This bug has been spun off from bug 517355 comment #55. Since the patches for bug 517355 and bug 521388 landed on the 1.9.2 branch (on 2009-10-09), mochitests and (sometimes) crashtests trigger the following leaks: leaked 3616 bytes during test execution leaked 2 instances of nsCLiveconnect with size 4 bytes each (8 bytes total) leaked 1 instance of nsComponentManagerImpl with size 276 bytes leaked 1 instance of nsJVMManager with size 4 bytes leaked 3 instances of nsLocalFile with size 1044 bytes each (3132 bytes total) leaked 1 instance of nsNPAPIPlugin with size 96 bytes leaked 1 instance of nsPluginHost with size 84 bytes leaked 3 instances of nsTArray_base with size 4 bytes each (12 bytes total) leaked 1 instance of nsVoidArray with size 4 bytes http://tinderbox.mozilla.org/showlog.cgi?log=Firefox3.6-Unittest/1255123815.1255126224.11988.gz http://tinderbox.mozilla.org/showlog.cgi?log=Firefox3.6-Unittest/1255124353.1255126125.10906.gz Starting Monday, I'll be trying to find and fix these.

Hey Steven, I've been digging through this and I noticed that nsPluginTag::TryUnloadPlugin lost its aForceShutdown parameter (and it didn't come back in your merge). Because of this, we never call the plugin's Shutdown method, which in the MRJ plugin releases the component manager and its allocated nsCLiveconnects. For reference, the "skip plugin Shutdown" code was removed in bug 488042. The then defunct parameter was removed in bug 500925.

(In reply to comment #2) I suspect this will cause crashes using Java. I'll test on Monday.

To stop these leaks interfering with automated testing, dbaron temporarily raised the leak threshold (on OS X) for mochitests and crashtests: http://hg.mozilla.org/releases/mozilla-1.9.2/rev/1226d4feb6cb This can be reversed once this bug is fixed.

(Following up comment #3) Odd that it didn't cause crashes before, I admit. I'll dig further into this on Monday. Thanks for the tip.

(Following up comment #0) > Since the patches for bug 517355 and bug 521388 landed on the 1.9.2 > branch (on 2009-10-09), mochitests and (sometimes) crashtests trigger > the following leaks: That's bug 517355 and *bug 521338*.

Comment on attachment 405706 [details] [diff] [review] Possible leak fix Yes, this patch seems to fix the leaks. And no Java crashes. Furthermore, looking again at the code, I needn't have worried about them.

Comment on attachment 405706 [details] [diff] [review] Possible leak fix This bug blocks a P1 blocker, so it should probably also block. But to observe the formalities ...