[petsc-dev] MatLUFactor() on MATSEQAIJ with no-inodes

[Jed Brown]

On Tue, 16 Mar 2010 00:00:51 -0300, Lisandro Dalcin <dalcinl at gmail.com> wrote:
> PS: I have a single warning:
> 
> In direct memory block for handle type KEYVAL, 3 handles are still allocated
> 
> This is because of my debug build of MPICH2, as PETSc does not frees
> all communicator keyvals it creates.

These also show up as valgrind warnings under Open MPI (presumably
MPICH2 frees them anyway in MPI_Finalize()).  The issue was discussed at
the end of a petsc-dev thread from November

  Subject: Should PetscObjectName be collective?

They could be resolved with the same approach as

  http://petsc.cs.iit.edu/petsc/petsc-dev/rev/ecbd2dcad71a

Jed