525573 - TestHarness.h doesn't initialize nsTraceMalloc in NS_TRACE_MALLOC builds

Status: RESOLVED DUPLICATE of bug 529134

(Core :: XPCOM, defect)

Description

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

This causes nsTraceMalloc not to allocate its PRLock, which in turn causes it to not protect its shared state.

This is very bad for multi-threaded unit tests in DEBUG builds: the deadlock detector uses nsTraceMalloc to get backtraces.  With new code I'm adding, I can cause an infinite loop in nsTraceMalloc due to state corruption in about 1/2 runs (on my machine with 16 HW threads).

Comment 1

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Attachment: v1

Comment 2

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Pushed http://hg.mozilla.org/mozilla-central/rev/247abc20a78b

Comment 3

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

This fails to link on OS X/Windows debug/libxul/tracemalloc, although everything is hunky dory on Linux for reasons beyond my ken.  Looking into build hackery.

Comment 4

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

I just confirmed that the OS X build error is caused by content/base/test *not* linking against libxul, while xpcom/test *is*.  Trying to determine how it's magically working on linux.

Comment 5

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

With bug 529134, trace-malloc will be implicitly initialized, so this bug is no longer necessary.

I really want RESOLVED -> SUBSUMED.