[Debichem-devel] Suggestion for the mdrun_mpi.{lam, mpich, openmpi}.1 man-pages

Nicholas Breen nbreen at ofb.net
Fri Feb 15 19:02:53 UTC 2008


On Fri, Feb 15, 2008 at 11:30:50AM +0100, Daniel Leidert wrote:
> Maybe you saw it. I did a quality review of the man-pages in debichem.
> The last package I was working on was GROMACS. I saw, that the
> mdrun_mpi.lam.1, mdrun_mpi.mpich.1 and mdrun_mpi.openmpi.1 are almost
> identical except one sentence. Further all these man-pages mention the
> generic command mdrun_mpi.1 in their synopsis, ... So I thought, if we
> can maybe ease this part a bit. Only write and maintain one man-page:
> mdrun_mpi.1. This file should contain a general description of this
> command [1] and IMO also mention the 3 different commands.
> 
> The man-pages for mdrun_mpi.{lam,mpich,openmpi}.1 can be "created" by
> simply a) linking them or b) put
> 
> .so man1/mdrun_mpi.1
> 
> into mdrun_mpi.{lam,mpich,openmpi}.1 and run dh_installman.
> 
> What's your opinion?

The reasoning behind the current split is twofold: first, none of the
MPI packages depend on each other nor on gromacs/-doc/-dev (by design;
this makes them suitable for nodes of a computing cluster without
pulling in extraneous dependencies), so there is no common package that
a man page could be placed in.  This makes (a) problematic.  Second,
they're designed to be installable in parallel, prohibiting any
overlapping files - which I believe (b) would fail at.  Within these
constraints, shipping one man page per package (even nearly identical as
it is) seems like the most efficient choice.

> [1] Mayne we should use the alternatives system to allow the usage
> of /usr/bin/mdrun_mpi. Which one should get the highest priority, if you
> agree to this idea?

Already does, check the postinst.  It creates links for the mdrun_mpi.1
man page as well.  LAM gets the highest priority as it's the preferred
MPI implementation in the gromacs community, though as it matures I
expect OpenMPI will eventually replace it.



- Nicholas



More information about the Debichem-devel mailing list