On Thu, 2008-03-06 at 09:19 -0600, Dirk Eddelbuettel wrote: > On 6 March 2008 at 09:28, Adam C Powell IV wrote: > | On Wed, 2008-03-05 at 23:29 +0100, Thomas Girard wrote: > | > Hello Adam, > | > > | > Le mercredi 05 mars 2008 à 07:34 -0500, Adam C Powell IV a écrit : > | > > I'm not getting this error... I don't know why you would have seen that > | > > MPI error, I'd patched openmpi before but upgrading to 1.2.5-2 undid my > | > > patch and it works for me. I reapplied my patch to hdf5 1.6.5-5.2 > | > > (minus the changelog bit) and retried, and didn't see your error; > | > > | > Okay, got it: /usr/lib/libmpi++.so is an aternative registered with > | > liblam++.so (which was needed for hdf5 compilation). Maybe salome should > | > build-conflict with lam4-dev? > | > > | > Hmmm... removing lam-dev makes /usr/lib/libmpi++.so an alternative > | > symlink to libpmpich++.so.1.0, and not to libmpi_cxx.so.0.0.0. It seems > | > only libmpi_cxx.so works: I have forced the symlink and the link error > | > disappeared. > | > | Ah right. "update-alternatives --config mpi" should let you select > | OpenMPI for all of those links. But to be on the safe side, I've added > | Build-Conflicts against lam4-dev and all of the mpich-dev variants. > > The Open MPI package recently fixed an oversight with respect to the C++ > library. Could you try the most recent one, ie 1.2.5-2, and see if that > works "out of the box" ? "It should", we think we now have most things > ironed out with respect to the stand C library.
Thanks Dirk; I think the issue was that with other MPI implementations also installed, the alternatives symlinks pointed to them, but my package tells Salomé to look in /usr/lib/openmpi/include for headers. So it was getting mpi.h from OpenMPI and failing to link with lam and mpich. Not a problem in general. We're both using 1.2.5-2 in unstable. -Adam -- GPG fingerprint: D54D 1AEE B11C CE9B A02B C5DD 526F 01E8 564E E4B6 Engineering consulting with open source tools http://www.opennovation.com/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
