1411644 - Intermittent content process leak of 1 CancelableRunnable in crash tests

Status: RESOLVED DUPLICATE of bug 1416927

(Core :: XPCOM, defect, P3)

Description

[Andrew McCreight [:mccr8]]

Looking at the log, the content process, for two different runs, was created during dom/base/test/chrome/test_bug1209621.xul so maybe this will be easy to reproduce. I see this on Linux, but not Linux64 or OSX.

Comment 1

[Andrew McCreight [:mccr8]]

This leak is actually intermittent, despite my hitting it twice on two runs of crash tests, apparently in the same tests.
https://treeherder.mozilla.org/#/jobs?repo=try&revision=900bdb32b890d45ec07404728348d0e550bfa6a1

It also showed up in this run in bc14, again on Linux32:
https://treeherder.mozilla.org/#/jobs?repo=try&revision=f37778db9f571857eeb86b4d0044a44756b56ec7

Comment 2

[Andrew McCreight [:mccr8]]

With my patches for bug 1410209, the leak now shows up as... RunnableFunction, which is not much more useful.

Comment 3

[Nathan Froyd [:froydnj]]

(In reply to Andrew McCreight [:mccr8] from comment #2)
> With my patches for bug 1410209, the leak now shows up as...
> RunnableFunction, which is not much more useful.

The next step is clearly a cross-platform implementation of lambda introspection.  *ducks*

Comment 4

[Andrew McCreight [:mccr8]]

nsMainThreadPtrHolder<>, which is a similar helper class, simple requires that you pass in a string for the name.

Comment 5

[Andrew McCreight [:mccr8]]

A better example of another class that requires you pass in a name is RunnableFunction. Yes, there are two.

This mName thing seems to work pretty well. My main concern is that it essentially mashes the random string people pass into the runnables into a global namespace used by the leak checker. It could be annoying for people to deal with.

Comment 6

[Andrew McCreight [:mccr8]]

Once bug 1410209 and bug 1411259 have landed, this will show up as an intermittent leak, and hopefully I can figure it out then.