test: one readiness wait in kill-settle; drop a fail-open test that pinned nothing

Merged
TH
3d ago
+17-41

Summary

Applies the take-or-leave review notes on #4058 and #4059. Test-only.

  • packages/host-kit/src/internal/exec-kill-settle.test.ts: the cancelled-command kill-policy test was the last one that polled a ready marker file. It now prints ready on stdout and waits through stdoutAnnouncement, like every other readiness wait in the file. That removes waitForTrapReady and the ready marker path.
  • scripts/check-affected/run.test.ts: drops "a fail-open plan fails on a unit-suite failure". It passed before #4058 too, so it pinned nothing. The fail-open plan test, which asserts the related command's --project=! exclusions and that check:unit runs, is the regression pin.

Validation

  • Planted reds in the kill-settle test:
    • Delaying the trap by 1 s still passes, because the stdout wait holds the abort.
    • Removing the wait with the same delay fails with the original released ENOENT.
  • The kill-settle file passed 17/17 five times at load average ~50.
  • pnpm check:affected:test passed 90/90.
  • pnpm check:affected --run (fail-open, 66 checks, exactCommit=true):
    • unit, vitest-related (with the unit-project exclusions), format, lint and typecheck passed.
    • The run stopped at provider-integration. ios-record-trace.test.ts, "host contention and recovers", timed out at 5 s at load average 86, and also fails alone at that load.
    • That file is untouched here, and the same test timed out in the local runs on #4058. As a result, 25 later checks did not run locally.

šŸ¤– Generated with Claude Code