On Thu, Aug 31, 2017 at 04:39:49PM +0100, Stefan Hajnoczi wrote: > On Wed, Aug 30, 2017 at 06:40:29PM -0400, John Snow wrote: > > > > > > On 08/30/2017 06:35 PM, Eric Blake wrote: > > > On 08/30/2017 05:28 PM, John Snow wrote: > > > > > >> I'm a little iffy on this patch; I know that ./check can take care of > > >> our temp files for us now, but because each python test is itself a > > >> little mini-harness, I'm a little leery of moving the teardown to setup > > >> and trying to pre-clean the confetti before the test begins. > > >> > > >> What's the benefit? We still have to clean up these files per-test, but > > >> now it's slightly more error-prone and in a weird place. > > >> > > >> If we want to try to preserve the most-recent-failure-files, perhaps we > > >> can define a setting in the python test-runner that allows us to > > >> globally skip file cleanup. > > > > > > On the other hand, since each test is a mini-harness, globally skipping > > > cleanup will make a two-part test fail on the second because of garbage > > > left behind by the first. > > > > > > > subtext was to have per-subtest files. > > > > > Patch 5 adds a comment with another possible solution: teach the python > > > mini-harness to either clean all files in the directory, or to relocate > > > the directory according to test name, so that each mini-test starts with > > > a fresh location, and cleanup is then handled by the harness rather than > > > spaghetti pre-cleanup. But any solution is better than our current > > > situation of nothing, so that's why I'm still okay with this patch as-is > > > as offering more (even if not perfect) than before. > > > > > > > I guess where I am unsure is really if this is better than what we > > currently do, which is to (try) to clean up after each test as best as > > we can. I don't see it as too different from trying to clean up before > > each test. > > > > It does give us the ability to leave behind a little detritus after a > > failed run, but it's so imperfect that I wonder if it's worth shifting > > this code around to change not much. > > An alternative is to define iotests.QMPTestCase.setUp() so it clears out > iotests.test_dir. Unfortunately this still requires touching up all > setUp() methods so that they call super(TheClass, self).setUp(). > > At least there would be no need to delete specific files by name (e.g. > blind_remove(my_img)). >
One reason to only remove specific files used in the test, is that it increases the chance that intermediate files will be left behind in case of test failure of a different test case. I think the real long-term solution is to run each unittest test case in its own subdirectory, so that no intermediate file removal is necessary, and each test case is self-contained. -Jeff
