Skip to content

ci(windows-nightly): run only the PR-skipped tests; move on when a kill never confirms (#167) - #174

Merged
Wolfe-Jam merged 3 commits into
mainfrom
fix/167-runner-moves-on
Oct 17, 2026
Merged

Wolfe-Jam merged 3 commits into
mainfrom
fix/167-runner-moves-on

Conversation

@Wolfe-Jam

@Wolfe-Jam Wolfe-Jam commented Oct 16, 2026 •

Copy link
Copy Markdown
Owner

Refs #167. Follow-up to #172.

What the 10-16 nightlies showed

  • Scheduled run 38906816290: both legs ran out the 60-minute step. parity-brake (Node 22) and scorer (Node 24) printed their passing summaries, then bun never exited. No HUNG line: the runner waited for an exit that never came.
  • Branch run 38949263198 (first commit here): Node 24 156/156 green in 3½ min. Node 22: wasm/kernel-stress ran 19/19 passing, then didn't exit; named at 5 min; taskkill said "could not be terminated … no running instance of the task"; the runner moved on 15 s later and finished in 8½ min.

So #167 has a second shape: bun finishing a file's tests and not exiting. All three files load WASM (scoring kernel / FAFb), a lead, not proven.

Changes

  1. scripts/test-per-file.mjs: logs HUNG (#167) the moment the limit hits; logs what taskkill /T /F reported; moves on 15 s after the kill whether or not the exit arrives; the junit entry says ran its tests but did not exit or did not finish.
  2. The nightly runs only the tests the Windows PR job skips (--nightly-only: the files that use windowsPr: e2e, commands/legacy-stamp-note, wjttc/hooks). Every PR's windows job already runs the other 153 files on Windows (a failing test fails the PR; a stall warns), so the nightly no longer repeats them. The release gate (pubpro Step 1.1b) becomes: those tests green on Node 22. Step limit 60 → 30 min, job 75 → 40.

Checked

  • macOS: a test past the limit is killed, named, next file runs; with the kill stubbed out, "no exit within 15s of the kill; moving on without it"; an ordinary file passes.
  • node scripts/test-per-file.mjs --nightly-only locally: 3 files, 0 failed.
  • Windows: branch run 38953954679 (this commit).

🤖 Generated with Claude Code

…onfirms (#167)

The 10-16 nightly: parity-brake (Node 22) and scorer (Node 24) printed
their summaries, then bun never exited. The 5-minute limit fired, but the
runner logged and moved on only after the process confirmed its exit, which
never came, so both jobs sat until the 60-minute step limit.

Now the runner logs HUNG the moment the limit hits, logs what taskkill
reported, and moves on 15 s after the kill whether or not the exit arrives.
The report says whether the file ran its tests and didn't exit, or didn't
finish them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@vercel

vercel Bot commented Oct 16, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment
Project Deployment Actions Updated
faf-cli Ignored Ignored Preview Oct 16, 2026 7:37pm UTC

Every PR's windows job already runs the suite on Windows, except the 3 tests
marked nightly-only (e2e, legacy-stamp-note, wjttc/hooks). The nightly re-ran
all 156 files to reach them, and each extra file was another chance for bun
to finish its tests and not exit: parity-brake, scorer and kernel-stress did
on 10-16, all passing. Now the nightly runs only the files that use
windowsPr (`--nightly-only`), so the release gate is those tests. Step limit
30 min, job 40.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@Wolfe-Jam Wolfe-Jam changed the title ci(windows-nightly): the per-file runner moves on when a kill never confirms (#167) ci(windows-nightly): run only the PR-skipped tests; move on when a kill never confirms (#167) Oct 16, 2026
R7-4 starts the CLI 16 times; on this PR's Windows (Node 24) job a child
hung there twice in a row (faf export --html, then faf loop), named by
namedHang. That is the class the PR job leaves to the nightly, so it
joins e2e, legacy-stamp-note and hooks there.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@Wolfe-Jam
Wolfe-Jam merged commit bf7e941 into main Oct 17, 2026
17 of 18 checks passed
@Wolfe-Jam
Wolfe-Jam deleted the fix/167-runner-moves-on branch October 17, 2026 06:03
@Wolfe-Jam Wolfe-Jam mentioned this pull request Oct 17, 2026
Wolfe-Jam added a commit that referenced this pull request Oct 17, 2026
`faf check` warns when a key_files path no longer exists (#173). Windows CI names a stuck test and moves on (#172, #174).

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Status: Done

Development

Successfully merging this pull request may close these issues.

1 participant