Repository navigation
test(architecture): landscape-record.test.sh flaked again with a broken pipe in a merge group #6376
Description
Context
plugins/architecture/skills/map-landscape/scripts/landscape-record.test.sh failed in a merge-group run on a change that does not touch it, and the queue removed the pull request.
- Run: 37240501487 (
gh-readonly-queue/main/pr-6344-fdd9489da0626e96d76832bd797b01c237ccad03), jobtest-bash (5). - Failing case:
drift: a moved checkout is not architecture drift,expected [0], got [3]. - Just before the case, the log shows:
landscape-record.sh: line 404: printf: write error: Broken pipejq: error: writing output failed: Broken pipe
- Effect: ci(check-plugins): fail a pull request on a plugin contract warning #6344, which changes only
scripts/validate-plugins.shand its suite, was dequeued at 2026-10-04T22:40:28Z (failed_checks) and had to be re-queued.
Closed #5785 ("landscape-record.test.sh flaked on main") reports the same symptom, so the cause is still present.
Proposed work
Find why a downstream reader closes the pipe early in this case, so that landscape-record.sh exits 3 instead of 0, and make the case deterministic.
Acceptance criteria
- The broken-pipe cause in the moved-checkout drift case is identified.
- The case passes repeatedly under the CI suite runner (
--jobs 3).
References
- Run 37240501487, job
test-bash (5) - ci(check-plugins): fail a pull request on a plugin contract warning #6344 (dequeued), test(architecture): landscape-record.test.sh flaked on main #5785 (same symptom, closed)
Activity
github-actions commentedon Oct 4, 2026
Thanks for the detailed report. As I understand it:
landscape-record.test.sh's "moved checkout is not architecture drift" case fails intermittently in merge-group runs with a broken pipe (printf/jqwrite errors) that makeslandscape-record.shexit 3 instead of 0, which dequeues unrelated PRs (here, #6344) from the merge queue. The report ties this to #5785, a closed issue on the same test file flaking on main, but that one's failure was a different case and symptom (an ordering/timing assertion mismatch, not a broken pipe), so this looks like a recurrence of general flakiness in the same test file rather than a duplicate of #5785's exact cause. The report already includes a specific CI run/job reference and clear acceptance criteria (identify the pipe-closing reader, make the case pass repeatedly under--jobs 3), so no additional information is needed from my side.Automated triage. A maintainer reviews it.