[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: updating mpi-defaults (decommissioning lam)

On Thu, Apr 07, 2011 at 02:58:42PM +1000, Drew Parsons wrote:
> mpi-defaults current makes a number of architectures broken [1] because
> they have the default mpi implementation set to lam (where openmpi is
> not available)
> A patch changing the default mpi from lam to mpich has been stagnating
> in git for more than a year, and missed the release of squeeze.
> I have the will to upload the new mpi-defaults (I'll apply the patch in
> #575259 while I'm at it) and would like to do so.  Please speak up if
> you would prefer me not to take this action.

Let's do it.  I believe the list of remaining issues [1] before we can start
removing LAM and MPICH1 entirely is quite small.  There are two categories:
first, the build failures that Lucas Nussbaum identified last year [2] caused
by including mpich2 in mpi-default-dev:

On Wed, Mar 17, 2010 at 08:33:40AM +0100, Lucas Nussbaum wrote:
> The following 11 packages failed to build:
> apbs looks for mpi.h in the wrong place
> petsc Nonexistent directory: /usr/lib/mpich2 for key with-mpi-dir (?)

Both of these expect /usr/lib/mpich2/{include,lib} to exist.  The other three
MPI implementations all have such directories, either as the actual location of
the corresponding files or symlinks to them.  MPICH2 maintainers: would you
consider adding these directories as well?

> blacs-mpi missing target 'build-mpich2' in debian/rules

Bug submitted, #626336.  Looks easy to fix, but I wanted to check some of the
details with the maintainer before writing a patch.

> gdcm ?

#624902, already fixed.  Not MPI related.

> igstk No rule to make target `/usr/lib/libmpi_cxx.so' (?)
> kwwidgets No rule to make target `/usr/lib/libmpi_cxx.so'

I can't figure out why either of these need MPI in the first place.  Neither of
them build any binary packages now that depend on any MPI implementation, and
nothing in their source references MPI.

> life matrixpetsc.hpp:49: error: using typedef-name 'MPI_Win' after 'struct'

Has been removed from unstable, #622138.

> mumps ld: cannot find -lblacs-mpich2

Depends on blacs-mpi, above.

> pgapack cd: 4: can't cd to lib/linux/

Bug with patch submitted as #626330.

> rmpi configure: error: "Cannot find mpi.h header file"

Needs to specify a directory in its configure step.  No bug filed yet, I can
write a quick patch depending on the /usr/lib/mpich2 outcome.

> scalapack missing target 'build-mpich2' in debian/rules

Will be the same fix as blacs-mpi.

The second category are packages that still Build-Depend directly on LAM or
MPICH1, part of the list at [1].  There are only six of them left!  I've
submitted patches for five; the last, xmpi (#571448), is part of LAM/MPI itself
and cannot have its B-Ds changed.  They should be removed from the archive
together.  One of others, python-scientific, is orphaned; IANADD, so could
someone please take a look at the patch [3] and make a QA upload?

Let's not let another release cycle slip away!

Nicholas Breen

[1] http://bugs.debian.org/cgi-bin/pkgreport.cgi?tag=old-mpi-eol;users=debian-science@lists.debian.org
[2] http://lists.debian.org/debian-science/2010/03/msg00021.html
[3] http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=571452#10

Reply to: