For cases where the alpha+non-zero prefix must be removed from a node name, be sure to do it everywhere we access node names - otherwise, modex methods such as pmi will fail to correctly identify procs on the same node

This commit was SVN r27022.

contrib/platform/lanl/cray_xe6/debug-nopanasas.conf

@@ -81,7 +81,7 @@ orte_tmpdir_base = /var/tmp
 orte_allocation_required = 1
 
 ## Deal with the allocator
-plm_base_strip_prefix_from_node_names = 1
+orte_strip_prefix_from_node_names = 1
 
 ## MPI behavior
 ## Do NOT specify mpi_leave_pinned so system

contrib/platform/lanl/cray_xe6/optimized-nopanasas.conf

@@ -81,7 +81,7 @@ orte_tmpdir_base = /var/tmp
 orte_allocation_required = 1
 
 ## Deal with the allocator
-plm_base_strip_prefix_from_node_names = 1
+orte_strip_prefix_from_node_names = 1
 
 ## MPI behavior
 ## Do NOT specify mpi_leave_pinned so system

orte/mca/plm/base/plm_base_launch_support.c

@@ -620,7 +620,7 @@ void orte_plm_base_daemon_callback(int status, orte_process_name_t* sender,
                     nodename = ptr;
                 }
             }
-            if (orte_plm_globals.strip_prefix_from_node_names) {
+            if (orte_process_info.strip_prefix_from_node_names) {
                 /* remove all leading characters and zeroes */
                 ptr = nodename;
                 while (idx < (int)strlen(nodename) &&

orte/mca/plm/base/plm_base_open.c

@@ -89,8 +89,6 @@ orte_plm_base_module_t orte_plm = {
  */
 int orte_plm_base_open(void)
 {
-    int value;
-
     /* Debugging / verbose output.  Always have stream open, with
        verbose set by the mca open system... */
     orte_plm_globals.output = opal_output_open(NULL);
@@ -104,11 +102,6 @@ int orte_plm_base_open(void)
     /* default to assigning daemons to nodes at launch */
     orte_plm_globals.daemon_nodes_assigned_at_launch = true;
 
-    mca_base_param_reg_int_name("plm", "base_strip_prefix_from_node_names",
-                                "Whether to strip leading characters and zeroes from node names returned by daemons",
-                                false, false, (int)false, &value);
-    orte_plm_globals.strip_prefix_from_node_names = OPAL_INT_TO_BOOL(value);
-
     /* Open up all the components that we can find */
 
     if (ORTE_SUCCESS != 

orte/mca/plm/base/plm_private.h

@@ -58,8 +58,6 @@ typedef struct {
     opal_buffer_t tree_spawn_cmd;
     /* daemon nodes assigned at launch */
     bool daemon_nodes_assigned_at_launch;
-    /* handle allocator-to-actual nodename matches */
-    bool strip_prefix_from_node_names;
 } orte_plm_globals_t;
 /**
  * Global instance of PLM framework data

orte/util/proc_info.c

@@ -31,6 +31,7 @@
 #ifdef HAVE_SYS_TYPES_H
 #include <sys/types.h>
 #endif
+#include <ctype.h>
 
 #include "opal/mca/base/base.h"
 #include "opal/mca/base/mca_base_param.h"
@@ -76,7 +77,8 @@ ORTE_DECLSPEC orte_proc_info_t orte_process_info = {
     /*  .app_rank =                     */   -1,
     /*  .peer_modex =                   */   -1,
     /*  .peer_init_barrier =            */   -1,
-    /*  .peer_fini_barrier =    */   -1
+    /*  .peer_fini_barrier =            */   -1,
+    /*  .strip_prefix_from_node_names = */ false
 };
 
 static bool init=false;
@@ -84,7 +86,7 @@ static bool init=false;
 int orte_proc_info(void)
 {
     
-    int tmp;
+    int tmp, idx;
     char *uri, *ptr;
     char hostname[ORTE_MAX_HOSTNAME_SIZE];
     
@@ -140,9 +142,33 @@ int orte_proc_info(void)
     /* get the process id */
     orte_process_info.pid = getpid();
 
+    mca_base_param_reg_int_name("orte", "strip_prefix_from_node_names",
+                                "Whether to strip leading characters and zeroes from node names returned by daemons",
+                                false, false, (int)false, &tmp);
+    orte_process_info.strip_prefix_from_node_names = OPAL_INT_TO_BOOL(tmp);
+
     /* get the nodename */
     gethostname(hostname, ORTE_MAX_HOSTNAME_SIZE);
+    /* we have to strip node names here, if user directs, to ensure that
+     * the names exchanged in the modex match the names found locally
+     */
+    if (orte_process_info.strip_prefix_from_node_names) {
+        /* remove all leading characters and zeroes */
+        idx = 0;
+        while (idx < (int)strlen(hostname) &&
+               (hostname[idx] <= '0' || '9' < hostname[idx])) {
+            idx++;
+        }
+        if ((int)strlen(hostname) <= idx) {
+            /* there were no non-zero numbers in the name */
             orte_process_info.nodename = strdup(hostname);
+        } else {
+            orte_process_info.nodename = strdup(&hostname[idx]);
+        }
+    } else {
+        orte_process_info.nodename = strdup(hostname);
+    }
+    opal_output(0, "HOSTNAME: %s", orte_process_info.nodename);
 
     /* get the number of nodes in the job */
     mca_base_param_reg_int_name("orte", "num_nodes",

orte/util/proc_info.h

@@ -127,6 +127,7 @@ struct orte_proc_info_t {
     orte_grpcomm_coll_id_t peer_modex;   /**< modex collective id */
     orte_grpcomm_coll_id_t peer_init_barrier;   /**< barrier id during init */
     orte_grpcomm_coll_id_t peer_fini_barrier;   /**< barrier id during finalize */
+    bool strip_prefix_from_node_names;
 };
 typedef struct orte_proc_info_t orte_proc_info_t;