[Pkg-openmpi-maintainers] Bug#457088: Bug#457088: mpi.h is missing

[Dirk Eddelbuettel]

On Wed, Dec 19, 2007 at 07:40:13PM +0100, Ondrej Certik wrote:
> On Dec 19, 2007 7:15 PM, Dirk Eddelbuettel <edd at debian.org> wrote:
> > On Wed, Dec 19, 2007 at 06:49:15PM +0100, Ondrej Certik wrote:
> > > Package: libopenmpi-dev
> > > Version: 1.2.4-5
> > > Severity: serious
> > >
> > > --- Please enter the report below this line. ---
> > >
> > > Hi,
> > >
> > > the mpi.h is missing. I found it in
> > >
> > > /usr/lib/openmpi/include/
> > >
> > > please add at least a symlink or something to
> > >
> > > /usr/include
> > >
> > > the is breaks the python-petsc4py package for example:
> > >
> > > http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=456869
> > >
> > > It used to work before. I could fix it by patching the build scripts to include /usr/lib/openmpi/include/, but
> > > the right solution imho is to fix libopenmpi-dev.
> >
> > Nope, as we as package maintainers simply changed the location.  Study
> > the  configure options you have in your client (source) code.
> >
> > For what it's worth, I just made the same change in Rmpi (being its
> > maintainer).
> >
> > mpi.h not being in /usr/include is not a bug. and certainly not a
> > serious bug.  Please show some more restraint.
> >
> > You can always ask us off-line, as you have seen in the last few days
> > we tend to respond quickly.
> 
> I am sorry I put the severity wrong. But you, as the maintainer can
> adjust it as you see fit.
> 
> As to the real problem - I just asked about exactly this on debian-python:
> 
> http://lists.debian.org/debian-python/2007/12/msg00030.html
> 
> and I think include files belong to /usr/include, accordig to the FHS:
> 
> http://www.debian.org/doc/packaging-manuals/fhs/fhs-2.3.html#USRINCLUDEDIRECTORYFORSTANDARDINCLU
> 
> So it's imho a bug and should be reopened. (I know how to fix
> petsc4py, I just tried and confirmed it works, but I think it's a bug
> that should be fixed in openmpi, not petsc4py). Why do you think the
> header files shouldn't be in /usr/include?

As you may know, we are trying to let Open MPI coexist with MPICH and
LAM.  

/usr/include is fine for _non-conflicting_ files as good as a general
rule, but not so clear if several packages provide it.  By 'staying
out of each others way' it is easier to build against several MPI
implementations.

So no bug here.

Thanks, Dirk

-- 
Three out of two people have difficulties with fractions.