On Tue, Nov 26, 2013 at 2:01 PM, Branko Čibej <[email protected]> wrote:
> On 26.11.2013 13:42, Stefan Fuhrmann wrote: > > On Tue, Nov 26, 2013 at 12:42 PM, Branko Čibej <[email protected]>wrote: > >> On 25.11.2013 17:32, [email protected] wrote: >> >> Author: stefan2 >> Date: Mon Nov 25 16:32:52 2013 >> New Revision: 1545338 >> >> URL: http://svn.apache.org/r1545338 >> Log: >> * subversion/tests/libsvn_client/client-test.c >> subversion/tests/libsvn_fs/fs-test.c >> subversion/tests/libsvn_fs/locks-test.c >> (test_funcs): Move particularly expensive tests to the top of the >> list to maximize parallism. None of these test lists >> imply a specific "simply-to-complex" ordering. >> >> >> You have a pool cleanup bug in the --parallel C tests mode. After these >> last two commits: >> > > Hm. I don't see these problems here nor on our Windows > buildbot (at least svn-windows-local runs in parallel mode). > > > Well, the malloc implementation on OSX has an invalid-free check; I know > Linux doesn't, and I don't know about Windows. > > Valgrind did not show any problem either. > > > Dunno about that. Philip reported a valgrind error. > > So, all threaded tests fail, i.e. any fix should be reproducible > on your system. > > > Yes. > > Hm. So, this happens after all tests completed. I tried some > fix in r1545634 but Philip just told me that it did not help. > > > Shouldn't cleanup_pool be cleared *before* test_pool? Handlers registered > in cleanup_pool may refer to data allocated from test_pool ... > I agree but that is the order that we use for non-threaded testing, so I kept it. -- Stefan^2.
