Bugzilla Bug 210639

sci-visualization/paraview-3.2.1  with USE="doc hdf5 mpi* python qt4 threads
-examples" failed to build, see log file.

I think the real error occur here:

Linking CXX shared library ../../../bin/libXdmf.so
/usr/lib/gcc/x86_64-pc-linux-gnu/4.1.2/../../../../x86_64-pc-linux-gnu/bin/ld:
/usr/lib64/libmpich.a(send.o): relocation R_X86_64_32 against `a local symbol'
can not be used when making a shared object; recompile with -fPIC
/usr/lib64/libmpich.a: could not read symbols: Bad value
collect2: ld returned 1 exit status
make[2]: *** [bin/libXdmf.so] Error 1
make[1]: *** [Utilities/Xdmf2/libsrc/CMakeFiles/Xdmf.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs....

It seems to be a problem with the mpich lib.


Reproducible: Always

Created an attachment (id=143918) [details]
The log file

So what's up with the missing shared library for sys-cluster/mpich? Ditto Bug
176849 and Bug 202684?

sci-visualization/paraview depends sys-cluster/mpich and not on virtual/mpi nor
|| (sys-cluster/mpich sys-cluster/mpich2). And sys-cluster/mpich doesn't
install a libmich.so and the libmich.a isn't build with -fPIC compiler flag.
And to do USE=-mpi is not an option.

And in detail, I don't know why there is no libmich.so build.

Personally, I much rather try enabling support for openmpi
in paraview (if possible) instead of spending time fixing the now
almost 3 year old mpich. Would this be an acceptable
option for you? I've done this in the past for vtk and lam-mpi
which is now deprecated in favour of openmpi.

Thanks,
Markus

If openmpi will work, that would be the perfect option!

Oliver

(In reply to comment #4)
> Personally, I much rather try enabling support for openmpi
> in paraview (if possible) instead of spending time fixing the now
> almost 3 year old mpich.

Sounds great, ditto for the other two packages mentioned in Comment #2 ;)

I just committed a patch that should allow paraview
to compile against openmpi and mpich2.

Please let me know if this works for you as well.

Thanks,
Markus

Thanks for that patch. Now I can compile against the openmpi lib. And with the
mpi useflag the volume rendering works! So I will close this bug.

Oliver

Great and thanks for testing!
I'll apply the same set of patches to vtk as well
then.

Markus

Just a small question why not changing the dep to virtual/mpi.

As long as mpich is in the tree it could be used even if it is old and
outdated. As far as i know paraview should work with every mpi implementation
which is currently in gentoo. This should apply to vtk too, at least for
mpich2, openmpi and lam-mpi according to the ebuild.

I think this is a conflict with the virtual itself as it is listing mpich as
the first value so the old and outdated implementation will be used first if
none is present! Why not use something like this in the virtual:

RDEPEND="|| (
                sys-cluster/openmpi
                sys-cluster/mpich2
                sys-cluster/lam-mpi
                sys-cluster/mpich
        )"

(In reply to comment #10)
> Just a small question why not changing the dep to virtual/mpi.
> 
> As long as mpich is in the tree it could be used even if it is old and
> outdated. As far as i know paraview should work with every mpi implementation
> which is currently in gentoo. This should apply to vtk too, at least for
> mpich2, openmpi and lam-mpi according to the ebuild.

Unfortunately, this is not the case. mpich only provides static libs and
hence paraview can't link against it on any arch that needs pic code such
as amd64.

> 
> I think this is a conflict with the virtual itself as it is listing mpich as
> the first value so the old and outdated implementation will be used first if
> none is present! Why not use something like this in the virtual:
> 
> RDEPEND="|| (
>                 sys-cluster/openmpi
>                 sys-cluster/mpich2
>                 sys-cluster/lam-mpi
>                 sys-cluster/mpich
>         )"
> 

This is currently in the works by the cluster herd but requires that
a stable version of openmpi is available on most arches, which takes
some time.

Best,
Markus