[Pkg-openmpi-maintainers] New Debian package version?

[Manuel Prinz]

Am Mittwoch, den 25.07.2007, 12:06 +0000 schrieb Dirk Eddelbuettel:
> On Wed, Jul 25, 2007 at 01:20:53PM +0200, Manuel Prinz wrote:
> >  * QA: The Debcheck page hasn't been update for ages, but the problem
> > mentioned still remains: libopenmpi1 is optional an depends on
> > libibverbs1 which is extra. The easiest way to fix that is to make
> > OpenMPI extra, too. Looking at the Policy definition of optional and
> > extra, extra fits better to OpenMPI anyway (IMHO).
> 
> Ahh. Yes, switching to extra is the easiest. What do lam, mpich2, ... have?

Both are "extra". (Exception: mpi-doc of MPICH is optional, which
doesn't violate the policy. We may adopt that but I don't see a point in
that.)

I'll change the priority of the source package in debian/control to
"extra" then.
 
> >  * libibverbs: The library is for use of InfiniBand. Our cluster doesn't
> > use (or better: have) InfiniBand, so OpenMPI throws warnings when an MPI
> > job is run. My question is whether we really need to depend on that
> > library or if we can/should provide a seperate package without
> > libibverbs dependancy or document in README.Debian how to turn it off?
> > My experience shows that the warnings irritates MPI users and I'd like
> > to avoid that, if possible.
> 
> Didn't that change? I think 1.2.3 no longer warns, and/or someone told us 
> how to turn the option off via config. I can check if you want.

You may be right here, stupid me didn't realize that he's still running
1.2.2 on the production system. I'll upgrade and check again. Sorry for
the noise!

> Another issue we need to address is a follow-up re the build failures for 
> missing 'atomistic' ops to contact the two or three people who offered help.
> We should take them up on it, but I am overly busy for the the next two weeks 
> or so. Anybody else?

I'm quite busy with two packages I prepare and want to get on mentors.
But I guess I'll have some time for that next week or so.

> We should look at the five open bugs. For arches with no hope of building, 
> we should exclude them in debian/control.

Seems reasonable. I may get some help with sparc during the next days.
Nevertheless, I have neither experience with nor access to other archs
than i386 or amd64, so I'm not quite sure how to deal with that issue.

> It's easy to comment patches in or out with the dpatch system, so go for it.

Done.

Best regards
Manuel