← Back to team overview

dolfin team mailing list archive

Re: UFC Python wrapper location change?

 

On Aug 20 2010, Johan Hake wrote:

On Friday August 20 2010 13:58:18 Garth N. Wells wrote:
On Aug 20 2010, Anders Logg wrote:
>On Fri, Aug 20, 2010 at 08:29:44PM +0100, Garth N. Wells wrote:
>> On Fri, 2010-08-20 at 21:05 +0200, Anders Logg wrote:
>> > On Fri, Aug 20, 2010 at 11:01:40AM -0700, Johan Hake wrote:
>> > > On Friday August 20 2010 03:10:27 Anders Logg wrote:
>> > > > On Fri, Aug 20, 2010 at 11:03:21AM +0100, Garth N. Wells wrote:
>> > > > > On Fri, 2010-08-20 at 11:56 +0200, Anders Logg wrote:
>> > > > > > On Fri, Aug 20, 2010 at 10:54:04AM +0100, Garth N. Wells >> > > > > > wrote: >> > > > > > > I'm running into a Python error (below), which I suspect >> > > > > > > is due to the relocation of the UFC Python module (now in >> > > > > > > its own subdirectory). Do I need to add >> > > > > > > >> > > > > > > from ufc import * >> > > > > > > >> > > > > > > somewhere? >> > > > > > > >> > > > > > > Garth >> > > > > > > >> > > > > > > Traceback (most recent call last):
>> > > > > > >   File "demo.py", line 20, in <module>
>> > > > > > > >> > > > > > > from dolfin import * >> > > > > > > >> > > > > > > File >> > > > > > > >> > > > > > > >> > > > > > > >> > > > > > > "/home/garth/Work/FEniCS/lib/python2.6/site-packages/dolfin/ >> > > > > > > __init__. py", line 13, in <module> >> > > > > > > >> > > > > > > import dolfin.cpp as cpp >> > > > > > > >> > > > > > > File >> > > > > > > >> > > > > > > >> > > > > > > >> > > > > > > "/home/garth/Work/FEniCS/lib/python2.6/site-packages/dolfin/ >> > > > > > > cpp.py", line 8902, in <module> >> > > > > > > >> > > > > > > class GenericFunction(ufc.function,Variable): >> > > > > > > AttributeError: 'module' object has no attribute >> > > > > > > 'function' >> > > > > > >> > > > > > Looks like a missing __init__.py in the ufc directory, but I
>> > > > > > thought that was added?
>> > > > > >> > > > > It's present on my machine. >> > > > > >> > > > > >> > > > > Make sure that you clean out any pkg-config and old versions
>> > > > > for UFC.
>> > > > >> > > > It's all installed in one directory which I just delete before
>> > > > testing.
>> > > >> > > You also need to get rid of old ufc.py{c} and _ufc.so sticking
>> > > around.
>> > >> > They are also somewhere under the same install tree and I just >> > delete the entire tree so that shouldn't be an issue. >> > >> > > Does it work now when function is imported? >> > >> > Yes, everything builds fine (just finished on Lucid) but something >> > is wrong which prevents me from running the demos. Will need to >> > debug. >> >> Are you compiling the demos with CMake/make or SCons?
>
>I wasn't sourcing the correct file before.
>
>I'm getting a ton of these now when I compile the Poisson demo:
>
>/home/logg/scratch/lib/fenics-dev/lib/libpytrilinos.so: undefined
>reference to `PyObject_SetAttrString'
>/home/logg/scratch/lib/fenics-dev/lib/libpytrilinos.so: undefined
>reference to `PyType_IsSubtype'
>/home/logg/scratch/lib/fenics-dev/lib/libpytrilinos.so: undefined
>reference to `PyErr_Fetch'
>/home/logg/scratch/lib/fenics-dev/lib/libpytrilinos.so: undefined
>reference to `PyCObject_Import'
>/home/logg/scratch/lib/fenics-dev/lib/libpytrilinos.so: undefined
>reference to `PyExc_RuntimeError'
>
>Any ideas? I have never really used Trilinos before so I'm not
>familiar with how it's setup, but do we really use the Python wrappers
>for Trilinos? Perhaps they can be disabled in the build to get around
>this.

Yes, just disable PyTrilinos. I've always had trouble with it. I think
Sandia need to pay Johan Hake to sort themselves out.

That will be expensive...

I have no problems (now) to import PyTrilinos. It tooks some time to get it going. Is there any way I can export my CMake settings for compiling PyTrilinos?


Have a look in CMakeCache.txt.

It was not that a pleasent experience compiling it and I have forgot what ever options I chose in the cmake-gui...


I like cmake-gui for Trilinos too - it's very civilised. The Trilinos build system has improved out of sight. PETSc next?

Garth

Johan

Garth

>--
>Anders




Follow ups

References