fix(test): the H4 fixture teardown waits for exit + retries the rmSync (#830)

[aarontrowbridge]

Closes #830.

The session-API availability fixture's teardown was fire-and-forget: SIGTERM + an unref'd SIGKILL timer, then an IMMEDIATE recursive rmdir of the temp HOME — while the vendored binary's background package-install was still writing into it. ENOTEMPTY on CI, reproduced twice on the same run (zod/src/v3/tests, then effect/dist/unstable — the nondeterminism of a write-race), never on a warm director-local run. It blocked every open PR's fast check (seen blocking #828).

Fix, both halves of the race: (a) killAndWait() — SEQUENTIAL shutdown awaited to process exit (SIGTERM → 10s bounded wait → SIGKILL → 5s bounded wait; the exit/error listeners are cleaned on timeout); (b) a RETRYING rmSync (maxRetries 20 × retryDelay 250ms — Node natively retries ENOTEMPTY/EBUSY) bounding any residual grandchild writer.

Honesty note on the RED: a cross-process write-race has no deterministic RED — the reviewer's own note on the #814 review said exactly this of the lock-release race. The verification is the fixture passing locally (director-run) and the CI fast job green across reruns; the race window that CI hit twice is now closed at the root (no write can be in flight after a child-exit await) and bounded at the surface (retried removal).

Verification: mode_block suite (32 tests) green locally, full extension suite green, smoke PASS — director-run, scrubbed env.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Team

Run ID: 4a7301ab-184d-47ad-a142-ca387108b5aa

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}