test: fix flaky wpt/test-timers

[Trott]

Change the wpt tests to not run in parallel in our Python test runner.
On my local machine, this changes the time needed to run the suite from
about 2 seconds to about 5 seconds, but it makes the test suite much
more reliable.

Fixes: #37672

[benjamingr]

I don't understand this well enough to review but just wanted to say thanks for all the work on these tasks 🙇

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/36591/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/36605/

[targos]

Feel free to skip the tests again in the mean time

[Trott]

The test that fails (negative-settimeout.any.js) is short and straightforward. It checks that a timer with a negative timeout value runs before a timer with a timeout value of 10.

setup({single_test: true});setTimeout(done,-100);setTimeout(assert_unreached,10);

While the ordering may be guaranteed in browsers, I don't think it is in Node.js due to the linked list stuff that underpins timers. But maybe I'm confused. /ping @nodejs/timers for confirmation.

If the ordering in the above case is not in fact guaranteed in Node.js, then skipping the test is certainly the right thing to do.

[Linkgoron]

The test that fails (negative-settimeout.any.js) is short and straightforward. It checks that a timer with a negative timeout value runs before a timer with a timeout value of 10.

setup({single_test: true});setTimeout(done,-100);setTimeout(assert_unreached,10);

While the ordering may be guaranteed in browsers, I don't think it is in Node.js due to the linked list stuff that underpins timers. But maybe I'm confused. /ping @nodejs/timers for confirmation.

If the ordering in the above case is not in fact guaranteed in Node.js, then skipping the test is certainly the right thing to do.

@Trott I think that there's something wrong with done(). Rewriting the test to the following is not flaky on my machine (while the original is):

setup({single_test: true});setTimeout(()=>{clearTimeout(y);done();},-100);vary=setTimeout(assert_unreached,10);

I assume that done doesn't clear the timeout, and there's an edge-case where if there's a very small difference (or maybe if they're on the same tick) between the actual running time of done and assert_unreached it fails the test anyway. I think that the above fix is a slightly more correct way to test that the first runs before the second anyway.

[Trott]

OK. The Parallel to Simple change fixed most of them. The one it didn't fix is now marked flaky in the status file. This is ready for review.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/36683/ ✅

[Trott]

Stress test against this PR: https://ci.nodejs.org/view/Stress/job/node-stress-single-test/224/ ✅

[Trott]

Stress test and CI were both green on the first run. It would be great to get this frequent-CI-failure addressed. @nodejs/testing

[targos]

Thanks!

[Trott]

Landed in 66f0eb7"