[openfabrics-ewg] OFED-1.1-rc2 review comments: openib.spec

[Doug Ledford]

On Mon, 2006-08-28 at 08:38 -0400, Jeff Squyres wrote:
> On 8/27/06 2:18 PM, "Doug Ledford" <dledford at redhat.com> wrote:
> 
> >> This *may* be possible for Open MPI, but it'll require an audit.  I have a
> >> dim recollection of one of our developers saying a long time ago that this
> >> would be quite difficult for OMPI for some obscure reason...
> > 
> > Nah, wasn't hard.  The OpenMPI binaries are actually mostly links to
> > orte* binaries and ortewrapper.  So, just remove the links prior to
> > packaging.  It doesn't conflict on the main binary names much at all.
> 
> There are a few issues here:
> 
> - While the real names of the executables are "orte_*" or "opal_*", users
> will expect "mpi*".  So we need those sym links.

That's what alternatives is for.

>   Auditing the
> creation/removal of those sym links is what I had in mind to ensure that we
> did the prefix stuff properly.

When I said that you had to coordinate between openmpi and lam (and
possible mvapich) in their spec files to make sure that they have the
same master/slave relationships on links and the same link nick-names,
that is in essence your audit.  No one should ever use orte* or
opalwrapper or lamexec or lamrun or lamcc, they use mpirun or mpiexec or
mpicc and based upon which alternative is currently in place, those
links will point to the right binaries plus there will be a link
in /etc/ld.so.conf.d/mpi.conf that points the the right libdir to find
libmpi plus there will be links to the right mpi* man pages.

Hmm...that just made me realize that if the user manually selects a
different mpi then we'll need to rerun ldconfig, but there's no way to
trigger that via the alternatives commands.  Will have to document that.

> - Not only do users expect those filenames, part of our code actually looks
> at filename(argv[0]) to determine its behavior (lightbulb -- this might be
> what our developer was referring to long ago when he said that prefix stuff
> would be problematic).

Using alternatives, the called program name will still be mpi*.

> > I'm syncing my latest user space rpms to web page right now so you can
> > take a look whenever you want.
> 
> Ok.  That's on your http://people.redhat.com/dledford/Infiniband/ page?

Yep.  I currently have fc6/rhel5 rpms up there, but can put rhel4 rpms
up there if people need them.

> All the stuff mentioned above is on my to-do list for this upcoming week.
> 
-- 
Doug Ledford <dledford at redhat.com>
              GPG KeyID: CFBFF194
              http://people.redhat.com/dledford

Infiniband specific RPMs available at
              http://people.redhat.com/dledford/Infiniband
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://lists.openfabrics.org/pipermail/ewg/attachments/20060828/f862fb8b/attachment.sig>