[petsc-dev] problems with master and MPICH version 1

[Jed Brown]

Barry Smith <bsmith at mcs.anl.gov> writes:

>   We are having some problems with master and MPICH 1 in the nightly tests
>
> http://ftp.mcs.anl.gov/pub/petsc/nightlylogs/archive/2015/06/08/examples_master_arch-linux-mpich1_steamroller.log
>
> I've done a lot of debugging including with valgrind and cannot
> determine the problem. I'm concluding that it is a problem with how
> they handle attributes since changing the number of attributes can
> produce crashes or prevent crashes. It is flaky, like memory
> corruption problems but valgrind is happy.

MPICH uses integer tables instead of pointers, hiding a lot of
information from Valgrind.  When run in a debugger or Valgrind, what is
the trace when SEGV is raised?  (I don't currently have an MPICH1
built.)

It may well be an MPICH1 bug, but there is definitely no support at all
for MPICH1 and hopefully people have stopped using it long ago.  If we
do decide to end MPICH1 support, we can merge 'jed/mpi-2' (not
necessarily for this release).
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 818 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20150608/fb7572e9/attachment.sig>