I removed all but this test from <...\toolkit\components\feeds\test\xml\> |clear && env XPCOM_MEM_LEAK_LOG=1 make -C .../toolkit/components/feeds check| [Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.1a2pre) Gecko/20080825125202 Minefield/3.1a2pre] (home, optim default) (W2Ksp4) [ 154 nsStringBuffer 8 3624 16966 453 ( 4827,21 +/- 3266,99) 27484 453 ( 6484,84 +/- 4463,57) ] [Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.1b1pre) Gecko/20080901134927 SeaMonkey/2.0a1pre] (home, optim default) (W2Ksp4) [ 154 nsStringBuffer 8 3600 17462 450 ( 4476,82 +/- 3310,71) 28649 450 ( 6025,43 +/- 4544,52) ]

Ah, I had already noticed this in bug 448980 comment 0. NB: I don't know what this file has different than all the other feed test files to trigger this.

[Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.1b1pre) Gecko/20080901143156 SeaMonkey/2.0a1pre] (home, debug default) (W2Ksp4) Fwiw, [ nsStringStats => mFreeCount: 18345 -- LEAKED 450 !!! ]

Something going wrong with the nsScriptNameSpaceManager, maybe?

With this debugging patch, you can see that the nsScriptNameSpaceManager is being leaked. If you set XPCOM_MEM_ALLOC_LOG=alloc.log and XPCOM_MEM_LOG_CLASSES=nsScriptNameSpaceManager you can even see the stack to its allocation ;) I don't think I'll be able to get much further in debugging this bug.

Comment on attachment 337173 [details] [diff] [review] patch to instrument nsScriptNameSpaceManager [Checkin: Comment 13] Sure. The real issue here is that we never call nsJSRuntime::Shutdown in this case....

What would normally call nsJSRuntime::Shutdown? I don't see any obvious callers in the tree.

I guess nsDOMScriptObjectFactory::Observe is supposed to call it. I wonder why it's not. Is there a way to get this shutdown in a debugger?

This looks like a regression from DOM_AGNOSTIC (bug 255942). We never create the nsDOMScriptObjectFactory, so we never register the observer for XPCOM shutdown. nsJSRuntime::GetNameSpaceManager does get called though, and creates and caches the script namespace manager. I'd rather remove the Shutdown method on nsIScriptRuntime completely, after this patch both JS and Python would have an empty Shutdown method. Other language runtimes (hah!) can just use an XPCOM shutdown observer to release stuff.

Peter, did you mean to ask jst for review?

Grr, yes. And it looks like I also forgot to qrefresh before attaching. Trying again.

This doesn't leak for me, but it might be safer to still release mLanguageArray in nsDOMScriptObjectFactory::Observer.

Comment on attachment 337173 [details] [diff] [review] patch to instrument nsScriptNameSpaceManager [Checkin: Comment 13] http://hg.mozilla.org/mozilla-central/rev/4a57478cce2a

[Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.1b1pre) Gecko/20080911004812 Minefield/3.1b1pre] (home, optim default) (W2Ksp4) New leak report :-) [ 1 nsScriptNameSpaceManager 450 nsStringBuffer ]

(In reply to comment #14) Moreover, 40 of the 69 |make check| tests which leak leak (1) nsScriptNameSpaceManager ! We have a winner here ;->

Comment on attachment 337929 [details] [diff] [review] v1 [Checkin: Comment 16] http://hg.mozilla.org/mozilla-central/rev/a530d5f2f15a with additional [ /layout/build/nsLayoutStatics.cpp 7 -#include "nsDOMScriptObjectFactory.h" ]

(In reply to comment #15) [Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.1b1pre) Gecko/20080911163138 Minefield/3.1b1pre] (home, optim default) (W2Ksp4) Compiled m-c up-to-before comment 16 checkin: *36 tests report exactly this leak. *test_bug299716.js: + 1 nsStringBuffer *test_placesTxn.js: + yet a few more objects *test_0030_general.js and test_0110_general.js + yet 1 XPCWrappedNative and 1 XPCWrappedNativeProto

(In reply to comment #17) [Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.1b1pre) Gecko/20080911182230 Minefield/3.1b1pre] (home, optim default) (W2Ksp4) Compiled m-c up-to-including comment 16 checkin: *The 36 tests are fixed. *The other 4 have their other leaks remaining only. V.Fixed Jesse, Boris, Peter: thanks :-))

(In reply to comment #17) > *test_bug299716.js: > + 1 nsStringBuffer I filed Bug 454842 - test_bug299716.js leaks 1 nsStringBuffer, due to (2) |env.set("XPCOM_DEBUG_BREAK", "...");| > *test_placesTxn.js: > + yet a few more objects > *test_0030_general.js and test_0110_general.js > + yet 1 XPCWrappedNative and 1 XPCWrappedNativeProto I updated Bug 449240 - |make check|: 30 tests report the same 6 leaked objects (plus others)