This is a regression from one of these bugs: bug 721543, bug 721548, bug 721515. When I look in the debug logs post landing, I see the following leak listed: == BloatView: ALL (cumulative) LEAK STATISTICS, default process 18305 |<----------------Class--------------->|<-----Bytes------>|<----------------Objects---------------->|<--------------References-------------->| Per-Inst Leaked Total Rem Mean StdDev Total Rem Mean StdDev 0 TOTAL 21 224 6595 5 ( 321.40 +/- 429.36) 17243 2 ( 1810.46 +/- 2094.06) 34 Mutex 12 12 64 1 ( 32.25 +/- 18.40) 0 0 ( 0.00 +/- 0.00) 50 ReentrantMonitor 16 16 15 1 ( 6.38 +/- 3.31) 0 0 ( 0.00 +/- 0.00) 168 nsTArray_base 4 4 484 1 ( 155.02 +/- 66.06) 0 0 ( 0.00 +/- 0.00) 169 nsThread 124 124 4 1 ( 2.29 +/- 1.11) 66 1 ( 15.30 +/- 5.79) 171 nsTimerImpl 68 68 3 1 ( 1.40 +/- 0.55) 9 1 ( 1.94 +/- 0.75) This occurs for the majority of the xpcshell-tests given.

So, STR, please.

STR: 1) Build Firefox (or Thunderbird) in debug mode using latest code. 2) run make -c <objdir> xpcshell-tests Results: See the bloat logs in the majority of the results. Example logs: https://tbpl.mozilla.org/php/getParsedLog.php?id=9001372&tree=Firefox https://tbpl.mozilla.org/php/getParsedLog.php?id=9002509&tree=Firefox https://tbpl.mozilla.org/php/getParsedLog.php?id=9002824&tree=Firefox https://tbpl.mozilla.org/php/getParsedLog.php?id=9002738&tree=Firefox https://tbpl.mozilla.org/php/getParsedLog.php?id=9003706&tree=Firefox https://tbpl.mozilla.org/php/getParsedLog.php?id=9003238&tree=Firefox Just running xpcshell-tests on my local build of Thunderbird reproduced this. So you shouldn't need to make a packaged test or anything.

I guess some of the little marking arrays aren't being cleaned up properly. I wonder if this is due to some XPCshell weirdness, or because the tests are so short. With mochitest, many tests are run, so maybe it has proper time to finish up.

XPCShell is in many ways non-normal execution environment, so it is possible that cycle collector isn't called the same way as in browser.

Don't create the timer if we're shutting down.

Comment on attachment 593471 [details] [diff] [review] patch [Approval Request Comment] Regression caused by (bug #): bug 721543 User impact if declined: N/A Testing completed (on m-c, etc.): Landed m-c Risk to taking this patch (and alternatives if risky): Some leaks in Aurora may not be noticed because of this minor (shutdown) leak. String changes made by this patch: N/A

Comment on attachment 593471 [details] [diff] [review] patch [Triage Comment] Approve for Aurora 13 to uncover hidden leaks.