dolfin team mailing list archive

[Johannes Ring <johannr@xxxxxxxxx>]

On Fri, Jan 20, 2012 at 3:00 PM, Johannes Ring <johannr@xxxxxxxxx> wrote:
> On Fri, Jan 20, 2012 at 2:05 PM, Anders Logg <logg@xxxxxxxxx> wrote:
>> On Fri, Jan 20, 2012 at 01:59:12PM +0100, Johannes Ring wrote:
>>> On Fri, Jan 20, 2012 at 1:57 PM, Anders Logg <logg@xxxxxxxxx> wrote:
>>> > On Fri, Jan 20, 2012 at 01:52:34PM +0100, Johannes Ring wrote:
>>> >> On Fri, Jan 20, 2012 at 1:44 PM, Johan Hake <johan.hake@xxxxxxxxx> wrote:
>>> >> > On Friday January 20 2012 13:39:51 Anders Logg wrote:
>>> >> >> On Fri, Jan 20, 2012 at 12:12:59PM +0100, Martin Sandve Alnćs wrote:
>>> >> >> > Is anyone planning to get the dolfin trunk buildbot green anytime
>>> >> >> > soon?
>>> >> >> >
>>> >> >> > I have some ufl bugfixes I'd rather not push before everything is
>>> >> >> > green. No rush though.
>>> >> >>
>>> >> >> Who broke it?
>>> >> >
>>> >> > Looks like my outcommenting of one test triggered the lucid buildbot to fail
>>> >> > on another test... Not a good sign, but I can take a look at that error.
>>> >> >
>>> >> > Not sure about the macbots.
>>> >>
>>> >> There is a segmentation fault on osx-10.7 (bug #918178) and osx-10.6
>>> >> fails with a PETSc error when running some of the demos:
>>> >>
>>> >> [0]PETSC ERROR: --------------------- Error Message
>>> >> ------------------------------------
>>> >> [0]PETSC ERROR: Unknown type. Check for miss-spelling or missing
>>> >> external package needed for type
>>> >>  seehttp://www.mcs.anl.gov/petsc/petsc-as/documentation/installation.html#external!
>>> >> [0]PETSC ERROR: Unable to find requested PC type hypre!
>>> >>
>>> >> The demo.log on the buildbot contains the complete error message:
>>> >>
>>> >>   http://fenicsproject.org:8090/builders/dolfin-osx-10.7/builds/12/steps/dolfin%20check/logs/demo.log
>>> >
>>> > That file seems to be empty.
>>>
>>> Sorry, it should be this one:
>>>
>>> http://fenicsproject.org:8090/builders/dolfin-osx-10.6/builds/72/steps/dolfin%20check/logs/demo.log
>>
>> The only hint I see from that log is that hypre is not available.
>> Nothing regarding hypre has been changed between 1.0.0 and trunk in
>> the code. Has something changed in the configuration? Maybe hypre is
>> not available for the trunk build?
>
> I have used Dorsal with petsccore (no Hypre) on the OS X 10.6
> buildbot. It is the same for both 1.0.x and trunk buildbots, however,
> I see now that 1.0.x has PETSc 3.1, while trunk has PETSc 3.2. I will
> see if I can get Dorsal to build petsc (instead of petsccore) on OS X
> 10.6. It works for 10.7.

I successfully got Dorsal to build 'petsc' on OS X yesterday, although
I had to install OpenMPI from MacPorts, which resulted in a lot of
other packages that had to be rebuilt. This was all successful and
DOLFIN built fine, however the tests failed to run on the buildbot:

  http://fenicsproject.org:8090/builders/dolfin-osx-10.6/builds/79/steps/dolfin%20check/logs/stdio

I found that the problem lies in
site-packages/dolfin/importhandler/dynamicloadmpi.py, where we force
load libmpi.0.dylib, the system installed OpenMPI library, and not the
one installed by MacPorts, which is libmpi.1.dylib. There are symlinks
to these in /usr/lib and /opt/local/lib, respectively, so in theory we
should be able to replace libmpi.0.dylib with libmpi.dylib in
dynamicloadmpi.py. However, it turns out that the stuff in
dynamicloadmpi.py is not needed on OS X. In fact, it has never been
used on Lion since Lion doesn't come with any OpenMPI compilers and
there is therefore no libmpi.0.dylib available (only libmpi.1.dylib
from MacPorts).

I have tested this on both of the OS X buildbots and I suggest we
replace the code in dynamicloadmpi.py with this:

  import ctypes
  ctypes.CDLL('libmpi.so', ctypes.RTLD_GLOBAL)

Doing so should result in one more green buildbot (osx-10.6). Any
objections before I push?

Johannes