feat(ext/web): Add AbortSignal.timeout().

[andreubotella]

Closes #13582.

[crowlKats]

Should be fairly simple with Deno.unrefTimer & Deno.refTimer, though they are unstable

[andreubotella]

I wasn't sure about overriding addEventListener and removeEventListener though, with such short timeframe before the release of 1.19.

[lucacasonato]

I don't think this is going to make it for 1.19. There are 2 hours left before denoland/deno freezes, and I'd really like to have the ref handling work correctly before we merge this. Also we need to figure out how this integrates with test sanitizer.

I think it would make more sense if we don't actually start the timer until the first listener is registered. We should also cancel the timer once the last listener unregisters.

Let's get this landed for 1.20 :-)

[andreubotella]

I think it would make more sense if we don't actually start the timer until the first listener is registered. We should also cancel the timer once the last listener unregisters.

Come to think about it, this would be hard to do right, since per the spec the signal is only aborted in a timer task:

const signal = AbortSignal.timeout(10);
console.log(signal.aborted);  // false

const start = Date.now();
while (Date.now() - start < 20) {}

console.log(signal.aborted);  // must still be false!

[lucacasonato]

Could be solved with a macrotask callback I think.

[andreubotella]

Could be solved with a macrotask callback I think.

Okay, so let's say that someone checks if the signal has aborted, and the timeout has indeed passed. Do we queue a macrotask now? Because if it has expired a while ago and the event loop hasn't been blocked during that time, the signal should have aborted by the time we're checking. (The spec technically would allow for this, but it's far from ideal.)

[andreubotella]

I'm not sure that an implementation that only has an active timer while there are listeners is workable, so I've implemented this with Deno.unrefTimer() and added tests. I guess that idea could still be pursued after this implementation is merged.

[bartlomieju]

@crowlKats @lucacasonato would be great to get this in for 1.20. Please review

[crowlKats]

LGTM, thanks!