Update the TM launcher so it provides an appropriate error message when encountering an invalid launch_id. This is a first step towards fixing ticket #1016, but needs to be followed by a more complete solution.

This commit was SVN r14578.

orte/mca/pls/tm/help-pls-tm.txt

@@ -16,6 +16,22 @@
 # 
 # $HEADER$
 #
+[tm-bad-launchid]
+The TM (PBS / Torque) process starter cannot spawn the specified
+application on a remote node due to an invalid launch_id.
+
+Node name: %s
+Launch id: %d
+
+This is most likely due to use of the "--hostfile" option to the
+command line. At this time, Open MPI/OpenRTE do not support this
+method of operation. Instead, the system expects to directly read
+information regarding the nodes to be used from the environment.
+
+Removing "--hostfile" from the command line will likely allow the
+application to be launched. This will be fixed in a future release
+to support the use of "--hostfile" on the command line.
+#
 [multiple-prefixes]
 Multiple different --prefix options were specified to mpirun for the
 same node.  This is a fatal error for the TM (PBS / Torque) process

orte/mca/pls/tm/pls_tm_module.c

@@ -164,6 +164,22 @@ static int pls_tm_launch_job(orte_jobid_t jobid)
         goto cleanup;
     }
 
+    /* Iterate through each of the nodes and check to see if we have
+     * a valid launch_id (must be > 0). If not, then error out as
+     * we cannot do anything
+     */
+    for (item =  opal_list_get_first(&map->nodes);
+         item != opal_list_get_end(&map->nodes);
+         item =  opal_list_get_next(item)) {
+        orte_mapped_node_t* node = (orte_mapped_node_t*)item;
+        
+        if (node->launch_id < 0) {
+            opal_show_help("help-pls-tm.txt", "tm-bad-launchid",
+                           true, node->nodename, node->launch_id);
+            goto cleanup;
+        }
+    }    
+        
     /* if the user requested that we re-use daemons,
      * launch the procs on any existing, re-usable daemons
      */