Update the PMIx support to the PMIx 1.1.2 release.

ompi/dpm/dpm.c

@@ -536,7 +536,7 @@ static int construct_peers(ompi_group_t *group, opal_list_t *peers)
         } else {
             proc_name = proct->super.proc_name;
         }
-      
+
         /* add to the list of peers */
         nm = OBJ_NEW(opal_namelist_t);
         nm->name = proc_name;
@@ -1203,7 +1203,7 @@ static bool ompi_dpm_group_is_dyn (ompi_group_t *group, ompi_jobid_t thisjobid)
 {
     int size = group ? ompi_group_size (group) : 0;
 
-    for (int i = 1 ; i < size ; ++i) {
+    for (int i = 0 ; i < size ; ++i) {
         opal_process_name_t name = ompi_group_get_proc_name (group, i);
 
         if (thisjobid != ((ompi_process_name_t *) &name)->jobid) {

opal/Makefile.am

@@ -75,6 +75,7 @@ nobase_opal_HEADERS = $(headers)
 endif
 
 include class/Makefile.am
+include errhandler/Makefile.am
 include memoryhooks/Makefile.am
 include runtime/Makefile.am
 include threads/Makefile.am

opal/errhandler/Makefile.am

@@ -0,0 +1,17 @@
+# -*- makefile -*-
+#
+# Copyright (c) 2015      Intel, Inc.  All rights reserved.
+# $COPYRIGHT$
+#
+# Additional copyrights may follow
+#
+# $HEADER$
+#
+
+# This makefile.am does not stand on its own - it is included from opal/Makefile.am
+
+headers += \
+        errhandler/opal_errhandler.h
+
+lib@OPAL_LIB_PREFIX@open_pal_la_SOURCES += \
+        errhandler/opal_errhandler.c

opal/errhandler/opal_errhandler.c

@@ -0,0 +1,34 @@
+/*
+ * Copyright (c) 2015      Intel, Inc.  All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+
+#include "opal_config.h"
+
+#include "opal/errhandler/opal_errhandler.h"
+
+opal_errhandler_fn_t errhandler = NULL;
+void *cbdata = NULL;
+
+void opal_register_errhandler(opal_errhandler_fn_t newerr, void *cbd)
+{
+    errhandler = newerr;
+    cbdata = cbd;
+}
+
+void opal_deregister_errhandler(void)
+{
+    errhandler = NULL;
+    cbdata = NULL;
+}
+
+void opal_invoke_errhandler(int status, opal_proc_t *proc)
+{
+    if (NULL != errhandler) {
+        errhandler(status, proc, cbdata);
+    }
+}

opal/errhandler/opal_errhandler.h

@@ -0,0 +1,25 @@
+/*
+ * Copyright (c) 2015      Intel, Inc.  All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+
+#ifndef OPAL_ERRHANDLER_H
+#define OPAL_ERRHANDLER_H
+
+#include "opal_config.h"
+
+#include "opal/util/proc.h"
+
+typedef void (*opal_errhandler_fn_t)(int status, opal_proc_t *proc, void *cbdata);
+
+OPAL_DECLSPEC void opal_register_errhandler(opal_errhandler_fn_t errhandler, void *cbdata);
+
+OPAL_DECLSPEC void opal_deregister_errhandler(void);
+
+OPAL_DECLSPEC void opal_invoke_errhandler(int status, opal_proc_t *proc);
+
+#endif

opal/mca/pmix/base/base.h

@@ -32,11 +32,17 @@ OPAL_DECLSPEC int opal_pmix_base_select(void);
 
 OPAL_DECLSPEC extern bool opal_pmix_base_allow_delayed_server;
 
-OPAL_DECLSPEC void opal_pmix_base_register_handler(opal_pmix_errhandler_fn_t err);
-OPAL_DECLSPEC void opal_pmix_base_deregister_handler(void);
+OPAL_DECLSPEC void opal_pmix_base_register_handler(opal_list_t *info,
+                                                   opal_pmix_notification_fn_t errhandler,
+                                                   opal_pmix_errhandler_reg_cbfunc_t cbfunc,
+                                                   void *cbdata);
+OPAL_DECLSPEC void opal_pmix_base_deregister_handler(int errhandler,
+                                                     opal_pmix_op_cbfunc_t cbfunc,
+                                                     void *cbdata);
 OPAL_DECLSPEC void opal_pmix_base_errhandler(int status,
                                              opal_list_t *procs,
-                                             opal_list_t *info);
+                                             opal_list_t *info,
+                                             opal_pmix_release_cbfunc_t cbfunc, void *cbdata);
 OPAL_DECLSPEC int opal_pmix_base_exchange(opal_value_t *info,
                                           opal_pmix_pdata_t *pdat,
                                           int timeout);

opal/mca/pmix/base/pmix_base_fns.c

@@ -38,26 +38,40 @@
 
 #define OPAL_PMI_PAD  10
 
-/********     ERRHANDLER SUPPORT     ********/
-static opal_pmix_errhandler_fn_t errhandler = NULL;
-
-void opal_pmix_base_register_handler(opal_pmix_errhandler_fn_t err)
+/********     ERRHANDLER SUPPORT FOR COMPONENTS THAT
+ ********     DO NOT NATIVELY SUPPORT IT
+ ********/
+static opal_pmix_notification_fn_t errhandler = NULL;
+
+void opal_pmix_base_register_handler(opal_list_t *info,
+                                     opal_pmix_notification_fn_t err,
+                                     opal_pmix_errhandler_reg_cbfunc_t cbfunc,
+                                     void *cbdata)
 {
     errhandler = err;
+    if (NULL != cbfunc) {
+        cbfunc(OPAL_SUCCESS, 0, cbdata);
+    }
 }
 
 void opal_pmix_base_errhandler(int status,
                                opal_list_t *procs,
-                               opal_list_t *info)
+                               opal_list_t *info,
+                               opal_pmix_release_cbfunc_t cbfunc, void *cbdata)
 {
     if (NULL != errhandler) {
-        errhandler(status);
+        errhandler(status, procs, info, cbfunc, cbdata);
     }
 }
 
