Bug 40654 – Database access in worker threads results in memory leaks due to lack of autorelease pool

• Status: NEW
• Component: WebCore Misc.

[Mark Rowe (bdash)]

You can see the following stderr output from a number of worker tests while running the regression tests:

2010-06-15 22:13:15.128 DumpRenderTree[29876:766f] *** __NSAutoreleaseNoPool(): Object 0x11a5a2f70 of class NSCFString autoreleased with no pool in place - just leaking

If you look at <http://build.webkit.org/results/SnowLeopard%20Intel%20Leaks/r61233%20(8108)/DumpRenderTree2-leaks.txt> you can also see that these objects are in fact leaked.

[Darin Adler]

http://trac.webkit.org/changeset/61626
http://trac.webkit.org/changeset/61627

[Alexey Proskuryakov]

See also: bug 40655.

[Simon Fraser (smfr)]

So callOnMainThread() requires an autorelease pool?

[Simon Fraser (smfr)]

Here's where the autorelease spew is coming from:

#0  0x00007fff80fe6d94 in __NSAutoreleaseNoPool ()
#1  0x00007fff80f32099 in _CFAutoreleasePoolAddObject ()
#2  0x00007fff80f31e06 in -[NSObject(NSObject) autorelease] ()
#3  0x00000001019352ed in HardAutorelease (obj=0x105c3c960) at FoundationExtras.h:73
#4  0x0000000101935311 in WTF::StringImpl::operator NSString* (this=0x105c07c70) at /Volumes/Monster/Development/apple/webkit/WebKit.git/WebCore/platform/text/mac/StringImplMac.mm:30
#5  0x000000010085119f in WTF::String::operator NSString* (this=0x108fa7440) at WTFString.h:277
#6  0x00000001008a3158 in WebDatabaseTrackerClient::dispatchDidModifyDatabase (this=0x107810ec0, origin=0x10790cc10, databaseIdentifier=@0x108fa7440) at /Volumes/Monster/Development/apple/webkit/

WebDatabaseTrackerClient::dispatchDidModifyDatabase() is doing Obj-C stuff.

[Simon Fraser (smfr)]

Or is it wrong for dispatchDidModifyDatabase() to be called on the worker thread?

[Simon Fraser (smfr)]

Bug 40655 should fix that.