On 09.02.2015 11:11, Philip Martin wrote: > Branko Čibej <br...@wandisco.com> writes: > >> The latest last-minute changes introduced a heisencrash in ra-test, and >> made some C tests run via DAV look for global credentials (e.g., the >> keychain on the Mac). I haven't tracked these down yet, but surely we >> can keep trunk more stable than that. > Does the ra-test crash happen in --parallel mode only? Running C tests > in parallel is new and ra-test.c doesn't follow the rules for calling > svn_ra_initialize: there is no protection against concurrent calls and > the pool passed has the wrong lifetime.
It turns out that it's not a crash. It also turns out that it's not new. The test hangs deep in BDB within cleanup-fs during the destruction of the top-level pool; but not always. The following script consistently reproduces this for me on trunk: cd $(builddir)/subversion/tests/cmdline while ../libsvn_ra/ra-test --fs-type=bdb; do echo '-----'; done It usually takes a few iterations, but it will fail eventually during cleanup for test #3. It's specific to BDB, and it could be specific to OSX. I verified that this is not new by building a version of trunk from three months ago; same issue. I'm just surprised we haven't seen it before (hence the suspicion that it's OSX-only). -- Brane
