341792 - major memory leak in nsLocalFile::AppendNative

Status: VERIFIED WORKSFORME

(Core :: XPCOM, defect, P1)

Description

[Josh Aas]

When doing some memory leak profiling, almost every single leak comes from nsLocalFile::AppendNative. The leaked memory is allocated with this stack:

nsLocalFile::AppendRelativeNativePath(nsACString_internal const&)
nsLocalFile::AppendNative(nsACString_internal const&)
CFURLCreateCopyAppendingPathComponent
CFStringCreateWithFormat
_CFStringCreateWithFormatAndArgumentsAux
CFStringCreateCopy
__CFStringCreateImmutableFunnel3
_CFRuntimeCreateInstance
MDNXZoneMalloc

This adds up to a significant chunk of memory leaked in very little time.

Comment 1

[Mike Schroepfer]

Josh you on this one?

Comment 2

[Josh Aas]

I'll get to this within a week.

Comment 3

[Josh Aas]

Using the leaks tool in Instruments on 10.5, I don't see a single instance of this leak on any more. Will  test on 10.4 soon to see if this might be an OS or leak tools issue. Probably though this leak just went away at some point.

Comment 4

[Josh Aas]

This leak is totally gone on 10.4 as well, using the same tools I found it with in the first place. Not a single leaked byte of memory from AppendNative.

Could be anything that fixed it, like cycle collector getting objects that allocated memory in AppendNative. Also, Boris fixed bug 389191 and while it probably isn't related it might have something to do with this. I looked at AppendNative, the CFURLCreateCopyAppendingPathComponent memory isn't leaking directly in that function.

Comment 5

[Mark Mentovai]

Could've even been an OS leak fixed in a Tiger update.

Comment 6

[Carsten Book [:Tomcat]]

can not reproduce this leak, but will do a Full Functional Test on Mac 10.5 to detect leaks. So i will close this bug as verified fixed.