Correct the cutoff point for full modex operation as it is based on the number of nodes in the system, not the number of procs in the signature.

This commit was SVN r32666.

orte/orted/pmix/pmix_server_sendrecv.c

@@ -752,7 +752,7 @@ static void process_message(pmix_server_peer_t *peer)
         /* if we are in a group collective mode, then we need to prep
          * the data as it should be included in the modex */
         OBJ_CONSTRUCT(&save, opal_buffer_t);
-        if (sig->sz < orte_full_modex_cutoff) {
+        if (orte_process_info.num_procs < orte_full_modex_cutoff) {
             /* need to include the id of the sender for later unpacking */
             opal_dss.pack(&save, &id, 1, OPAL_UINT64);
             opal_dss.copy_payload(&save, &xfer);
@@ -1019,7 +1019,7 @@ static void process_message(pmix_server_peer_t *peer)
         }
         OBJ_RELEASE(sig);
         /* include any data that is to be globally shared */
-        if (found) {
+        if (found && 0 < save.bytes_used) {
             bptr = &save;
             if (OPAL_SUCCESS != (rc = opal_dss.pack(reply, &bptr, 1, OPAL_BUFFER))) {
                 ORTE_ERROR_LOG(rc);