Some minor cleanups. Get singletons working. Cleanup abort handling so it gets properly identified.

This commit was SVN r26261.

orte/mca/ess/base/ess_base_std_app.c

@@ -390,6 +390,10 @@ void orte_ess_base_app_abort(int status, bool report)
                                      ORTE_RML_NON_PERSISTENT, report_sync, NULL)) {
             return;
         }
+        while (!sync_recvd) {
+            opal_progress();
+        }
+        return;
     }
     
     /* - Clean out the global structures 

orte/mca/ess/singleton/ess_singleton_module.c

@@ -242,7 +242,12 @@ static int rte_init(void)
         ORTE_ERROR_LOG(rc);
         return rc;
     }
-        
+    
+    /* set the collective ids */
+    orte_process_info.peer_modex = 0;
+    orte_process_info.peer_init_barrier = 1;
+    orte_process_info.peer_fini_barrier = 2;
+
     return ORTE_SUCCESS;
 }
 

orte/mca/odls/base/odls_base_default_fns.c

@@ -1889,7 +1889,11 @@ void orte_odls_base_default_report_abort(orte_process_name_t *proc)
     orte_proc_t *child;
     opal_buffer_t *buffer;
     int rc, i;
-    orte_ns_cmp_bitmask_t mask;
+
+        OPAL_OUTPUT_VERBOSE((5, orte_odls_globals.output,
+                             "%s GOT ABORT REPORT FOR %s",
+                             ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
+                             ORTE_NAME_PRINT(proc)));
 
     /* find this child */
     for (i=0; i < orte_local_children->size; i++) {
@@ -1897,11 +1901,9 @@ void orte_odls_base_default_report_abort(orte_process_name_t *proc)
             continue;
         }
 
-        mask = ORTE_NS_CMP_ALL;
-        
-        if (OPAL_EQUAL ==
-                orte_util_compare_name_fields(mask, proc, &child->name)) { /* found it */
-            child->state = ORTE_PROC_STATE_CALLED_ABORT;
+        if (proc->jobid == child->name.jobid &&
+            proc->vpid == child->name.vpid) { /* found it */
+            child->aborted = true;
             /* send ack */
             buffer = OBJ_NEW(opal_buffer_t);
             if (0 > (rc = orte_rml.send_buffer_nb(proc, buffer,
@@ -1993,7 +1995,7 @@ void odls_base_default_wait_local_proc(pid_t pid, int status, void* cbdata)
         /* set the exit status appropriately */
         proc->exit_code = WEXITSTATUS(status);
 
-        if (ORTE_PROC_STATE_CALLED_ABORT == proc->state) {
+        if (proc->aborted) {
             /* even though the process exited "normally", it happened
              * via an orte_abort call, so we need to indicate this was
              * an "abnormal" termination.

orte/orted/orted_main.c

@@ -72,6 +72,7 @@
 #include "orte/mca/errmgr/errmgr.h"
 #include "orte/mca/ess/ess.h"
 #include "orte/mca/grpcomm/grpcomm.h"
+#include "orte/mca/grpcomm/base/base.h"
 #include "orte/mca/rml/rml.h"
 #include "orte/mca/rml/rml_types.h"
 #include "orte/mca/odls/odls.h"
@@ -510,7 +511,7 @@ int orte_daemon(int argc, char *argv[])
         proc = OBJ_NEW(orte_proc_t);
         proc->name.jobid = jdata->jobid;
         proc->name.vpid = 0;
-
+        proc->alive = true;
         proc->state = ORTE_PROC_STATE_RUNNING;
         proc->app_idx = 0;
         /* obviously, they are on my node */
@@ -519,8 +520,16 @@ int orte_daemon(int argc, char *argv[])
         OBJ_RETAIN(node);  /* keep accounting straight */
         opal_pointer_array_add(jdata->procs, proc);
         jdata->num_procs = 1;
+        /* and obviously they are one of my local procs */
+        OBJ_RETAIN(proc);
+        opal_pointer_array_add(orte_local_children, proc);
         jdata->num_local_procs = 1;
         
+        /* the singleton will use the first three collectives
+         * for its modex/barriers
+         */
+        orte_grpcomm_base.coll_id += 3;
+
         /* need to setup a pidmap for it */
         buffer = OBJ_NEW(opal_buffer_t);
         opal_dss.pack(buffer, &jdata->jobid, 1, ORTE_JOBID); /* jobid */
@@ -603,7 +612,7 @@ int orte_daemon(int argc, char *argv[])
                        orted_globals.singleton_died_pipe,
                        OPAL_EV_READ,
                        pipe_closed,
-                       &orted_globals.singleton_died_pipe);
+                       pipe_handler);
         opal_event_add(pipe_handler, NULL);
     }
 

orte/runtime/orte_globals.c

@@ -845,6 +845,7 @@ static void orte_proc_construct(orte_proc_t* proc)
     proc->last_errmgr_state = ORTE_PROC_STATE_UNDEF;
     proc->state = ORTE_PROC_STATE_UNDEF;
     proc->alive = false;
+    proc->aborted = false;
     proc->app_idx = 0;
 #if OPAL_HAVE_HWLOC
     proc->locale = NULL;

orte/runtime/orte_globals.h

@@ -442,6 +442,8 @@ struct orte_proc_t {
      * and has not yet terminated
      */
     bool alive;
+    /* flag if it called abort */
+    bool aborted;
     /* exit code */
     orte_exit_code_t exit_code;
     /* the app_context that generated this proc */

orte/tools/orterun/orterun.c

@@ -935,9 +935,6 @@ int orterun(int argc, char *argv[])
     }
 
  DONE:
-    /* update the exit status, in case it wasn't done */
-    ORTE_UPDATE_EXIT_STATUS(orte_exit_status);
-
     /* cleanup and leave */
     orte_finalize();