On Tue, Mar 10, 2009 at 02:56:08PM +0100, Branko Cibej wrote:
> Bert Huijben wrote:
> >> Isn't the whole point of the test suite to identify bugs? So, if you
> >> unconditionally Skip a test that fails on some platforms, it'll never
> >> get fixed because it'll never get run, and no-one ever reviews skipped
> >> tetss. Might as well remove the test then, what's the point?
> >>
> >> Or maybe fix the bug, who knows.
> >>
> >
> > Windows fails, mac-os partly passes, one linux buildbot passes, another linux buildbot fails.. (Unrelated to filesystem, ra layer, etc.)
> >
> > Which check would you like to apply?
> >
> >
> > These tests were merged to trunk in a broken state from the svnpatch branch and they hinder the other work that is going on. The buildbots are there to identify errors.. But until these tests are passing on all platforms they hide all the other new errors.
> >
>
> Ok, this may sound too pontificating, but I have to ask ... why was the
> merge even committed if it creates new errors? We agreed not too long
> ago that we'd agreed a long time ago that trunk should be stable ... or
> at least that tests on trunk should pass.
Yes, tests should pass, but keep in mind that in this particular case,
Arfrever might not even have noticed these failures before committing
because they only appear on certain systems, possibly not his.
I'd say leave it in trunk, and let the tests fail, and try to fix the bug.
We'll have to fix the bug anyway. Backing this out or tweaking the tests
creates more work on top of that. If the failure was trivially reproducible,
I'd prefer backing out. But if only certain systems show the failure,
it's better not to back out to see where it fails and where it does not.
I think we can afford the risk of hiding new errors for a short while.
Bert, in the meantime, just don't commit code with errors in it :P
Is anyone working on the real fix?
Stefan
Received on 2009-03-10 15:17:33 CET