On 08/05/13 at 02:18pm, Allan McRae wrote: > On 05/08/13 14:16, Andrew Gregory wrote: > > On 08/05/13 at 10:52am, Allan McRae wrote: > >> On 02/08/13 22:34, Andrew Gregory wrote: > >>> This patchset converts the output of all of our tests to tap [1] and fully > >>> integrates them with automake so that tests can be run in parallel with > >>> `make > >>> check`. The test suite may also be run with other test harnesses such as > >>> perl's prove which can do such interesting things as remember which tests > >>> failed and run only those on subsequent invocations. The documentation > >>> for > >>> integrating tests is here [2]. > >>> > >>> [1] http://podwiki.hexten.net/TAP/TAP.html?page=TAP > >>> [2] > >>> http://www.gnu.org/software/automake/manual/html_node/Parallel-Test-Harness.html > >>> > >> > >> Have you any ideas on how to fix the "unexpected" pass on the time test > >> for x86_64 to not have the test suite return non-zero? I believe this > >> is essential. > >> > >> Allan > > > > I think that "unexpected" passes are rightly considered failures. The test > > should reflect what we actually expect to happen. We should either update > > the > > test so that it succeeds or fails uniformly on all systems or set > > expectfailure > > only on systems where we actually expect it to fail. Personally, I would > > prefer that the test use the maximum values that the testing system could be > > expected to support and unset expectfailure, but the easier solution is to > > just > > set expectfailure only on 32 bit systems. > > Setting expected failure on 32bit systems would actually be my preferred > solution in this case. Can our test suite handle that?
I don't have any 32-bit systems readily available to test it at the moment, but checking either platform.architecture [1] or sys.maxsize [2] should be sufficient. [1] http://docs.python.org/2/library/platform.html#platform.architecture [2] http://docs.python.org/2/library/sys.html#sys.maxsize
