1865239 - Try to provide a way for OnDataFinish to be done strictly before OnStopRequest.

Status: RESOLVED FIXED

(Core :: Networking, task, P3)

Description

[Emilio Cobos Álvarez (:emilio)]

Being able to rely on that would simplify the implementation of bug 1864817 quite a bit.

It seems that, if I'm reading this code correctly, there is no race if either:

• The ODA/ODF event target is the same as the OnStopRequest target (because in that case ODA is guaranteed to be enqueued first).
• The ODA/ODF event target is the socket thread (which is true of the CSS loader, and a bunch of others, IIUC, that's what code like this tries to do). Because in that case the task is executed sync and ODA/ODF is done before OnStopRequest is enqueued.

If that's correct, it'd be good to have that more explicitly documented at least, so that it's not so sketchy to rely on that. Maybe even worth having a more explicit API to make that even clearer.

At least for the CSS loader, it'd be nice to be able to rely on that, since that means we could make ODF do something like taking care of the happy path in ODF, and set a flag to nerf OnStopRequest

Comment 1

[Sunil Mayya]

Attachment: Bug 1865239 - update ChannelEventQueue and nsIThreadRetargetableStreamListener documentation. r=#necko,emilio

Comment 2

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/eb4ff33091f5
update ChannelEventQueue and nsIThreadRetargetableStreamListener documentation. r=necko-reviewers,jesup DONTBUILD

Comment 3

[Cristian Tuns]

https://hg.mozilla.org/mozilla-central/rev/eb4ff33091f5