test(worker-threads): reap the spawned child after a timeout

[szegedi]

Symptom

win32-test-22 on #385 printed its mocha epilogue inside the first minute, then sat in_progress on Run ./.prebuildify/test for half an hour:

✔ should not crash when worker is terminated (19288ms)
91 passing (1m)
1 failing
1) Worker Threads should work:
   Error: Timeout of 20000ms exceeded.

Tests done in a minute, job still running 30 minutes later.

Cause

A mocha timeout rejects the test but does not kill the process the test spawned, and npm test is nyc mocha … with no --exit, so mocha waits for the event loop to drain before exiting. A live child keeps its process handle on that loop, so a child that outlives its test holds the entire run open.

When the child is merely slow this is invisible — mocha waits the extra couple of seconds and exits, which is why it never shows up on Linux (worker.js finishes in ~4.5s there). When the child is wedged, the run never ends and the job burns a runner until the CI job limit instead of failing in ~20s.

Fix

promisify(execFile) exposes the ChildProcess on the returned promise, so capture it and kill it from afterEach — mocha still runs hooks after a timeout. Measured locally with the child replaced by a 10-minute sleep and the test timeout forced low:

	total
with the reap	1.19s
without the reap	never exits (killed by a 30s watchdog)

Chose this over adding --exit to the mocha invocation: --exit would paper over any handle leak, and this suite exists partly to catch worker threads that fail to exit.

Scope

This bounds the damage only. It does not address why that child wedges on win32, which is the actual bug and a separate investigation. It is worth noting that Windows takes the plain V8 CpuProfiler path (no SIGPROF, and worker.js sets withContexts/useCPED to false off-platform), and this codebase already carries detectV8Bug / v8ProfilerStuckEventLoopDetected machinery for V8 sampler stalls.

Also unaddressed here: should work has a 20s budget while its heavier in-process sibling gets 30s, and Windows runs these ~10x slower than Linux. That inconsistency is worth revisiting, but raising the timeout would not have fixed this run — the parent waits on the child regardless of the test timeout.

[github-actions]

Overall package size

Self size: 2.45 MB
Deduped: 3.15 MB
No deduping: 3.15 MB

Dependency sizes

| name | version | self size | total size | |------|---------|-----------|------------| | pprof-format | 2.3.0 | 503.97 kB | 503.97 kB | | source-map | 0.8.0 | 185.66 kB | 185.66 kB | | node-gyp-build | 4.8.4 | 13.86 kB | 13.86 kB |

_{🤖 This report was automatically generated by heaviest-objects-in-the-universe}