How can microtasks run without a task?

[annevk]

I was looking at the final patch of #7768 and it struck me that it implies a microtask checkpoint can be performed when there are no tasks on the queue. However, that assumes a microtask can be enqueued without a corresponding task, right? I don't think that's something we mean to support.

Should what is now step 3 of https://html.spec.whatwg.org/#event-loop-processing-model be nested under step 2?

cc @smaug----

[smaug----]

Yeah, microtask's generic model is end-of-outermost script execution or end of task. So under 2 sounds reasonable.

[smaug----]

(Separately, I'm not quite sure step 7 follows what the browsers do. At least in Gecko 'Update the rendering' happens effectively inside a task)

[domenic]

However, that assumes a microtask can be enqueued without a corresponding task, right? I don't think that's something we mean to support.

I'm not sure. Basically saying that we don't support this means we don't support implementations enqueuing microtasks, unless they do so from inside a task.

I think there are at least some instances of this, e.g. all the calls to "await a stable state".

[annevk]

Hmm yeah, "await a stable state" looks like a bug, because that could interfere with the microtask queue of a currently running task, which seems bad? (This is a problem we have been trying to remove from specifications over the years, as many of them enqueued microtasks without a corresponding task.)

I guess there's a larger question as to whether the microtask queue could be owned by a task. That might avoid some of this.

[annevk]

This depends on #2882 to some extent I suppose. We could use an existing task source for "await a stable state", but that would not be particularly accurate either I think.