Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.63%. Comparing base (80e0f14) to head (3bcacb1).
⚠️ Report is 1 commits behind head on main.

@@            Coverage Diff             @@
##             main   #63057      +/-   ##
==========================================
- Coverage   89.65%   89.63%   -0.02%     
==========================================
  Files         708      708              
  Lines      220413   220413              
  Branches    42275    42280       +5     
==========================================
- Hits       197607   197571      -36     
- Misses      14658    14704      +46     
+ Partials     8148     8138      -10     

@IlyasShabi
Thanks for working on this. I think this can be fixed in the fixture instead of marking the test as flaky.

different-registry-per-thread.mjs registers object literals with process.finalization.register(), but those objects are not strongly referenced after registration. process.finalization only keeps a WeakRef, and the documented behavior is that the callback is not called on process exit if the ref was GCed before exit:

That matches the failure here: the child process exits with status 0, but stdout only contains shutdown on worker.

Keeping the registered refs alive until exit should make this deterministic, e.g. by storing the main-thread and worker refs in a module-level array before registering them.

The V8 update may still be related, but it seems more likely that it changed GC timing enough to expose this fragile test assumption.

One concrete shape for the fixture fix could be to keep the registered refs alive in a top-level array, for both the main thread and the worker.

const registeredRefs = [];

const ref = { foo: 'foo' };
registeredRefs.push(ref);
process.finalization.register(ref, () => {
  process.stdout.write('shutdown on main thread\n');
});