Only consider pending fetch-later requests for deferred fetch quota

[TimvdLippe]

Quota computation shouldn't include aborted or already
completed requests. WPT uses AbortController to ensure
quota is freed up. This also makes sense, since they are
no longer relevant. Therefore, only consider pending
deferred fetches when computing quota.

• At least two implementers are interested (and none opposed):
  • Chromium
  • …
• Tests are written and can be reviewed and commented upon at:
  • https://wpt.fyi/results/fetch/fetch-later/quota/accumulated-oversized-payload.https.window.html?label=experimental&label=master&aligned
• Implementation bugs are filed:
  • Chromium: n/a already passing tests
  • Gecko: n/a no implementation yet
  • WebKit: n/a no implementation yet
• The top of this comment includes a clear commit message to use.

(See WHATWG Working Mode: Changes for more details.)

---

Preview | Diff

[TimvdLippe]

FYI @noamr PTAL

[noamr]

(2) is fine
Not sure about 1... Isn't posting an empty body a legit use case? The null use case was about not allowing streaming.

[TimvdLippe]

Well, Chrome currently blocks such requests: https://wpt.fyi/results/fetch/fetch-later/quota/empty-payload.https.window.html?label=experimental&label=master&aligned If I didn't implement that case, Servo wouldn't match Chrome's behaviour.

[noamr]

Well, Chrome currently blocks such requests: https://wpt.fyi/results/fetch/fetch-later/quota/empty-payload.https.window.html?label=experimental&label=master&aligned If I didn't implement that case, Servo wouldn't match Chrome's behaviour.

@mingyc the spec only prevents null bodies (streaming) rather than empty... Do you remember why chromium also blocks empty requests?

@TimvdLippe we should check whether this is an implementation oversight or a spec oversight before defaulting to "spec what chromium does"... For the time being I think that the other part of the PR is ok.

[TimvdLippe]

I wasn't aware of the streaming part. Since the test wasn't marked as tentative, I thought this part of the spec was meant to imply the empty case behaviour. I don't mind either behaviour (not familiar with streaming or anything), so happy to update it to whatever you folks decide.

[TimvdLippe]

Friendly ping @mingyc on Chromiums perspective

[noamr]

Friendly ping @mingyc on Chromiums perspective

I suggest to make the other change and propose a WPT pr to align with the spec re empty request bodies

[TimvdLippe]

Updated this PR and updated Servo + tests, see the WPT export in web-platform-tests/wpt#58726 Please review that PR.

[mingyc]

Sorry about missing this thread

@mingyc the spec only prevents null bodies (streaming) rather than empty... Do you remember why chromium also blocks empty requests?

I roughly remember there was some discussion about posting with empty body should be replaced by a GET request (can't find it now). Isn't null body the same as empty body in this context?

[annevk]

No they are distinct. One has a Content-Length.

[TimvdLippe]

@annevk @noamr Can I write Chromium and Webkit as interested implementers for this PR?

[noamr]

@annevk @noamr Can I write Chromium and Webkit as interested implementers for this PR?

yes, though do you mind opening a bug for the 0-length thing?

[TimvdLippe]

Sure, done: #1920