Flaky asan test?

My change 225282 broke an asan test [1], but by the time I got around
to trying to revert it, I noticed that the test had started passing.
Moreover, it seems to have been "fixed" by 225291 which at least does
not seem related to the breakage directly.

I'm puzzled now -- is the failing asan test flaky? Should I still
revert 225282?

Thank you for your time!

[1]: http://lab.llvm.org:8011/builders/libcxx-libcxxabi-x86_64-linux-ubuntu-asan/builds/98/steps/test.libcxx/logs/stdio

-- Sanjoy

+eric, who introduced the test in r222076.
Eric, how does the test fail under asan?
Do you expect a leak report?

Just to note, my change did make an existing bug more likely. The bug
is being tracked at http://llvm.org/bugs/show_bug.cgi?id=22179, and I
have a fix up for review: http://reviews.llvm.org/D6914

Hmm… Stil don’t see how this may affect this test…

If I run check-libcxx in the asan build (-DLLVM_USE_SANITIZER=Address) I see that the test reports a true leak.
However, if I run the test many times asan/lsan miss the leak occasionally.
Sergey, could you please take a look? The leak happens in the 2-nd thread, we sometimes fail to report it.