549743 - Abort if cycle collected object is refcounted from non-main-thread

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Description

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Attachment: Patch to fix

To avoid race conditions that cause leaks and crashes sometimes, we should make sure that people that use cycle collected objects from threads other than the main thread know about it.

The easiest way to do that is to abort as soon as someone does it.

This needs to happen early in a release cycle so that developers have time to notice and fix their code.

Now is a good such time.

Comment 1

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Oh, and I don't care strongly about the name here. Feel free to come up with better suggestions.

Comment 2

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Comment on attachment 429849 [details] [diff] [review]
Patch to fix

How about "AbortIfOffMainThreadAndHaveFastMainThreadCheck"?  It's a little more verbose, but it says what it does.

r=dbaron

Comment 3

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Dbaron suggested "IfCheckFastAbortIfOffMainThread" on irc. I'll go with that.

Comment 4

[Michael Kraft [:morac]]

As someone who's add-on is now frequently triggering this abort at startup in the trunk loads, I'm confused as to why that's happening.  It appears to be triggered by a call to nsWindowMediator::GetMostRecentWindow which my add-on isn't making.

See http://forums.mozillazine.org/viewtopic.php?f=19&t=1790145 for details.

Am I to take it that threads can't be safely used until the initial "domwindowopened" notification fires?

Comment 5

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Please file a new bug and attach a stack. And cc me.

Comment 6

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

(In reply to comment #5)
> Please file a new bug and attach a stack. And cc me.

A stack or a link to a crash report.

Comment 7

[Michael Kraft [:morac]]

Never mind I tracked my problem down to a call elsewhere in my code.  While looking at bug reports though I did see a few that occurred with no extensions installed though.

Comment 8

[(not currently active) Ted Mielczarek]

Were there literally no extensions listed? You should look at the uptime field, it's possible that it was a startup crash before we collected the extension data.

Comment 9

[Michael Kraft [:morac]]

Up-time was 1 second so most likely that's what happened.

Comment 10

[Jesse Ruderman]

Sicking landed this a few weeks ago:
http://hg.mozilla.org/mozilla-central/rev/0d5553264cbf.