dolfin team mailing list archive

[Johannes Ring <johannr@xxxxxxxxx>]

On Mon, Dec 3, 2012 at 1:24 PM, Anders Logg <logg@xxxxxxxxx> wrote:
> Most buildbots are now green, except for timeouts on the following buildbots:
>
> http://fenicsproject.org:8010/builders/dolfin-trunk-full-mpich/builds/176/steps/make%20run_unittests/logs/stdio
>
> http://fenicsproject.org:8010/builders/dolfin-trunk-full-precise-i386/builds/248/steps/make%20run_unittests/logs/stdio
>
> http://fenicsproject.org:8010/builders/dolfin-trunk-full-squeeze-amd64/builds/289/steps/make%20run_unittests/logs/stdio
>
> The timeouts occur when running in parallel the FunctionSpace unit
> tests (first bot) and for the XMLVector unit tests (last 2 bots).

Actually, when testing this manually on these buildbots, they all fail
the FunctionSpace unit test with the following output:

$ mpirun -np 3 python FunctionSpace.py
Process 0: Number of global vertices: 729
Process 0: Number of global cells: 3072
Process 0: Number of global vertices: 8
Process 0: Number of global cells: 6
Process 0: *** Warning: Mesh is empty, unable to create connectivity 2 --> 3.
Process 0: *** Warning: Mesh is empty, unable to create entities of dimension 2.
Process 0: *** Warning: Mesh is empty, unable to create entities of dimension 2.
Process 0: *** Warning: Mesh is empty, unable to create entities of dimension 2.
E^Cmpirun: killing job...

However, they all work fine when using SCOTCH instead of ParMETIS
(parameters["mesh_partitioner"] = "SCOTCH").

Note that none of these buildbots failed the XMLVector unit test. I'm
not sure why it looks like that in the buildbot stdio output.

Also note that these three buildbots also fails on
unit/la/python/LinearOperator.py and unit/mesh/python/BoundaryMesh.py
(and possibly more). I have reported two different bugs about this,
but I think they can be merged:

https://bugs.launchpad.net/dolfin/+bug/1083174
https://bugs.launchpad.net/dolfin/+bug/1084505

> Is there a way to view the output from the unit tests other than what
> is currently printed? We write the output to fail.log but only for
> tests that actually terminate and produce some output.

I'm not sure. I usually just log in on the failing buildbot and run
the test manually. Then I report the failure on the list or file a new
bug.

Johannes