Editorial: Refactor for auto-reporting timing from fetch

[noamr]

Depends on whatwg/fetch#1413

This clarifies that a resource timing entry is always for a
fetch and not for a request or response.

---

Preview | Diff

[annevk]

This needs updating to align with the revised approach.

[noamr]

This needs updating to align with the revised approach.

Fixed

[annevk]

So the result here is that for synchronous XMLHttpRequest we don't create a timing entry, right?

[noamr]

So the result here is that for synchronous XMLHttpRequest we don't create a timing entry, right?

No, there is no behavioral change. In sync XHR the task that reports the timing would run on the parallel queue, but that's immaterial because the main-thread queue is waiting on it anyway.

[annevk]

I think there's two problems and they might also have been present with the previous text (though in a different way):

1. When you're "in parallel" you don't have access to the main thread.
2. "Pause" means nothing else can happen in the main thread (other than what it allows for). Having state changes happen during pause would be weird.

Perhaps sync XHR needs to invoke "report timing steps" itself?

[noamr]

I think there's two problems and they might also have been present with the previous text (though in a different way):

1. When you're "in parallel" you don't have access to the main thread.
2. "Pause" means nothing else can happen in the main thread (other than what it allows for). Having state changes happen during pause would be weird.

Perhaps sync XHR needs to invoke "report timing steps" itself?

Sure, that would be an easy solution. I'll revise the fetch spec to only report if not in a parallel queue

[noamr]

I think there's two problems and they might also have been present with the previous text (though in a different way):

1. When you're "in parallel" you don't have access to the main thread.
2. "Pause" means nothing else can happen in the main thread (other than what it allows for). Having state changes happen during pause would be weird.

Perhaps sync XHR needs to invoke "report timing steps" itself?

Fixed in whatwg/fetch@9f0745a and Report timing for sync XHR

[noamr]

@annevk: I believe this is now ready for a re-review as it links to the corrected fetch.

[domenic]

This is blocking #353. I will give my own review and ping @annevk about if that's enough.

[domenic]

Looks good modulo one phrasing suggestion.

[domenic]

This would be a bit more conventional as "Report timing for this's fetch controller given the current global object".

[annevk]

I'll push a fixup commit.