Add a bunch of debug to help track down the problem, and eventually find another place where comparison of signatures was incorrectly performed - use the dss compare operation to be consistent and safe

This commit was SVN r32620.

orte/mca/grpcomm/base/grpcomm_base_stubs.c

@@ -197,11 +197,7 @@ orte_grpcomm_coll_t* orte_grpcomm_base_get_tracker(orte_grpcomm_signature_t *sig
             /* if only one is NULL, then we can't possibly match */
             break;
         }
-        /* if the size doesn't match, then they can't be the same */
+        if (OPAL_EQUAL == opal_dss.compare(sig, coll->sig, ORTE_SIGNATURE)) {
-        if (sig->sz != coll->sig->sz) {
-            continue;
-        }
-        if (0 == memcmp(sig->signature, coll->sig->signature, coll->sig->sz)) {
             OPAL_OUTPUT_VERBOSE((1, orte_grpcomm_base_framework.framework_output,
                                  "%s grpcomm:base:returning existing collective",
                                  ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
@@ -217,9 +213,13 @@ orte_grpcomm_coll_t* orte_grpcomm_base_get_tracker(orte_grpcomm_signature_t *sig
 
         return NULL;
     }
-    OPAL_OUTPUT_VERBOSE((1, orte_grpcomm_base_framework.framework_output,
+    if (1 < opal_output_get_verbosity(orte_grpcomm_base_framework.framework_output)) {
-                         "%s grpcomm:base: creating new coll",
+        char *tmp=NULL;
-                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
+        (void)opal_dss.print(&tmp, NULL, sig, ORTE_SIGNATURE);
+        opal_output(0, "%s grpcomm:base: creating new coll for procs %s",
+                    ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), tmp);
+        free(tmp);
+    }
     coll = OBJ_NEW(orte_grpcomm_coll_t);
     OBJ_RETAIN(sig);
     coll->sig = sig;

orte/mca/grpcomm/direct/grpcomm_direct.c

@@ -437,8 +437,8 @@ static void xcast_recv(int status, orte_process_name_t* sender,
         nm = (orte_namelist_t*)item;
 
         OPAL_OUTPUT_VERBOSE((5, orte_grpcomm_base_framework.framework_output,
-                             "%s grpcomm:direct:send_relay sending relay msg to %s",
+                             "%s grpcomm:direct:send_relay sending relay msg of %d bytes to %s",
-                             ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
+                             ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), (int)rly->bytes_used,
                              ORTE_NAME_PRINT(&nm->name)));
         OBJ_RETAIN(rly);
         /* check the state of the recipient - no point

orte/orted/pmix/pmix_server.c

@@ -676,6 +676,13 @@ static void pmix_server_release(int status,
     pmix_server_peer_t *peer;
     opal_buffer_t *reply;
 
+    if (2 < opal_output_get_verbosity(pmix_server_output)) {
+        char *tmp=NULL;
+        (void)opal_dss.print(&tmp, NULL, trk->sig, ORTE_SIGNATURE);
+        opal_output(0, "%s pmix_server release called on tracker %s",
+                    ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), tmp);
+        free(tmp);
+    }
     opal_output_verbose(2, pmix_server_output,
                         "%s pmix:server:release coll release recvd",
                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));

orte/orted/pmix/pmix_server_sendrecv.c

@@ -715,6 +715,14 @@ static void process_message(pmix_server_peer_t *peer)
                 goto reply_fence;
             }
         }
+        if (4 < opal_output_get_verbosity(pmix_server_output)) {
+            char *tmp=NULL;
+            (void)opal_dss.print(&tmp, NULL, sig, ORTE_SIGNATURE);
+            opal_output(0, "%s %s called with procs %s",
+                        ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
+                        (PMIX_FENCENB_CMD == cmd) ? "FENCE_NB" : "FENCE", tmp);
+            free(tmp);
+        }
         /* get the URI for this process */
         cnt = 1;
         if (OPAL_SUCCESS != (rc = opal_dss.unpack(&xfer, &local_uri, &cnt, OPAL_STRING))) {

orte/runtime/data_type_support/orte_dt_print_fns.c

@@ -895,6 +895,36 @@ int orte_dt_print_attr(char **output, char *prefix,
 
 int orte_dt_print_sig(char **output, char *prefix, orte_grpcomm_signature_t *src, opal_data_type_t type)
 {
+    char *prefx;
+    size_t i;
+    char *tmp, *tmp2;
+
+    /* deal with NULL prefix */
+    if (NULL == prefix) asprintf(&prefx, " ");
+    else prefx = strdup(prefix);
+    
+    /* if src is NULL, just print data type and return */
+    if (NULL == src) {
+        asprintf(output, "%sData type: ORTE_SIG\tValue: NULL pointer", prefx);
+        free(prefx);
+        return OPAL_SUCCESS;
+    }
+
+    if (NULL == src->signature) {
+        asprintf(output, "%sORTE_SIG\tValue: NULL", prefx);
+        free(prefx);
+        return ORTE_SUCCESS;
+    }
+
+    /* there must be at least one */
+    asprintf(&tmp, "%sORTE_SIG\tValue: ", prefx);
+
+    for (i=0; i < src->sz; i++) {
+        asprintf(&tmp2, "%s%s", tmp, ORTE_NAME_PRINT(&src->signature[i]));
+        free(tmp);
+        tmp = tmp2;
+    }
+    *output = tmp;
     return ORTE_SUCCESS;
 }