[Python-modules-team] Bug#456869: [PATCH] petsc4py: FTBFS: /usr/lib/petsc/include/petsc.h:138:17: error: mpi.h: No such file or directory

[Ondrej Certik]

Hi Manuel,

> Hello Ondrej,
>
> attached you'll find a patch that solved the FTBFS of your package for
> me. It patches the source directly, so you have to convert it so it can
> be used with your favorite patch system.
>
> The problem is that you can't find the MPI includes, as you already
> stated in #457088. This is an issue because you do not use the mpicc
> compiler wrapper to compile. The best solution would be to add
>
>   DEB_PYTHON_BUILD_ARGS += --compiler=/usr/bin/mpicc
>
> to debian/rules but distutils do not seem to handle that. (Which I think
> is a bug in distutils, mpicc is just a gcc wrapper.) So the patch adds
> the directory where the header files are located to the include search
> path.
>
> I'm not familiar with your package at all, so I'm not quite sure if it
> even uses MPI. But if it does, it would be a good idea to Build-Depend
> on at least one MPI package. Currently, you're not doing this, and
> relying on packages not in Build-Depends is an RC bug. If your package
> does not make use of MPI at all, I'm not sure why had problems in the
> first place. You shouldn't need the patch then.


Thanks very much for it! Yes, this is the patch I did to fix the
package, except that originally I was using /usr/lib/openmpi/inlude/
and
that's why I filled that bug. I patched the package and now I am waiting to find
a sponsor to upload it.

Thanks again for the work you and Dirk are doing on the openmpi
package and especially the quick responses.
And sorry if I made some confusion.

Ondrej