Fix segv in MPI_Graph_create_undef_c Intel test.

When you call MPI_Graph_create with a old_comm of size N, and pass
nnodes=(N=1), then the Nth proc is supposed to get MPI_COMM_NULL out.
The code in this base function didn't properly handle the proc(s) that
are supposed to get MPI_COMM_NULL out.

cmr=v1.7.5:reviewer=hjelmn

This commit was SVN r31145.

ompi/mca/topo/base/topo_base_graph_create.c

@@ -9,6 +9,7 @@
  *                         University of Stuttgart.  All rights reserved.
  * Copyright (c) 2004-2005 The Regents of the University of California.
  *                         All rights reserved.
+ * Copyright (c) 2014 Cisco Systems, Inc.  All rights reserved.
  * $COPYRIGHT$
  * 
  * Additional copyrights may follow
@@ -73,6 +74,15 @@ int mca_topo_base_graph_create(mca_topo_base_module_t *topo,
     graph->index = NULL;
     graph->edges = NULL;
 
+    /* Don't do any of the other initialization if we're not supposed
+       to be part of the new communicator (because nnodes has been
+       reset to 0, making things like index[nnodes-1] be junk).
+
+       JMS: This should really be refactored to use
+       comm_create_group(), because ompi_comm_allocate() still
+       complains about 0-byte mallocs in debug builds for 0-member
+       groups. */
+    if (MPI_UNDEFINED != new_rank) {
         graph->index = (int*)malloc(sizeof(int) * nnodes);
         if (NULL == graph->index) {
             free(graph);
@@ -99,6 +109,7 @@ int mca_topo_base_graph_create(mca_topo_base_module_t *topo,
                 topo_procs[i] = ompi_group_peer_lookup(old_comm->c_local_group,i);
             }
         }
+    }
 
     /* allocate a new communicator */
     new_comm = ompi_comm_allocate(nnodes, 0);