SIMIX_network_copy_data() is a mess because it tries to allow passing

[simgrid.git] / src / msg / msg_mailbox.c

diff --git a/src/msg/msg_mailbox.c b/src/msg/msg_mailbox.c
index 1f1954d..c897b6a 100644 (file)
--- a/src/msg/msg_mailbox.c
+++ b/src/msg/msg_mailbox.c
@@ -20,11 +20,10 @@ msg_mailbox_t MSG_mailbox_create(const char *alias)
 {
   msg_mailbox_t mailbox = xbt_new0(s_msg_mailbox_t, 1);
 
-  mailbox->tasks = xbt_fifo_new();
   mailbox->cond = NULL;
   mailbox->alias = alias ? xbt_strdup(alias) : NULL;
-  mailbox->hostname = NULL;
-
+  mailbox->rdv = SIMIX_rdv_create(alias);
+  
   return mailbox;
 }
 
@@ -42,12 +41,9 @@ void MSG_mailbox_free(void *mailbox)
 {
   msg_mailbox_t _mailbox = (msg_mailbox_t) mailbox;
 
-  if (_mailbox->hostname)
-    free(_mailbox->hostname);
-
-  xbt_fifo_free(_mailbox->tasks);
   free(_mailbox->alias);
-
+  SIMIX_rdv_destroy(_mailbox->rdv);
+  
   free(_mailbox);
 }
 
@@ -56,59 +52,25 @@ smx_cond_t MSG_mailbox_get_cond(msg_mailbox_t mailbox)
   return mailbox->cond;
 }
 
-void MSG_mailbox_remove(msg_mailbox_t mailbox, m_task_t task)
-{
-  xbt_fifo_remove(mailbox->tasks, task);
-}
-
 int MSG_mailbox_is_empty(msg_mailbox_t mailbox)
 {
-  return (NULL == xbt_fifo_get_first_item(mailbox->tasks));
-}
-
-m_task_t MSG_mailbox_pop_head(msg_mailbox_t mailbox)
-{
-  return (m_task_t) xbt_fifo_shift(mailbox->tasks);
+  return (NULL == SIMIX_rdv_get_head(mailbox->rdv));
 }
 
 m_task_t MSG_mailbox_get_head(msg_mailbox_t mailbox)
 {
-  xbt_fifo_item_t item;
-
-  if (!(item = xbt_fifo_get_first_item(mailbox->tasks)))
-    return NULL;
+  smx_comm_t comm = SIMIX_rdv_get_head(mailbox->rdv);
 
-  return (m_task_t) xbt_fifo_get_item_content(item);
-}
-
-
-m_task_t MSG_mailbox_get_first_host_task(msg_mailbox_t mailbox, m_host_t host)
-{
-  m_task_t task = NULL;
-  xbt_fifo_item_t item = NULL;
-
-  xbt_fifo_foreach(mailbox->tasks, item, task, m_task_t)
-    if (task->simdata->source == host) {
-    xbt_fifo_remove_item(mailbox->tasks, item);
-    return task;
-  }
-
-  return NULL;
+  if(!comm)
+    return NULL; 
+  
+  return (m_task_t)SIMIX_communication_get_data(comm);
 }
 
 int
 MSG_mailbox_get_count_host_waiting_tasks(msg_mailbox_t mailbox, m_host_t host)
 {
-  m_task_t task = NULL;
-  xbt_fifo_item_t item = NULL;
-  int count = 0;
-
-  xbt_fifo_foreach(mailbox->tasks, item, task, m_task_t) {
-    if (task->simdata->source == host)
-      count++;
-  }
-
-  return count;
+  return SIMIX_rdv_get_count_waiting_comm (mailbox->rdv, host->simdata->smx_host);
 }
 
 void MSG_mailbox_set_cond(msg_mailbox_t mailbox, smx_cond_t cond)
@@ -121,25 +83,13 @@ const char *MSG_mailbox_get_alias(msg_mailbox_t mailbox)
   return mailbox->alias;
 }
 
