1900681 - XMLHttpRequestMainThread::DisconnectFromOwner should not call Abort for `mForWorker` XHRs because it double-clocks the state machine

Status: RESOLVED FIXED

(Core :: DOM: Networking, defect, P2)

Description

[Andrew Sutherland [:asuth] (he/him)]

When looking at a weird stack trace :jmarshall was getting from the WorkerRef debugging logic for XHRs I noticed that an unintentional aspect of the worker XHR creating a main-thread XHR on its owning window's global is that we will get DisconnectFromOwner notifications from the window that will race the worker teardown which I do not believe were anticipated / compensated for inside the worker XHR implementation. It's a straightforward fix for us to not call Abort in that case.

Comment 1

[Andrew Sutherland [:asuth] (he/him)]

Attachment: Bug 1900681 - Worker XHR self-manages XHRMT lifecycle. r=#dom-workers!

We don't want the XHRMT to self-Abort() when the XHRWT state machine
is built assuming this doesn't happen. This is likely the source of a
number of weird hangs and crashes we have hardened the implementation
against in the past, and potentially still some ongoing hangs and
weirdness.

Comment 2

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/8e8ee3fc21c3
Worker XHR self-manages XHRMT lifecycle. r=dom-worker-reviewers,smaug

Comment 3

[Cristian Tuns]

https://hg.mozilla.org/mozilla-central/rev/8e8ee3fc21c3