Completing the MPICH transition

Anton Gladky gladk at debian.org
Tue Oct 29 13:41:35 UTC 2013


Hi Sébastien,

thanks for heads-up. Both mpich versions can coexist, but we should
really request
binNMU to finish the transition and probably remove mpich2 before Jessie release
(CC-ing original mpich-maintainer).

Best regards,

Anton


2013/10/29 Sébastien Villemot <sebastien at debian.org>:
> Hi,
>
> In trying to fix the FTBFS of petsc on those arches where MPICH is the
> default MPI implementation, I realized that the problem is due to the
> MPICH transition. As you know, version 3 of MPICH has been uploaded as
> src:mpich. It took over the libmpich2-dev package from src:mpich2, and
> made it a transitional package. And the new shared library is called
> libmpich10, while the former was libmpich2-3.
>
> My understanding is that this transition is in an unfinished state.
>
> First, I had to fix mpi-defaults (in version 1.0.2), since it was still
> returning "mpich2" on those archs where MPICH is the default.
>
> Also, there are a lot of packages still depending on libmpich2-3,
> including src:scalapack and src:blacs-mpi on which src:petsc
> build-depends. Unless versions 2 and 3 of MPICH can coexist in the same
> application (which I doubt, but please correct me), the current
> situation makes it impossible to fix petsc (and probably quite a few
> other packages).
>
> Anyways, shouldn't we request a proper transition slot from the Release
> Team? That would involve binNMU-ing all packages still depending on
> libmpich2-3. And then removing src:mpich2.
>
> Am I missing something? Should I request the transition slot?
>
> Cheers,
>
> --
>  .''`.    Sébastien Villemot
> : :' :    Debian Developer
> `. `'     http://www.dynare.org/sebastien
>   `-      GPG Key: 4096R/381A7594
>
>
> --
> debian-science-maintainers mailing list
> debian-science-maintainers at lists.alioth.debian.org
> http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/debian-science-maintainers



More information about the debian-science-maintainers mailing list