Try to fix a valgrind leak.

Try to fix a valgrind leak.

This shows up as
Suppression:
{
   <insert_a_suppression_name_here>
   Memcheck:Leak
   fun:_Znw*
   fun:_ZNSt8_Rb_treeISsSt4pairIKSsSt3setIiSt4lessIiESaIiEEESt10_Select1stIS7_ES3_ISsESaIS7_EE11_M_get_nodeEv
   fun:_ZNSt8_Rb_treeISsSt4pairIKSsSt3setIiSt4lessIiESaIiEEESt10_Select1stIS7_ES3_ISsESaIS7_EE14_M_create_nodeERKS7_
   fun:_ZNSt8_Rb_treeISsSt4pairIKSsSt3setIiSt4lessIiESaIiEEESt10_Select1stIS7_ES3_ISsESaIS7_EE9_M_insertEPSt18_Rb_tree_node_baseSE_RKS7_
   fun:_ZNSt8_Rb_treeISsSt4pairIKSsSt3setIiSt4lessIiESaIiEEESt10_Select1stIS7_ES3_ISsESaIS7_EE13insert_uniqueESt17_Rb_tree_iteratorIS7_ERKS7_
   fun:_ZNSt3mapISsSt3setIiSt4lessIiESaIiEES1_ISsESaISt4pairIKSsS4_EEE6insertESt17_Rb_tree_iteratorIS8_ERKS8_
   fun:_ZNSt3mapISsSt3setIiSt4lessIiESaIiEES1_ISsESaISt4pairIKSsS4_EEEixERS7_
   fun:_ZN23ExtensionMessageService24DispatchEventToRenderersERKSsS1_bRK4GURL
   fun:_ZL13DispatchEventP7ProfilePKcSs
   fun:_ZL26DispatchSimpleBrowserEventP7ProfileiPKc
   fun:_ZN27ExtensionBrowserEventRouter17OnBrowserRemovingEPK7Browser
   fun:_ZN11BrowserList13RemoveBrowserEP7Browser
   fun:_ZN7BrowserD0Ev
   fun:object_cxxDestructFromClass
   fun:_ZN12_GLOBAL__N_113ZombieDeallocEP11objc_objectP13objc_selector
   fun:-[NSResponder dealloc]
   fun:-[NSWindowController dealloc]
   fun:-[BrowserWindowController dealloc]
   fun:-[NSWindowController release]
   fun:__delayedPerformCleanup
   fun:CFRunLoopTimerInvalidate
   fun:CFRunLoopRunSpecific
}
, with

c++filt __ZNSt3mapISsSt3setIiSt4lessIiESaIiEES1_ISsESaISt4pairIKSsS4_EEEixERS7_ being map::operator[] and the symbol above it being map::insert().

(e.g. http://build.chromium.org/buildbot/memory/builders/Chromium%20Mac%20UI%20(valgrind)(3)/builds/3538/steps/memory%20test:%20ui/logs/stdio ).

Seems to happen only on mac.

This shouldn't leak, but maybe we're hitting an STL leak (I tried to build a reduced test case for a few minutes and failed). Maybe it goes away with this.

BUG=none
TEST=One fewer "Suppression:" line on mac ui_tests valgrind bots (2 instead of 3 atm).

TBR=mpcomplete

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=52995

[Nico, 2010-07-20 03:10:01]

Not sure why this is suddenly showing up on the memory bots, doesn't look like
anything in this area changed recently.

This is mostly stabbing in the dark.

[Paweł Hajdan Jr., 2010-07-20 03:12:49]

I don't know this code at all, so I'm removing myself from reviewers. The change
seems fine, thanks for working on this.

[Nico, 2010-07-20 04:06:56]

mpcomplete: Since the change is pretty trivial, and since valgrind cycle times
are so long, I'll tbr this. Please take a look anyway.

[Matt Perry, 2010-07-20 17:55:40]

Seems reasonable, though if this were leaking, I would expect the
ExtensionMessageService to also be leaking. I have seen memory errors with STL
containers in this class before for some reason.

LGTM.