view serialization accounts for dead members now

src/ssg-rpc.c

@@ -262,7 +262,7 @@ static int ssg_group_view_serialize(
     unsigned int i;
     hg_size_t view_size = 0;
     int tmp_size;
-    void *view_buf, *p;
+    void *view_buf, *buf_p, *str_p;
 
     *buf = NULL;
     *buf_size = 0;
@@ -270,19 +270,33 @@ static int ssg_group_view_serialize(
     /* first determine view size */
     for (i = 0; i < view->size; i++)
     {
-        view_size += (strlen(view->member_states[i].addr_str) + 1);
+        if (view->member_states[i].addr_str)
+            view_size += (strlen(view->member_states[i].addr_str) + 1);
+        else
+            view_size += 1;
     }
 
     view_buf = malloc(view_size);
     if(!view_buf)
         return SSG_FAILURE;
 
-    p = view_buf;
+    buf_p = view_buf;
     for (i = 0; i < view->size; i++)
     {
-        tmp_size = strlen(view->member_states[i].addr_str) + 1;
-        memcpy(p, view->member_states[i].addr_str, tmp_size);
-        p += tmp_size;
+        char null = '\0';
+
+        if (view->member_states[i].addr_str)
+        {
+            tmp_size = strlen(view->member_states[i].addr_str) + 1;
+            str_p = view->member_states[i].addr_str;
+        }
+        else
+        {
+            tmp_size = 1;
+            str_p = &null;
+        }
+        memcpy(buf_p, str_p, tmp_size);
+        buf_p += tmp_size;
     }
 
     *buf = view_buf;

src/ssg.c

@@ -737,11 +737,20 @@ void ssg_apply_membership_update(
     ssg_group_t *g,
     ssg_membership_update_t update)
 {
-    if (!g) return;
+    hg_class_t *hgcl = NULL;
+    
+    if(!ssg_inst || !g) return;
+
+    hgcl = margo_get_class(ssg_inst->mid);
+    if (!hgcl) return;
 
     if (update.type == SSG_MEMBER_REMOVE)
     {
+        HG_Addr_free(hgcl, g->view.member_states[update.member].addr);
+        free(g->view.member_states[update.member].addr_str);
+        g->view.member_states[update.member].addr_str = NULL;
         g->view.member_states[update.member].is_member = 0;
+        /* XXX: need to update size ... g->view.size--; */
     }
     else
     {
@@ -842,7 +851,7 @@ static int ssg_group_view_create(
     {
         view->member_states[i].addr_str = NULL;
         view->member_states[i].addr = HG_ADDR_NULL;
-        view->member_states[i].is_member = 1;
+        view->member_states[i].is_member = 0;
         lookup_ults[i] = ABT_THREAD_NULL;
     }
 
@@ -870,6 +879,8 @@ static int ssg_group_view_create(
          */
         j = (r + i) % view->size;
 
+        if (group_addr_strs[j] == NULL || strlen(group_addr_strs[j]) == 0) continue;
+
         view->member_states[j].addr_str = strdup(group_addr_strs[j]);
         if (!view->member_states[j].addr_str)
         {
@@ -926,6 +937,7 @@ static int ssg_group_view_create(
             sret = SSG_FAILURE;
             break;
         }
+        view->member_states[i].is_member = 1;
     }
 
 fini: