ffc team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Thu, Oct 08, 2009 at 02:27:12PM +0200, Kristian Oelgaard wrote:
> Quoting Anders Logg <logg@xxxxxxxxx>:
>
> > The FFC buildbot is currently failing on jaunty-amd64 and
> > linux64-exp. The difference seems to be a variable in the generated
> > code named FE0_f0 on some systems (including mine) and FE0_f1 on the
> > buildbot.
> >
> > Johannes: Do you know if there is a difference in the configuration of
> > Python versions etc on those machines that can affect this?
> >
> > Kristian: How are those variable names generated?
>
> I blame dict and set (although it is my fault). When I extract unique elements
> in all forms I do it by updating a set. Afterwards, I do something like:

I suspected something like set or dict was the problem.

> for i, e in enumerate(list(set())):
>   ...

How about

  for i, e in enumerate(sorted(list(set()))):

?

> where (i,e) are not guaranteed to be the same pairs, not even for subsequent
> runs on the same machine. I guess I should fix this to make our tests robust,
> but it is difficult since I don't get the error.
> If you update those references that show the discrepancy, then it _should_ fail
> on my machine and I can see if I can remove the 'error'.

I don't know if that's possible. It doesn't fail on my machine.

You can check what's wrong here:

http://fenics.org:8080/builders/ffc-jaunty-amd64/builds/83/steps/ffc%20check/logs/stdio

If you make a change, just push and then force a rebuild on the
buildbot and hope for the best.

--
Anders

Attachment: signature.asc
Description: Digital signature