launchpad-dev team mailing list archive

[Michael Hudson <michael.hudson@xxxxxxxxxxxxx>]

On Thu, 7 Oct 2010 19:33:43 +0100, Graham Binns <graham@xxxxxxxxxxxxx> wrote:
> On 7 October 2010 19:21, Francis J. Lacoste
> <francis.lacoste@xxxxxxxxxxxxx> wrote:
> > And additionnally, spot checks seem to indicate that these tests continue to
> > run successfully in subsequent builds.
> >
> > So:
> >
> > a) these tests were not skipped
> > b) they pass in the buildbot environment
> >
> > Another case, where the difference between our dev, production / test
> > environement is hurting us.
> >
> 
> Note that the test that I fixed, which is now in devel, was failing
> due to an ordering problem and always passed locally; I'm not sure
> that there's much we can do to avoid those except remind reviewers to
> be vigilant for potential issues.

Back in the SQLObject days we had a hack that would add "ORDER BY
random()" to any query that didn't have an ORDER BY already.  Do we
still have that?  Although in this case it seems we had an ORDER BY,
just not a sufficienly discriminating one.  Could you add ", random()"
to any query that does have an ORDER BY?

--

Separately, it seems this problem may have been perpetuated by simply
changing a doctest to match observed output, and not understanding why
the output had changed.  Let's not do that again? [1]

Cheers,
mwh

[1] Of course, this is harder if the tests are not clear about what they
    are testing and why.  Most of you heard me rant about this at the
    epic.