-const char *MSG_mailbox_get_hostname(msg_mailbox_t mailbox)
-{
-  return mailbox->hostname;
-}
-
-void MSG_mailbox_set_hostname(msg_mailbox_t mailbox, const char *hostname)
-{
-  mailbox->hostname = xbt_strdup(hostname);
-}
-
 msg_mailbox_t MSG_mailbox_get_by_alias(const char *alias)
 {
 
   msg_mailbox_t mailbox = xbt_dict_get_or_null(msg_mailboxes, alias);
 
-  if (!mailbox) {
+  if (!mailbox)
     mailbox = MSG_mailbox_new(alias);
-    MSG_mailbox_set_hostname(mailbox, MSG_host_self()->name);
-  }
 
   return mailbox;
 }
@@ -155,308 +105,112 @@ msg_mailbox_t MSG_mailbox_get_by_channel(m_host_t host, m_channel_t channel)
 }
 
 MSG_error_t
-MSG_mailbox_get_task_ext(msg_mailbox_t mailbox, m_task_t * task,
-                         m_host_t host, double timeout)
+MSG_mailbox_get_task_ext(msg_mailbox_t mailbox, m_task_t *task, m_host_t host,
+                         double timeout)
 {
-  m_process_t process = MSG_process_self();
-  m_task_t t = NULL;
-  m_host_t h = NULL;
-  simdata_task_t t_simdata = NULL;
-  simdata_host_t h_simdata = NULL;
-  double start_time = SIMIX_get_clock();
+  xbt_ex_t e;
+  MSG_error_t ret = MSG_OK;
+  smx_comm_t comm;
 
-  smx_cond_t cond = NULL;       //conditional wait if the task isn't on the channel yet
+  /* We no longer support getting a task from a specific host */
+  if (host) THROW_UNIMPLEMENTED;
 
   CHECK_HOST();
 
+  memset(&comm,0,sizeof(comm));
+
+  /* Kept for compatibility with older implementation */
+  xbt_assert1(!MSG_mailbox_get_cond(mailbox),
+              "A process is already blocked on this channel %s", 
+              MSG_mailbox_get_alias(mailbox));
+
   /* Sanity check */
   xbt_assert0(task, "Null pointer for the task storage");
 
   if (*task)
-    CRITICAL0
-      ("MSG_task_get() was asked to write in a non empty task struct.");
-
-  /* Get the task */
-  h = MSG_host_self();
-  h_simdata = h->simdata;
-
-  SIMIX_mutex_lock(h->simdata->mutex);
+    CRITICAL0("MSG_task_get() was asked to write in a non empty task struct.");
 
