On Mon, Dec 3, 2012 at 1:18 PM, Jakub Jelinek <ja...@redhat.com> wrote: > On Mon, Dec 03, 2012 at 11:06:48AM +0400, Konstantin Serebryany wrote: >> This patch copies the asan tests almost, but not quite, verbatim from >> upstream. >> Since the patch is not in attachment (and gmail messes up with inlined >> patches) I can't see the exact changes. > > Sending patches inline rather than as attachments is the preferred way for > gcc-patches, you can always grab it from the gcc-patches ml archive: > http://gcc.gnu.org/ml/gcc-patches/2012-11/msg02557.html > (then click on Raw text).
Got it, thanks. > > I'm attaching the diff for asan_test.cc from llvm anyway. > >> I see #ifdef ASAN_AVOID_EXPENSIVE_TESTS, which I don't really like >> because I'd rather fix the test than disable it. > > The test isn't disabled, just by default limited to 30 threads instead of > 1000, because that really will ruin testing for everybody with ulimit -u > in 1024-ish range. Even 500 threads would be undesirable for that. Which is the same as disabling it. Unfortunately, we don't have a good automated way to test asan performance, so this test is guarding us from performance degradation in the asan's pthread wrappers. If there is a bug there, we may not notice it on 30 threads, but may (with a high probability) on 1000 threads. Anyway, does http://llvm.org/viewvc/llvm-project?rev=169118&view=rev solve the problem? It adds ASAN_AVOID_EXPENSIVE_TESTS and checks the results of all pthread_{create,join} calls in tests. --kcc > >> Can we commit the tests 100% verbatim, and then fix them as separate commits >> (preferably, by fixing the tests upstream and doing the merge with >> libsanitizer/merge.sh)? > > I'd prefer delay committing the patch over causing random testsuite failures > elsewhere. > > Jakub
