Also remove the comm from processes when SIMIX_post_comm is not called

[simgrid.git] / src / simix / smx_network.c

diff --git a/src/simix/smx_network.c b/src/simix/smx_network.c
index 3c4440e..c3265f6 100644 (file)
--- a/src/simix/smx_network.c
+++ b/src/simix/smx_network.c
@@ -134,7 +134,7 @@ smx_action_t SIMIX_rdv_get_request(smx_rdv_t rdv, e_smx_comm_type_t type,
   xbt_fifo_item_t item;
   void* req_data = NULL;
 
-  xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t){
+  xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t) {
     if (action->comm.type == SIMIX_COMM_SEND) {
       req_data = action->comm.src_data;
     } else if (action->comm.type == SIMIX_COMM_RECEIVE) {
@@ -187,7 +187,7 @@ int SIMIX_comm_has_recv_match(smx_rdv_t rdv, int (*match_fun)(void*, void*), voi
   smx_action_t action;
   xbt_fifo_item_t item;
 
-  xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t){
+  xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t) {
     if (action->comm.type == SIMIX_COMM_RECEIVE
         && (!match_fun || match_fun(data, action->comm.dst_data))) {
       XBT_DEBUG("Found a matching communication action %p", action);
@@ -213,6 +213,7 @@ smx_action_t SIMIX_comm_new(e_smx_comm_type_t type)
 
   /* alloc structures */
   act = xbt_mallocator_get(simix_global->action_mallocator);
+
   act->type = SIMIX_ACTION_COMMUNICATE;
   act->state = SIMIX_WAITING;
 
@@ -241,11 +242,12 @@ smx_action_t SIMIX_comm_new(e_smx_comm_type_t type)
  */
 void SIMIX_comm_destroy(smx_action_t action)
 {
-  XBT_DEBUG("Destroy action %p (refcount:%d)", action, action->comm.refcount);
+  XBT_DEBUG("Destroy action %p (refcount: %d), state: %d",
+      action, action->comm.refcount, action->state);
 
-  if (action->comm.refcount <= 0)
-    xbt_die("the refcount of comm %p is already 0 before decreasing it. "
-            "That's a bug!", action);
+  xbt_assert(action->comm.refcount > 0,
+      "The refcount of comm %p is already 0 before decreasing it. "
+      "That's a bug!", action);
 
   action->comm.refcount--;
   if (action->comm.refcount > 0)
@@ -546,6 +548,10 @@ XBT_INLINE void SIMIX_comm_start(smx_action_t action)
   }
 }
 
+/**
+ * \brief Answers the SIMIX requests associated to a communication action.
+ * \param action a finished communication action
+ */
 void SIMIX_comm_finish(smx_action_t action)
 {
   unsigned int destroy_count = 0;
@@ -622,7 +628,9 @@ void SIMIX_comm_finish(smx_action_t action)
       case SIMIX_LINK_FAILURE:
         TRY {
          XBT_DEBUG("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p)",
-             action, action->comm.src_proc->smx_host->name, action->comm.dst_proc->smx_host->name,
+             action,
+             action->comm.src_proc ? action->comm.src_proc->smx_host->name : NULL,
+             action->comm.dst_proc ? action->comm.dst_proc->smx_host->name : NULL,
              req->issuer->name, req->issuer);
           THROWF(network_error, 0, "Link failure");
         }
@@ -631,8 +639,22 @@ void SIMIX_comm_finish(smx_action_t action)
         }
         break;
 
+      case SIMIX_CANCELED:
+        TRY {
+          if (req->issuer == action->comm.dst_proc) {
+            THROWF(cancel_error, 0, "Communication canceled by the sender");
+          }
+          else {
+            THROWF(cancel_error, 0, "Communication canceled by the receiver");
+          }
+        }
+        CATCH(req->issuer->running_ctx->exception) {
+          req->issuer->doexception = 1;
+        }
+        break;
+
       default:
-        THROW_IMPOSSIBLE;
+        xbt_die("Unexpected action state in SIMIX_comm_finish: %d", action->state);
     }
 
     /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */
@@ -646,6 +668,7 @@ void SIMIX_comm_finish(smx_action_t action)
     }
 
     req->issuer->waiting_action = NULL;
+    xbt_fifo_remove(req->issuer->comms, action);
     SIMIX_request_answer(req);
     destroy_count++;
   }
@@ -654,6 +677,10 @@ void SIMIX_comm_finish(smx_action_t action)
     SIMIX_comm_destroy(action);
 }
 
+/**
+ * \brief This function is called when a Surf communication action is finished.
+ * \param action the corresponding Simix communication
+ */
 void SIMIX_post_comm(smx_action_t action)
 {
   /* Update action state */
@@ -678,7 +705,11 @@ void SIMIX_post_comm(smx_action_t action)
   XBT_DEBUG("SIMIX_post_comm: comm %p, state %d, src_proc %p, dst_proc %p, detached: %d",
       action, action->state, action->comm.src_proc, action->comm.dst_proc, action->comm.detached);
 
-  /* remove the action from pending communications of both processes (if they still exist) */
+  /* destroy the surf actions associated with the Simix communication */
+  SIMIX_comm_destroy_internal_actions(action);
+
+  /* remove the communication action from the list of pending communications
+   * of both processes (if they still exist) */
   if (action->comm.src_proc) {
     xbt_fifo_remove(action->comm.src_proc->comms, action);
   }
@@ -686,25 +717,24 @@ void SIMIX_post_comm(smx_action_t action)
     xbt_fifo_remove(action->comm.dst_proc->comms, action);
   }
 
-  /* destroy the surf actions associated with the Simix communication */
-  SIMIX_comm_destroy_internal_actions(action);
-
   /* if there are requests associated with the action, then answer them */
-  if (xbt_fifo_size(action->request_list))
+  if (xbt_fifo_size(action->request_list)) {
     SIMIX_comm_finish(action);
+  }
 }
 
 void SIMIX_comm_cancel(smx_action_t action)
 {
-  /* If the action is a waiting state means that it is still in a rdv */
+  /* if the action is a waiting state means that it is still in a rdv */
   /* so remove from it and delete it */
   if (action->state == SIMIX_WAITING) {
     SIMIX_rdv_remove(action->comm.rdv, action);
-    action->state = SIMIX_FAILED;
-  } else {
-    /* When running the MC there are no surf actions */
-    if(!MC_IS_ENABLED)
-      surf_workstation_model->action_cancel(action->comm.surf_comm);
+    action->state = SIMIX_CANCELED;
+  }
+  else if (!MC_IS_ENABLED /* when running the MC there are no surf actions */
+      && (action->state == SIMIX_READY || action->state == SIMIX_RUNNING)) {
+
+    surf_workstation_model->action_cancel(action->comm.surf_comm);
   }
 }