-  if (MSG_mailbox_get_cond(mailbox)) {
-    CRITICAL1("A process is already blocked on the channel %s",
-              MSG_mailbox_get_alias(mailbox));
-    SIMIX_cond_display_info(MSG_mailbox_get_cond(mailbox));
-    xbt_die("Go fix your code!");
+  /* Try to receive it by calling SIMIX network layer */
+  TRY{
+    SIMIX_network_recv(mailbox->rdv, timeout, task, NULL, &comm);
+    //INFO2("Got task %s from %s",(*task)->name,mailbox->alias);
+    (*task)->simdata->refcount--;
   }
-
-  while (1) {
-    /* if the mailbox is empty (has no task */
-    if (!MSG_mailbox_is_empty(mailbox)) {
-      if (!host) {
-        /* pop the head of the mailbox */
-        t = MSG_mailbox_pop_head(mailbox);
+  CATCH(e){
+    switch(e.category){
+      case host_error:
+        ret = MSG_HOST_FAILURE;
         break;
-      } else {
-        /* get the first task of the host */
-        if ((t = MSG_mailbox_get_first_host_task(mailbox, host)))
-          break;
-      }
-    }
-
-    if ((timeout > 0) && (SIMIX_get_clock() - start_time >= timeout)) {
-      SIMIX_mutex_unlock(h->simdata->mutex);
-      MSG_mailbox_set_cond(mailbox, NULL);
-      SIMIX_cond_destroy(cond);
-      MSG_RETURN(MSG_TRANSFER_FAILURE);
-    }
-
-    if (!cond) {
-      cond = SIMIX_cond_init();
-      MSG_mailbox_set_cond(mailbox, cond);
-    }
-
-    if (timeout > 0)
-      SIMIX_cond_wait_timeout(cond, h->simdata->mutex, timeout - start_time);
-    else
-      SIMIX_cond_wait(MSG_mailbox_get_cond(mailbox), h->simdata->mutex);
-
-    if (SIMIX_host_get_state(h_simdata->smx_host) == 0) {
-      SIMIX_mutex_unlock(h->simdata->mutex);
-      MSG_mailbox_set_cond(mailbox, NULL);
-      SIMIX_cond_destroy(cond);
-      MSG_RETURN(MSG_HOST_FAILURE);
-    }
-  }
-
-
-  DEBUG1("OK, got a task (%s)", t->name);
-  /* clean conditional */
-  if (cond) {
-    MSG_mailbox_set_cond(mailbox, NULL);
-    SIMIX_cond_destroy(cond);
-  }
-
-  SIMIX_mutex_unlock(h->simdata->mutex);
-
-  t_simdata = t->simdata;
-  t_simdata->receiver = process;
-  *task = t;
-
-  SIMIX_mutex_lock(t_simdata->mutex);
-
-  /* Transfer */
-  /* create SIMIX action to the communication */
-  t_simdata->comm =
-    SIMIX_action_communicate(t_simdata->sender->simdata->m_host->simdata->
-                             smx_host,
-                             process->simdata->m_host->simdata->smx_host,
-                             t->name, t_simdata->message_size,
-                             t_simdata->rate);
-
-  /* This is a hack. We know that both the receiver and the sender will
-     need to look at the content of t_simdata->comm. And it needs to be
-     destroyed. However, we don't known whether the receiver or the sender
-     will get to it first. So by setting with refcount to 2 we can enforce
-     that things happen correctly. An alternative would be to only do ++ and
-     -- on this refcount and to sprinkle them judiciously throughout the code,
-     which appears perhaps worse? Or perhaps the refcount field of
-     task->simdata can be used for this? At any rate, this will do for now */
-  t_simdata->comm->refcount = 2;
-
-  /* if the process is suspend, create the action but stop its execution, it will be restart when the sender process resume */
-  if (MSG_process_is_suspended(t_simdata->sender)) {
-    DEBUG1("Process sender (%s) suspended", t_simdata->sender->name);
-    SIMIX_action_set_priority(t_simdata->comm, 0);
-  }
-
-  process->simdata->waiting_task = t;
-  SIMIX_register_action_to_condition(t_simdata->comm, t_simdata->cond);
-
-  while (1) {
-    SIMIX_cond_wait(t_simdata->cond, t_simdata->mutex);
-
-    if (SIMIX_action_get_state(t_simdata->comm) != SURF_ACTION_RUNNING)
-      break;
-  }
-
-  SIMIX_unregister_action_to_condition(t_simdata->comm, t_simdata->cond);
-  process->simdata->waiting_task = NULL;
-
-  /* If sender still around (it didn't free the comm yet), note that it's not waiting anymore */
-  if (t_simdata->comm->refcount == 2) {
-    t->simdata->sender->simdata->waiting_task = NULL;
-  }
-
-  /* for this process, don't need to change in get function */
-  SIMIX_mutex_unlock(t_simdata->mutex);
-
-
-  if (SIMIX_action_get_state(t_simdata->comm) == SURF_ACTION_DONE) {
-    if (t_simdata->comm->refcount == 1) {
-      SIMIX_action_destroy(t_simdata->comm);
-      t_simdata->comm = NULL;
-    } else {
-      t_simdata->comm->refcount--;
-    }
-    t_simdata->refcount--;
-    MSG_RETURN(MSG_OK);
-  } else if (SIMIX_host_get_state(h_simdata->smx_host) == 0) {
-    if (t_simdata->comm->refcount == 1) {
-      SIMIX_action_destroy(t_simdata->comm);
-      t_simdata->comm = NULL;
-    } else {
-      t_simdata->comm->refcount--;
-    }
-    t_simdata->refcount--;
-    MSG_RETURN(MSG_HOST_FAILURE);
-  } else {
-    if (t_simdata->comm->refcount == 1) {
-      SIMIX_action_destroy(t_simdata->comm);
-      t_simdata->comm = NULL;
-    } else {
-      t_simdata->comm->refcount--;
+      case network_error:
+        ret = MSG_TRANSFER_FAILURE;
+        break;
+      case timeout_error:
+        ret = MSG_TIMEOUT_FAILURE;
+        break;      
+      default:
+        xbt_die("Unhandled SIMIX network exception");
     }
-    t_simdata->refcount--;
-    MSG_RETURN(MSG_TRANSFER_FAILURE);
+    xbt_ex_free(e);        
   }
+  
+  MSG_RETURN(ret);        
 }
 
