Merge pull request #93 from rhc54/topic/status

Preserve and return the app exit status

orte/mca/errmgr/dvm/errmgr_dvm.c

@@ -337,10 +337,12 @@ static void proc_errors(int fd, short args, void *cbdata)
             OBJ_RETAIN(pptr);
             ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
             /* update our exit code */
-            ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
+            jdata->exit_code = pptr->exit_code;
             /* just in case the exit code hadn't been set, do it here - this
              * won't override any reported exit code */
-            ORTE_UPDATE_EXIT_STATUS(ORTE_ERR_COMM_FAILURE);
+            if (0 == jdata->exit_code) {
+                jdata->exit_code = ORTE_ERR_COMM_FAILURE;
+            }
         }
         goto cleanup;
     }
@@ -405,7 +407,7 @@ static void proc_errors(int fd, short args, void *cbdata)
             /* retain the object so it doesn't get free'd */
             OBJ_RETAIN(pptr);
             ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
-            ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
+            jdata->exit_code = pptr->exit_code;
             /* kill the job */
             _terminate_job(jdata->jobid);
         }
@@ -423,7 +425,7 @@ static void proc_errors(int fd, short args, void *cbdata)
             /* retain the object so it doesn't get free'd */
             OBJ_RETAIN(pptr);
             ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
-            ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
+            jdata->exit_code = pptr->exit_code;
             /* kill the job */
             _terminate_job(jdata->jobid);
         }
@@ -441,13 +443,15 @@ static void proc_errors(int fd, short args, void *cbdata)
             /* retain the object so it doesn't get free'd */
             OBJ_RETAIN(pptr);
             ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
-            ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
+            jdata->exit_code = pptr->exit_code;
             /* now treat a special case - if the proc exit'd without a required
              * sync, it may have done so with a zero exit code. We want to ensure
              * that the user realizes there was an error, so in this -one- case,
              * we overwrite the process' exit code with the default error code
              */
-            ORTE_UPDATE_EXIT_STATUS(ORTE_ERROR_DEFAULT_EXIT_CODE);
+            if (0 == jdata->exit_code) {
+                jdata->exit_code = ORTE_ERROR_DEFAULT_EXIT_CODE;
+            }
              /* kill the job */
             _terminate_job(jdata->jobid);
        }
@@ -537,7 +541,7 @@ static void proc_errors(int fd, short args, void *cbdata)
             /* retain the object so it doesn't get free'd */
             OBJ_RETAIN(pptr);
             ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
-            ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
+            jdata->exit_code = pptr->exit_code;
             /* kill the job */
             _terminate_job(jdata->jobid);
         }
@@ -549,7 +553,7 @@ static void proc_errors(int fd, short args, void *cbdata)
                              ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
                              ORTE_NAME_PRINT(proc),
                              pptr->exit_code));
-        ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
+        jdata->exit_code = pptr->exit_code;
         /* track the number of non-zero exits */
         i32 = 0;
         i32ptr = &i32;
@@ -588,7 +592,7 @@ static void proc_errors(int fd, short args, void *cbdata)
             /* retain the object so it doesn't get free'd */
             OBJ_RETAIN(pptr);
             ORTE_FLAG_SET(jdata, ORTE_JOB_FLAG_ABORTED);
-            ORTE_UPDATE_EXIT_STATUS(pptr->exit_code);
+            jdata->exit_code = pptr->exit_code;
             /* kill the job */
             _terminate_job(jdata->jobid);
         }

orte/mca/state/dvm/state_dvm.c

@@ -629,12 +629,12 @@ static void dvm_notify(int sd, short args, void *cbdata)
 
     /* see if there was any problem */
     if (orte_get_attribute(&jdata->attributes, ORTE_JOB_ABORTED_PROC, (void**)&pptr, OPAL_PTR) && NULL != pptr) {
-        ret = opal_pmix_convert_rc(pptr->exit_code);
+        ret = opal_pmix_convert_rc(jdata->exit_code);
     /* or whether we got cancelled by the user */
     } else if (orte_get_attribute(&jdata->attributes, ORTE_JOB_CANCELLED, NULL, OPAL_BOOL)) {
         ret = opal_pmix_convert_rc(ORTE_ERR_JOB_CANCELLED);
     } else {
-        ret = PMIX_SUCCESS;
+        ret = opal_pmix_convert_rc(jdata->exit_code);
     }
 
     if (0 == ret && orte_get_attribute(&jdata->attributes, ORTE_JOB_SILENT_TERMINATION, NULL, OPAL_BOOL)) {

orte/runtime/orte_globals.c

@@ -618,6 +618,7 @@ OBJ_CLASS_INSTANCE(orte_app_context_t,
 
 static void orte_job_construct(orte_job_t* job)
 {
+    job->exit_code = 0;
     job->personality = NULL;
     job->jobid = ORTE_JOBID_INVALID;
     job->offset = 0;

orte/runtime/orte_globals.h

@@ -314,6 +314,8 @@ ORTE_DECLSPEC OBJ_CLASS_DECLARATION(orte_node_t);
 typedef struct {
     /** Base object so this can be put on a list */
     opal_list_item_t super;
+    /* record the exit status for this job */
+    int exit_code;
     /* personality for this job */
     char **personality;
     /* jobid for this job */

orte/tools/prun/prun.c

@@ -268,6 +268,8 @@ static void evhandler(size_t evhdlr_registration_id,
                         ORTE_JOBID_PRINT(jobid), jobstatus);
         }
     }
+    /* save the status */
+    lock->status = jobstatus;
     /* release the lock */
     OPAL_PMIX_WAKEUP_THREAD(lock);
 
@@ -390,7 +392,7 @@ static void clean_abort(int fd, short flags, void *arg);
 
 int prun(int argc, char *argv[])
 {
-    int rc, i;
+    int rc=1, i;
     char *param, *ptr;
     opal_pmix_lock_t lock, rellock;
     opal_list_t apps;
@@ -1078,6 +1080,8 @@ int prun(int argc, char *argv[])
 
     OPAL_PMIX_WAIT_THREAD(&rellock);
     OPAL_PMIX_DESTRUCT_LOCK(&rellock);
+    /* save the status */
+    rc = rellock.status;
 
     OPAL_PMIX_CONSTRUCT_LOCK(&lock);
     PMIx_Deregister_event_handler(evid, opcbfunc, &lock);
@@ -1089,7 +1093,7 @@ int prun(int argc, char *argv[])
     PMIx_tool_finalize();
     opal_progress_thread_finalize(NULL);
     opal_finalize();
-    return 0;
+    return rc;
 }
 
 static int parse_locals(opal_list_t *jdata, int argc, char* argv[])