-void opal_pmix_base_deregister_handler(void)
+void opal_pmix_base_deregister_handler(int errid,
+                                       opal_pmix_op_cbfunc_t cbfunc,
+                                       void *cbdata)
 {
     errhandler = NULL;
+    if (NULL != cbfunc) {
+        cbfunc(OPAL_SUCCESS, cbdata);
+    }
 }
 
 struct lookup_caddy_t {

opal/mca/pmix/external/Makefile.am

@@ -44,7 +44,5 @@ noinst_LTLIBRARIES = $(component_noinst)
 libmca_pmix_external_la_SOURCES =$(sources)
 libmca_pmix_external_la_CFLAGS =
 libmca_pmix_external_la_CPPFLAGS = $(opal_pmix_ext_CPPFLAGS)
-libmca_pmix_external_la_LDFLAGS = -module -avoid-version -L$(opal_pmix_ext_LDFLAGS)
-libmca_pmix_external_la_LIBADD = $(opal_pmix_ext_LIBS) \
-        $(OPAL_TOP_BUILDDIR)/opal/mca/event/lib@OPAL_LIB_PREFIX@mca_event.la \
-        $(OPAL_TOP_BUILDDIR)/opal/mca/hwloc/lib@OPAL_LIB_PREFIX@mca_hwloc.la
+libmca_pmix_external_la_LDFLAGS = -module -avoid-version $(opal_pmix_ext_LDFLAGS)
+libmca_pmix_external_la_LIBADD = $(opal_pmix_ext_LIBS)

opal/mca/pmix/external/pmix_ext_client.c

@@ -36,6 +36,18 @@ static pmix_proc_t my_proc;
 static char *dbgvalue=NULL;
 static int errhdler_ref = 0;
 
+static void completion_handler (void * cbdata) {
+    int * cond = (int *)cbdata;
+    *cond = 0;
+}
+
+#define PMIX_WAIT_FOR_COMPLETION(a)             \
+    do {                                        \
+        while ((a)) {                           \
+            usleep(10);                         \
+        }                                       \
+    } while (0);
+
 static void myerr(pmix_status_t status,
                   pmix_proc_t procs[], size_t nprocs,
                   pmix_info_t info[], size_t ninfo)
@@ -45,6 +57,7 @@ static void myerr(pmix_status_t status,
     opal_namelist_t *nm;
     opal_value_t *iptr;
     size_t n;
+    volatile int cond = 1;
 
     /* convert the incoming status */
     rc = pmix1_convert_rc(status);
@@ -64,11 +77,13 @@ static void myerr(pmix_status_t status,
         iptr = OBJ_NEW(opal_value_t);
         iptr->key = strdup(info[n].key);
         pmix1_value_unload(iptr, &info[n].value);
-        opal_list_append(&plist, &nm->super);
+        opal_list_append(&plist, &iptr->super);
     }
 
     /* call the base errhandler */
-    opal_pmix_base_errhandler(rc, &plist, &ilist);
+    opal_pmix_base_errhandler(rc, &plist, &ilist, completion_handler, (void *)&cond);
+    PMIX_WAIT_FOR_COMPLETION(cond);
+
     OPAL_LIST_DESTRUCT(&plist);
     OPAL_LIST_DESTRUCT(&ilist);
 }

opal/mca/pmix/external/pmix_ext_server_south.c

@@ -50,6 +50,18 @@ extern opal_pmix_server_module_t *host_module;
 static char *dbgvalue=NULL;
 static int errhdler_ref = 0;
 
+static void completion_handler (void * cbdata) {
+    int * cond = (int *)cbdata;
+    *cond = 0;
+}
+
+#define PMIX_WAIT_FOR_COMPLETION(a)             \
+    do {                                        \
+        while ((a)) {                           \
+            usleep(10);                         \
+        }                                       \
+    } while (0);
+
 static void myerr(pmix_status_t status,
                   pmix_proc_t procs[], size_t nprocs,
                   pmix_info_t info[], size_t ninfo)
@@ -58,6 +70,7 @@ static void myerr(pmix_status_t status,
     opal_list_t plist, ilist;
     opal_namelist_t *nm;
     opal_value_t *iptr;
+    volatile int cond = 1;
     size_t n;
 
     /* convert the incoming status */
@@ -78,11 +91,13 @@ static void myerr(pmix_status_t status,
         iptr = OBJ_NEW(opal_value_t);
         iptr->key = strdup(info[n].key);
         pmix1_value_unload(iptr, &info[n].value);
-        opal_list_append(&plist, &nm->super);
+        opal_list_append(&plist, &iptr->super);
     }
 
     /* call the base errhandler */
-    opal_pmix_base_errhandler(rc, &plist, &ilist);
+    opal_pmix_base_errhandler(rc, &plist, &ilist, completion_handler, (void *)&cond);
+    PMIX_WAIT_FOR_COMPLETION(cond);
+
     OPAL_LIST_DESTRUCT(&plist);
     OPAL_LIST_DESTRUCT(&ilist);
 }