Tom Boucher 7013406f0b fix(#1586): pin withPlanningLock liveness probe in perf-407 for determinism ...

#1531/#1532 replaced withPlanningLock's mtime-staleness with PID-liveness
(process.kill(pid,0)). perf-407 plants pid: process.pid + 1 and relied on the
old mtime model to force the retry/sleep path; under the new model that pid's
liveness is environment-dependent, so the retry path was taken on some runners
and skipped on others (sleepCallCount: 0 precondition failure) — flaky CI red
on next that blocks the merge queue.

Pin the planted holder live via the _setLockProbes seam that #1532 added, and
_resetLockProbes() in afterEach. The retry/sleep path is now exercised
deterministically on every runner. No assertion weakened; no variable renamed.
perf-316 is unaffected (its worker writes the parent's own, always-live pid).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

2026-06-22 12:33:13 -04:00

10 KiB

Raw Blame History

View Raw