Fix for bug Josh found -- if we kill as a result of the

OMPI_ERR_INIT_FINALIZE macro (meaning that MPI_INIT has not yet been
invoked or MPI_FINALIZE has already been invoked), then the
communicator argument that is passed to ompi_mpi_abort() is
meaningless and should not be consulted.

This commit was SVN r5313.

src/mpi/runtime/ompi_mpi_abort.c

@@ -76,17 +76,26 @@ ompi_mpi_abort(struct ompi_communicator_t* comm,
         return ret;
     }
 
-    /* kill everyone in the remote group execpt our jobid, if requested */
+    /* Corner case: if we're being called as a result of the
+       OMPI_ERR_INIT_FINALIZE macro (meaning that this is before
+       MPI_INIT or after MPI_FINALIZE), then comm is not setup yet /
+       has already been cleaned up. */
+
+    if (ompi_mpi_initialized && !ompi_mpi_finalized) {
+
+        /* kill everyone in the remote group execpt our jobid, if
+           requested */
         if (kill_remote_of_intercomm && OMPI_COMM_IS_INTER(comm)) {
             abort_procs(comm->c_remote_group->grp_proc_pointers,
                         comm->c_remote_group->grp_proc_count,
                         my_jobid);
         }
 
-    /* kill everyone in the local group, except our jobid */
+        /* kill everyone in the local group, except our jobid. */
         abort_procs(comm->c_local_group->grp_proc_pointers,
                     comm->c_local_group->grp_proc_count,
                     my_jobid);
+    }
 
 
     ret = orte_rmgr.terminate_job(my_jobid);