Hi Alexander! Am Dienstag, den 05.08.2008, 17:43 +0200 schrieb Alexander Achenbach: > The openmpi-bin package adds an alternative 'mpiexec' that controls > /usr/bin/mpiexec and its manpage. > > However, another MPI implementation (lam-runtime) adds a single > alternative class 'mpirun' that includes the 'mpiexec' command and > its manpage as a slave. > > This causes alternatives resolution to be order dependent: Even > though lam-runtime adds the slave 'mpiexec' along with 'mpirun' > at priority 30, 'mpiexec' may get overridden if openmpi-bin gets > installed (or configured) after it.
I remember that we had a similar problem before. > (On etch, a similar problem exists, except that 'mpiexec' also > includes the 'mpirun' command, extending the problem to its scope.) > > What about merging openmpi-bin's 'mpiexec' alternative as a slave > into 'mpirun'? This may be an option. I think we did not do that because MPICH provides different alternatives for both executables (mpirun and mpiexec) but can't verify right now. I think this needs thorough checking and maybe we need to get in touch with the maintainers of the other MPI implementations. update-alternatives only works as expected if everyone plays by the same rules. Best regards Manuel
signature.asc
Description: Dies ist ein digital signierter Nachrichtenteil
