OpenMPI, Emacs and libotf problems

Adam Williamson awilliam at redhat.com
Wed Jul 6 20:42:36 UTC 2011


On Wed, 2011-07-06 at 23:39 +0300, Jussi Lehtola wrote:
> On Wed, 06 Jul 2011 13:09:41 -0700
> Adam Williamson <awilliam at redhat.com> wrote:
> > > .. but on the other hand, the same logic applies in the opposite
> > > sense: if something requires OpenMPI's libotf.so.0, also the
> > > truetype libotf will satisfy the requirement. (Although openmpi
> > > apps typically link to a half a dozen other openmpi libs as well).
> > 
> > Nothing really could require OpenMPI's libotf as things stand, because
> > of what I wrote above: nothing can find it unless it uses a custom
> > linker path. If OpenMPI actually wanted the library to be something
> > other packages can use, it should really install it in a shared path
> > (and, as we've already discussed, rename it). If we're just talking
> > about different OpenMPI packages, they can handle the
> > intra-dependencies manually, I'd say.
> 
> Not really, since when the MPI environment is loaded the relevant
> library paths are added to LD_LIBRARY_PATH.
> 
> They're not installed in system locations, since e.g. all MPI libraries
> ship with libmpi.so, and there are many variants: OpenMPI, MPICH2,
> MVAPICH, and so on.

In that case it may be best just to get the library renamed, I guess.
-- 
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Fedora Talk: adamwill AT fedoraproject DOT org
http://www.happyassassin.net



More information about the devel mailing list