test-finalization-group-error failed on v8 8.4

[gengjiawen]

This one still failed:
cc @mmarchini

=== release test-finalization-group-error ===
Path: parallel/test-finalization-group-error
--- stdout ---
Mismatched noop function calls. Expected exactly 1, actual 2.
    at Proxy.mustCall (/home/runner/work/node-v8/node-v8/test/common/index.js:328:10)
    at Object.<anonymous> (/home/runner/work/node-v8/node-v8/test/parallel/test-finalization-group-error.js:27:40)
    at Module._compile (internal/modules/cjs/loader.js:1178:30)
    at Object.Module._extensions..js (internal/modules/cjs/loader.js:1198:10)
    at Module.load (internal/modules/cjs/loader.js:1027:32)
    at Function.Module._load (internal/modules/cjs/loader.js:923:14)
    at Function.executeUserEntryPoint [as runMain] (internal/modules/run_main.js:71:12)
    at internal/main/run_main_module.js:17:47
Command: out/Release/node --expose-gc --harmony-weak-refs /home/runner/work/node-v8/node-v8/test/parallel/test-finalization-group-error.js

Originally posted by @gengjiawen in #155 (comment)

[gengjiawen]

I debugged this, throw is triggered, but uncaughtException not noticed.

node-v8/test/parallel/test-finalization-group-error.js

Lines 8 to 10 in e1230d5

	const g = new globalThis.FinalizationRegistry(common.mustCallAtLeast(() => {
	throw new Error('test');
	}, 1));

cc @devsnek @syg

[syg]

Pretty sure this is because you need to change the flag to --harmony-weak-refs-with-cleanup-some. At the last TC39, Apple objected to cleanupSome, so the method was made optional. V8 split out cleanupSome to a different flag so the rest of the API may ship. That test uses cleanupSome, which is no longer exposed by default.

[gengjiawen]

I am aware of that, when I debug, I add --harmony-weak-refs-with-cleanup-some, but the test still failed.

[devsnek]

@syg could recent changes (maybe the torque rewrite?) have broken the uncaught exception reporting? there's been a similar issue with microtasks for a while.

[syg]

I debugged this locally. I noticed three issues with node's WeakRef integration. I'm not exactly sure how this test worked before.

1. The top-level line g.register({}, 42) doesn't actually cause the unreachable {} to be GC-able when calling gc(), so the async task wasn't being scheduled and wasn't running at all. I didn't dig into why exactly, but other V8 tests tend to do the registration in an IIFE, like (function () { g.register({}, 42); })();, which is a more foolproof for causing the object to be collected.

2. There's a single object registered, and FinalizationRegistry's API has changed to a per-item API. cleanupSome will drain all finalized objects, so if cleanupSome is called, no async task will be scheduled even if the scheduling was working.

3. As per the the deprecation message for Isolate::SetHostCleanupFinalizationGroupCallback, the scheduling of FinalizationRegistry cleanup tasks is automatic if no host cleanup callback is set. I see that this version of node is still using the hook. Things should "just work" if no hook is set.

Testing weak refs and finalizers is hard, good luck!

[targos]

This is made obsolete by nodejs/node#33373 which I cherry-picked to canary-base.