fenics team mailing list archive

[Balbir Thomas <balbir.thomas@xxxxxxxxx>]

Hi,

On Wed, Feb 11, 2009 at 3:26 PM, Johan Hake <hake@xxxxxxxxx> wrote:

>
> > Can you please hand the error message that says that you need swig
> >=1.3.37?


Well the debian/control file (of dolfin) lists it as a hard dependency in
the Build-Depends line "swig (>=1.3.37)" . I am glad to know that the swig
dependency is >= 1.3.35. This would mean that the segmentation fault
problems is not because of swig but because of MPI. I do have both MPICH and
OPENMPI installed. They were installed automatically when I choose some of
the debian science and math "tasks" (packages), meaning expect others to
face the same situation. From what you (Johan) say I suspect that MPICH is
the culprit. Though I have not decided how i am going to confirm this short
of uninstalling mpich and all the packages that depend on it, potentially
breaking my fresh Debian/Lenny install. Any suggestions ? I plan to play
around with different configure options and see how i can isolate the dolfin
build process and subsequent use from seeing mpich. This may just be a bug
in the packaging but could also be a problem with the pkg-config setup
(--cflags etc).

Another potential problem with the Build-Depends line is the fact that it
has "python-all". However on all current deb based distributions
"python-all" includes python2.4 (even on lenny). But Fenics is not
compatible with python2.4 as far as I know since it uses additions to python
since python2.5 such as set functions "any", "all" and lexical elements such
as generators etc. I believe python-all must be replaced by python (>=2.5)
on the build depends line.

>
> > I am not sure how the compatablility is for lenny, but you should be able
> to
> > build most fenics packages using dorsal,


I could use dorsal but shouldn't the debian source package be enough and be
the recomended way to build fenics on a deb based distribution ? After all
if Fenics is going to make it into one of the official Debian (or Debian
based) distributions it will need to build cleanly from the source package.
I maintain a local repository of deb packages on my computer to make it easy
for me to do automated upgrades and rebuilds using Debian's built-in tools,
for packages that are not part of the Debian distributions as yet.

 > <http://www.fenics.org/wiki/Dorsal>
 > It should support Etch and "testing"

>
> The problem with Etch and the source packages were the build dependencies.
For instance libsuitesparse among many others. Despite this I spent a couple
of days installing all the build dependencies on Etch and trying to port
Fenics to python2.4 but finally decided I may find it easier to work with
Debian/Lenny. While in principle it is possible to rewrite the build
dependencies so that builds are successfull on Etch too, I think it would
not be worth the effort given lenny seems to be right around the corner and
the fact that in Etch the recomended python version is 2.4 (but is 2.5 in
lenny). However there are problems with lenny too (in particular the
segfaults). Once again while I appreciate the Fenics developers efforts in
writing dorsal (and I will probably check it out), I would like to still
request them to provide source packages that may cleanly be built by
themselves. Perhaps dorsal can be a wrapper over the Debian/Ubuntu and
RedHat source packages. Is it so ? I havent' checked it out but will do so.
I hope you do see why I urge this case.

> When something happens in a C++ demo swig cannot be blamed.

>
Agreed. I thought so too and so I mentioned the fact.

> It looks like something happens in some MPI code. DOLFIN is confirmed to
work\
> with the OpenMPI that comes with the Ubuntu hardy/intrepid versions.

Agreed. And I believe this is the likely problem too. I will try and dig
deeper (as mentioned above) on this issue but am not sure how far I will
get. It will be great if one of your developers could also try to replicate
these issues on Debian/Lenny, as my familiarity with the Fenics source code
is limited. Kindly do post a patch to this thread if one of the Fenics
developers does find the problem. I will do so to should I ... .

Thanking you
B Thomas