Re: [Libmesh-users] LibMeshInit and MPI calls

[Roy S. <roy...@ic...>]

On Wed, 19 Mar 2008, David Knezevic wrote:

> I've just updated to the SVN head, and saw that libMesh::init is now
> deprecated. So I changed to use LibMeshInit, but I get a problem when I
> make MPI calls. For example, this runs fine:
>
> int main (int argc, char** argv)
> {
>  libMesh::init(argc,argv);
>  {
>
>    PetscErrorCode ierr;
>    PetscMPIInt    num_procs;
>    ierr = MPI_Comm_size(PETSC_COMM_WORLD,&num_procs);CHKERRQ(ierr);
>
>  }
>  return libMesh::close();
> }
>
> But changing to:
>
> int main (int argc, char** argv)
> {
>  LibMeshInit(argc,argv);
>
>  PetscErrorCode ierr;
>  PetscMPIInt    num_procs;
>  ierr = MPI_Comm_size(PETSC_COMM_WORLD,&num_procs);CHKERRQ(ierr);
>
> }
>
> I get the error:
> "Attempting to use an MPI routine after finalizing MPICH".
>
> I haven't spotted what's causing the problem, help would be appreciated.

Instead of LibMeshInit(argc, argv), try LibMeshInit init (argc, argv);

Your way, a temporary LibMeshInit object is being created (which
initializes MPI) but immediately destroyed (which finalizes MPI).  The
LibMeshInit object needs to last as long as your use of MPI/PETSc/
libMesh does.

Don't feel bad; I originally made the same mistake when converting the
examples.  I really like the C++ "constructors acquire resources,
destructors delete them" idiom, but it would really help if there was
a way to catch this sort of typo at compile time instead of run time.
---
Roy