Hi,
currently many tests use this code in its test objects:
raise error.TestFail -or- raise error.TestWarn
to end a test that "failed" (from our perspective). It then creates a few ugly exceptions in the output, see here, even if everything went just fine (from autotest perspective):
http://autoqa.fedoraproject.org/results/250-autotest/qa03.c.fedoraproject.or...
and finally it reports:
09/08 12:48:13 INFO | job:1254| FAIL rpmlint rpmlint timestamp=1283950093 localtime=Sep 08 12:48:13 09/08 12:48:13 INFO | job:1254| END FAIL rpmlint rpmlint timestamp=1283950093 localtime=Sep 08 12:48:13
(which is also visible in autotest web frontend).
However, since we now use jskladan's patchset defining AutoQATest parent class (using self.result variable), I no longer see a reason to raise those exceptions. Quite the contrary. Those exception don't even match our result states, which we have a much richer set (passed, failed, info, aborted, crashed and maybe some other).
I'd like to propose that we stop using those autotest-internal TestFail and TestWarn exceptions.
What will change: 1. We no longer see red/purple boxes in autotest frontend. 2. The AutoQATest parent class will be adjusted to automatically output the test result after a test has finished. 3. Tests will just need to store proper result in self.result variable, that's all. The methods may be then ended in a standard manner (return or no statement).
What will be the benefits: 1. We will know that if we see a traceback it's a problem. We won't have to examine in detail whether this traceback is "good" or "bad". 2. It will allow us to report any crashed test object (throwing an exception) to the mailing list. We have already started to do that and it's great: https://fedorahosted.org/pipermail/autoqa-results/2010-September/thread.html But currently it works only for those tests that haven't filled self.result before the crash occurred. We need to catch all exceptions.
What do you think?