test_runner: automatically wait for subtests to finish

[cjihrig]

Some people feel pretty strongly that the test runner should automatically wait for subtests to finish (by the way, that comment seems like something that someone probably should have moderated). The current behavior is for tests to run and cancel any subtests that have not finished in that time. The idea is that people would use await as needed in this case.

This is what the code would look like to make that automatic waiting happen. I would still need to update the docs if we want to land this. From a user perspective the changes are:

• Nothing ever needs to be awaited anymore. Existing code that does await subtests continues to work. This is a DX improvement.
• Subtests that timeout just hang instead of being cancelled if you aren't applying a test timeout. This is a worse DX IMO.

I'd like to know how other people feel about such a change.

cc: @nodejs/test_runner @mcollina

[nodejs-github-bot]

Review requested:

• @nodejs/test_runner

[anonrig]

How do you feel about this change?

[cjihrig]

Mostly indifferent. I don't like the bit of extra complexity or shipping breaking changes in general. It's also more magic automatically awaiting async operations (although suites already do this, so it's a bit more consistent), and artificially keeping the event loop alive. I can appreciate making things a little simpler though.

[mcollina]

I think I'm ok with this change, and I think adding the extra complexity matches user expectations.

I'll be hide that comment.

[MoLow]

I am in favor of this change (not because of that comment 😃)

[mcollina]

maybe this should be removed

[khaosdoctor]

I'm also in favor of this, mostly because today the behaviour of the subtests (and tests), at least to me, seem a bit different from what I'd expect. As an user I'd like to just write the tests without having to care if they're promises or not because the runner should (in theory) be able to handle all pending promises and automatically close them.

In the meantime, I suggest we add a better error message for those cases, so instead of "Test did not finish before its parent" we could include something like "Did you forget to await a subtest?" or something like this.

[nicholaswmin]

I don't like the bit of extra complexity or shipping breaking changes in general. It's also more magic automatically awaiting async operations

FTR I agree with this too as strange as it sounds."¯_(ツ)_/¯

[nicholaswmin]

Hi guys - Ive made that comment - I don't like this change. I think I was wrong. A misalignment of this nature is probably a bad thing. Sorry for the confusion.

I still do think that the test runner is going to be enormously problematic for people with less experience though. I wouldn't dare ask someone without substantial NodeJS experience to write a serious test suite with it.
It's not the test runner itself that's the problem but rather the surprising misalignment between the semantics
of await-able stuff and the fact the Eloop doesn't actually wait for them.

I stand behind my comment that test runners need to be particularly ergonomic because of their usage context.

But I'm also starting to think that having it align it with the peculiarities of Node is preferable to magic. If there's gonna be a change - i think it should be upstream, to how Promises and the EL interact.

Also that's the way i talk - I don't mean no disrespect to anyone; I can't even if I wanted - I spend more time with the test runner than with actual people so thank you for you work. I'll try to be more considerate in the future with my phrasing.

Sorry for the confusion, only my 2c.