 MSG_error_t
 MSG_mailbox_put_with_timeout(msg_mailbox_t mailbox, m_task_t task,
                              double timeout)
 {
-  m_process_t process = MSG_process_self();
-  const char *hostname;
+  xbt_ex_t e;
+  MSG_error_t ret = MSG_OK;
   simdata_task_t t_simdata = NULL;
-  m_host_t local_host = NULL;
-  m_host_t remote_host = NULL;
-  smx_cond_t cond = NULL;
-
+  m_process_t process = MSG_process_self();
+  
   CHECK_HOST();
 
+  /* Prepare the task to send */
   t_simdata = task->simdata;
   t_simdata->sender = process;
-  t_simdata->source = MSG_process_get_host(process);
+  t_simdata->source = MSG_host_self();
 
   xbt_assert0(t_simdata->refcount == 1,
               "This task is still being used somewhere else. You cannot send it now. Go fix your code!");
 
-  t_simdata->comm = NULL;
-
   t_simdata->refcount++;
-  local_host = ((simdata_process_t) process->simdata)->m_host;
   msg_global->sent_msg++;
 
-  /* get the host name containing the mailbox */
-  hostname = MSG_mailbox_get_hostname(mailbox);
-
-  remote_host = MSG_get_host_by_name(hostname);
-
-  if (!remote_host)
-    THROW1(not_found_error, 0, "Host %s not fount", hostname);
-
-
-  DEBUG4("Trying to send a task (%g kB) from %s to %s on the channel %s",
-         t_simdata->message_size / 1000, local_host->name,
-         remote_host->name, MSG_mailbox_get_alias(mailbox));
-
-  SIMIX_mutex_lock(remote_host->simdata->mutex);
-
-  /* put the task in the mailbox */
-  xbt_fifo_push(mailbox->tasks, task);
-
-  if ((cond = MSG_mailbox_get_cond(mailbox))) {
-    DEBUG0("Somebody is listening. Let's wake him up!");
-    SIMIX_cond_signal(cond);
-  }
-
-  SIMIX_mutex_unlock(remote_host->simdata->mutex);
-
-  SIMIX_mutex_lock(t_simdata->mutex);
-
   process->simdata->waiting_task = task;
+  
+  /* Try to send it by calling SIMIX network layer */
+  TRY{
+    /* Kept for semantical compatibility with older implementation */
+    if(mailbox->cond)
+      SIMIX_cond_signal(mailbox->cond);
+
+    SIMIX_network_send(mailbox->rdv, t_simdata->message_size, t_simdata->rate,
+                       timeout, task, sizeof(void*), &t_simdata->comm, task);
+  }
 
