src,http2: simplify native immediate queue running and harden http2 test

[addaleax]

test: make test-http2-buffersize more correct

Previously, this code could have closed the server before the
connection was actually received by the server, as the 'close'
event on the client side can be emitted before the connection
is established.

The following commit exacerbates this problem, so fix the test first.

src: simplify native immediate queue running

Make SetImmediate() behave more like process.nextTick()
(which matches how we use it) by also running tasks that have been
added during previous SetImmediate() calls.

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• commit message follows commit guidelines

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/28628/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/28687/ ✅

[addaleax]

Landed in f284290...78743f8

[codebytere]

@addaleax if this is something that should be in v12.x, could you please backport it? There are a nest of conflicts and i want to be extra careful 😅

[addaleax]

This is a follow-up to #31386, so I’ll try to backport those together

[apapirovski]

I missed this originally @addaleax but this seems like a semver-major to me, so probably shouldn't be backported? See https://nodejs.org/api/timers.html#timers_setimmediate_callback_args where we document the setImmediate behavior as:

If an immediate timer is queued from inside an executing callback, that timer will not be triggered until the next event loop iteration.

[addaleax]

@apapirovski This is about the C++ SetImmediate(), it has no effect on the JS API

[apapirovski]

@addaleax yup, I get that but the queues are somewhat tied to each other, right? Or well, at least I would expect them to be given the naming... (They certainly are tied in terms of execution timing, at the very least.)

[addaleax]

@apapirovski The queues are managed independently, no connection there.

The naming is the same because they use the same underlying mechanism, but effectively we use SetImmediate() as nextTick() in the C++ source code. I’d be okay with renaming it to something else, just not NextTick() – while that would be 100 % accurate, it’s not true for the JS nextTick() and that might be even more confusing ;)

[apapirovski]

@addaleax Yeah, that makes sense to me. I guess that was the original crux of the issue — the naming ties into the JS setImmediate but the behavior only partially does. That said, I also don't have any great ideas re: naming. Maybe it's just a matter of adding a comment to the native SetImmediate function that calls this difference out?