $s.Dispose() }
} catch {
$e = $_.Exception; while ($e.InnerException) { $e = $e.InnerException }
$msg = (($e.Message -split "`r?`n")[0]).TrimEnd('.', ' ')
if (($e.HResult -band 0xFFFF) -in 19, 39, 112, 1117, 1295) {
Write-Output "::error::infrastructure: ${{ matrix.label }}: a 32 MiB write under $p, on $vol, failed: $msg. The disk filled while the leg ran: read the failure above as the runner host's disk, not as a byte-equality failure."
$short = 1
} else {
Write-Output "::warning::a 32 MiB write under $p failed for another reason: $msg"
}
} finally {
Remove-Item -LiteralPath $f -Force -ErrorAction SilentlyContinue
}
}
if ($short -eq 0) { Write-Output "Every disk this leg writes to has room: the failure above is not the disk." }
exit $short