-  if (timeout > 0) {
-    xbt_ex_t e;
-    double time;
-    double time_elapsed;
-    time = SIMIX_get_clock();
-
-    TRY {
-      /*verify if the action that ends is the correct. Call the wait_timeout with the new time. If the timeout occurs, an exception is raised */
-      while (1) {
-        time_elapsed = SIMIX_get_clock() - time;
-        SIMIX_cond_wait_timeout(t_simdata->cond, t_simdata->mutex,
-                                timeout - time_elapsed);
-
-        if ((t_simdata->comm != NULL)
-            && (SIMIX_action_get_state(t_simdata->comm) !=
-                SURF_ACTION_RUNNING))
-          break;
-      }
-    }
-    CATCH(e) {
-      if (e.category == timeout_error) {
-        xbt_ex_free(e);
-        /* verify if the timeout happened and the communication didn't started yet */
-        if (t_simdata->comm == NULL) {
-          process->simdata->waiting_task = NULL;
-
-          /* remove the task from the mailbox */
-          MSG_mailbox_remove(mailbox, task);
-
-          if (t_simdata->receiver && t_simdata->receiver->simdata) {    /* receiver still around */
-            t_simdata->receiver->simdata->waiting_task = NULL;
-          }
-
-          SIMIX_mutex_unlock(t_simdata->mutex);
-          MSG_RETURN(MSG_TRANSFER_FAILURE);
-        }
-      } else {
-        RETHROW;
-      }
-    }
-  } else {
-    while (1) {
-      SIMIX_cond_wait(t_simdata->cond, t_simdata->mutex);
-
-      if (SIMIX_action_get_state(t_simdata->comm) != SURF_ACTION_RUNNING)
+  CATCH(e){
+    switch(e.category){
+      case host_error:
+        ret = MSG_HOST_FAILURE;
         break;
+      case network_error:
+        ret = MSG_TRANSFER_FAILURE;
+        break;
+      case timeout_error:
+        ret = MSG_TIMEOUT_FAILURE;
+        break;
+      default:
+        xbt_die("Unhandled SIMIX network exception");
     }
-  }
+    xbt_ex_free(e);
 
-  DEBUG1("Action terminated %s", task->name);
-  process->simdata->waiting_task = NULL;
-
-  if (t_simdata->comm->refcount == 2) { //receiver didn't free it yet: he's still around
-    t_simdata->receiver->simdata->waiting_task = NULL;
+    /* Decrement the refcount only on failure */
+    t_simdata->refcount--;
   }
 
-  SIMIX_mutex_unlock(task->simdata->mutex);
-
-  if (SIMIX_action_get_state(t_simdata->comm) == SURF_ACTION_DONE) {
-    if (t_simdata->comm->refcount == 1) {
-      SIMIX_action_destroy(t_simdata->comm);
-      t_simdata->comm = NULL;
-    } else {
-      t_simdata->comm->refcount--;
-    }
-    MSG_RETURN(MSG_OK);
-  } else if (SIMIX_host_get_state(local_host->simdata->smx_host) == 0) {
-    if (t_simdata->comm->refcount == 1) {
-      SIMIX_action_destroy(t_simdata->comm);
-      t_simdata->comm = NULL;
-    } else {
-      t_simdata->comm->refcount--;
-    }
-    MSG_RETURN(MSG_HOST_FAILURE);
-  } else {
-    if (t_simdata->comm->refcount == 1) {
-      SIMIX_action_destroy(t_simdata->comm);
-      t_simdata->comm = NULL;
-    } else {
-      t_simdata->comm->refcount--;
-    }
-    MSG_RETURN(MSG_TRANSFER_FAILURE);
-  }
+  process->simdata->waiting_task = NULL;
+   
+  MSG_RETURN(ret);        
 }