Cleanup the code a bit by simply adding our nspace to the top of the list of jobid <-> nspace correlations. Add two new APIs to opal_pmix for registering new jobid/nspace pairs and retrieving an nspace given a jobid - these are required to support connect/accept. No impact on the PMIx library.

(cherry picked from commit a4a3dfd)

Author: Ralph Castain

ompi/dpm/dpm.c

@@ -139,7 +139,13 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
         opal_argv_append_nosize(&members, nstring);
         free(nstring);
         /* have to add the number of procs in the job so the remote side
-         * can correctly add the procs by computing their names */
+         * can correctly add the procs by computing their names, and our nspace
+         * so they can update their records */
+        if (NULL == (nstring = (char*)opal_pmix.get_nspace(OMPI_PROC_MY_NAME->jobid))) {
+            opal_argv_free(members);
+            return OMPI_ERR_NOT_SUPPORTED;
+        }
+        opal_argv_append_nosize(&members, nstring);
         (void)asprintf(&nstring, "%d", size);
         opal_argv_append_nosize(&members, nstring);
         free(nstring);
@@ -171,6 +177,11 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
             }
             opal_argv_append_nosize(&members, nstring);
             free(nstring);
+            if (NULL == (nstring = (char*)opal_pmix.get_nspace(proc_list[i]->super.proc_name.jobid))) {
+                opal_argv_free(members);
+                return OMPI_ERR_NOT_SUPPORTED;
+            }
+            opal_argv_append_nosize(&members, nstring);
         }
         if (!dense) {
             free(proc_list);
@@ -246,6 +257,17 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
             OPAL_LIST_DESTRUCT(&mlist);
             goto exit;
         }
+        /* step over the nspace */
+        ++i;
+        if (NULL == members[i]) {
+            /* this shouldn't happen and is an error */
+            OMPI_ERROR_LOG(OMPI_ERR_BAD_PARAM);
+            OPAL_LIST_DESTRUCT(&mlist);
+            opal_argv_free(members);
+            free(rport);
+            rc = OMPI_ERR_BAD_PARAM;
+            goto exit;
+        }
         /* if the rank is wildcard, then we need to add all procs
          * in that job to the list */
         if (OPAL_VPID_WILDCARD == nm->name.vpid) {
@@ -295,6 +317,16 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
             OPAL_LIST_DESTRUCT(&rlist);
             goto exit;
         }
+        /* next entry is the nspace - register it */
+        ++i;
+        if (NULL == members[i]) {
+            OMPI_ERROR_LOG(OMPI_ERR_NOT_SUPPORTED);
+            opal_argv_free(members);
+            OPAL_LIST_DESTRUCT(&ilist);
+            OPAL_LIST_DESTRUCT(&rlist);
+            goto exit;
+        }
+        opal_pmix.register_jobid(nm->name.jobid, members[i]);
         if (OPAL_VPID_WILDCARD == nm->name.vpid) {
             jobid = nm->name.jobid;
             OBJ_RELEASE(nm);

opal/mca/pmix/cray/pmix_cray.c

@@ -78,6 +78,9 @@ static int cray_unpublish_nb(char **keys, opal_list_t *info,
 static const char *cray_get_version(void);
 static int cray_store_local(const opal_process_name_t *proc,
                           opal_value_t *val);
+static const char *cray_get_nspace(opal_jobid_t jobid);
+static void cray_register_jobid(opal_jobid_t jobid, const char *nspace);
+
 #if 0
 static bool cray_get_attr(const char *attr, opal_value_t **kv);
 #endif
@@ -109,7 +112,9 @@ const opal_pmix_base_module_t opal_pmix_cray_module = {
     .get_version = cray_get_version,
     .register_errhandler = opal_pmix_base_register_handler,
     .deregister_errhandler = opal_pmix_base_deregister_handler,
-    .store_local = cray_store_local
+    .store_local = cray_store_local,
+    .get_nspace = cray_get_nspace,
+    .register_jobid = cray_register_jobid
 };
 
 // usage accounting
@@ -814,6 +819,16 @@ static int cray_store_local(const opal_process_name_t *proc,
     return OPAL_SUCCESS;
 }
 
+static const char *cray_get_nspace(opal_jobid_t jobid)
+{
+    return NULL;
+}
+
+static void cray_register_jobid(opal_jobid_t jobid, const char *nspace)
+{
+    return;
+}
+
 static char* pmix_error(int pmix_err)
 {
     char * err_msg;

opal/mca/pmix/pmix.h

@@ -701,6 +701,12 @@ typedef void (*opal_pmix_base_module_deregister_fn_t)(void);
 typedef int (*opal_pmix_base_module_store_fn_t)(const opal_process_name_t *proc,
                                                 opal_value_t *val);
 
+/* retrieve the nspace corresponding to a given jobid */
+typedef const char* (*opal_pmix_base_module_get_nspace_fn_t)(opal_jobid_t jobid);
+
+/* register a jobid-to-nspace pair */
+typedef void (*opal_pmix_base_module_register_jobid_fn_t)(opal_jobid_t jobid, const char *nspace);
+
 /*
  * the standard public API data structure
  */
@@ -745,6 +751,8 @@ typedef struct {
     opal_pmix_base_module_register_fn_t               register_errhandler;
     opal_pmix_base_module_deregister_fn_t             deregister_errhandler;
     opal_pmix_base_module_store_fn_t                  store_local;
+    opal_pmix_base_module_get_nspace_fn_t             get_nspace;
+    opal_pmix_base_module_register_jobid_fn_t         register_jobid;
 } opal_pmix_base_module_t;
 
 typedef struct {

opal/mca/pmix/pmix1xx/pmix1.h

@@ -30,11 +30,24 @@
 
 BEGIN_C_DECLS
 
-OPAL_DECLSPEC extern opal_pmix_base_component_t mca_pmix_pmix1_component;
+typedef struct {
+  opal_pmix_base_component_t super;
+  opal_list_t jobids;
+  bool native_launch;
+} mca_pmix_pmix1_component_t;
+
+OPAL_DECLSPEC extern mca_pmix_pmix1_component_t mca_pmix_pmix1xx_component;
 
 OPAL_DECLSPEC extern const opal_pmix_base_module_t opal_pmix_pmix1xx_module;
 
 /****  INTERNAL OBJECTS  ****/
+typedef struct {
+    opal_list_item_t super;
+    opal_jobid_t jobid;
+    char nspace[PMIX_MAX_NSLEN + 1];
+} opal_pmix1_jobid_trkr_t;
+OBJ_CLASS_DECLARATION(opal_pmix1_jobid_trkr_t);
+
 typedef struct {
     opal_object_t super;
     pmix_proc_t p;