Repository navigation
fix(ci): run the formal checks on Windows and size test bounds for a loaded host - #126
Merged
Merged
Conversation
Upd4ting
marked this pull request as ready for review
November 20, 2026 22:28
Upd4ting
enabled auto-merge
November 20, 2026 22:28
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Wec9b4589-19. The nightly CI has failed every night since 2026-11-17, in two ways. This PR fixes both.
Windows legs (
Node 22 on Windows,Node 24 on Windows).scripts/test.shrunsformal-pipeline-chains/check.mjs. That script spawnedpnpmwith no shell, and Node on Windows only finds.exeand.comfiles, so it failed withspawn pnpm ENOENT.formal-json-value-identityandformal-merge-repeated-keysdid the same withpnpmand with the extensionlessnode_modules/.bin/ajs. All three now run thetypescript/bin/tscand@antelopejs/coreCLI scripts withprocess.execPath, as the other three formal checks already did. The build step is nowtscitself, withoutrimraf dist, because each check runs named test files only. Noshell: trueis used, so Node 24 prints no DEP0190 warning.Stress (3 of 3 runs failing). Every failure came from a wall-clock bound sized for an idle host, or from one wait that was too short:
TIMEOUT_MS(probe-steps,probe-locks,probe.spec,probe-url.spec,probe-options)TIMEOUT_MS1 s → 5 sprobe-manyTIMEOUT_MS(deadline 15 s)probe-stepslatency testDELAY_MS100 → 500 msGROUP_BUDGETSremovedpipeline-chainsfeed propertyremovedeventDELIVERY_MSfor that removalEach test still catches the bug it names. A quadratic regex over the 100k-character URL value takes 17 s here, above the 5 s bound. The quadratic respell the dedup budget guards took over 1.1 s per row on an idle host, above 0.4 s per row.
probe-manystill counts the prepared statements, which is what catches a probe that confirms every relation.flowchart LR T["scripts/test.sh chains"] --> C["check.mjs"] C -->|"before: spawn('pnpm', …)"| E["ENOENT on Windows"] C -->|"after: spawn(process.execPath, [tsc or ajs CLI, …])"| N["the same Node, on every OS"]Related issue
Wec9b4589-19, from ANT-3515. Nightly runs failing since 2026-11-17: 44002817791, 44223078548, 44444088606, 44660523649.
Verification
A manual run cannot reach the nightly tier, because the
ci-tieraction returnsnightlyonly forschedule. The next scheduled nightly is the final proof. Until then, these are the local reproductions.Windows ENOENT, reproduced with
pnpmremoved from PATH.Before (
check.mjsfromorigin/main):After (the build step runs too):
Stress, reproduced like the CI job. The script starts 4 busy loops per CPU (64 here), then runs the probe test files against two MySQL 8.4 servers in parallel.
Before:
After:
Also run beside the same load after the change:
distinct-large.test.js: passes twice on MySQL 8.4 and three times on MariaDB 11.8.pipeline-chainscheck: exit 0 on MySQL 8.4 and on MariaDB 11.8.Run without load:
pnpm run test:unit: 647 passing.pnpm run typecheck, plusoxlintandoxfmton the changed files: clean.Review
An independent review of the diff (Sonnet 5.5) found no defects in the spawn changes. It raised three points about how much the larger bounds weaken the tests:
probe-locks. With a 5 s timeout, the server's own lock wait (6 s) ends before the 10 s bound, so the time bound alone would not catch a probe with no read timer. The outcome assertion does catch it. With the read timer removed, the test fails withAssertionError: expected 'signed-in' to equal 'unreachable'(measured). No change.probe-url.spec. A slow-value bug that costs 1 to 5 s against a closed port would now pass. The bug that test names, a quadratic over the 100k-character value, takes 17 s. Under load the closed-port probes took up to 2.6 s, so a tighter bound would flake. Kept as is.Opened. Late events from two or more open attempts could still leak. Every failure in the four nightly runs was the single late removal fixed here. A count-based handshake must also handle a removal whose insert came before the feed opened, so it waits for evidence of that case.Follow-ups
Checklist