Bug fixed: use a timeout on receive.

[loba.git] / communicator.cpp

diff --git a/communicator.cpp b/communicator.cpp
index 94cdb04160337353a0a1bab8b2797d88c0ce415d..90cf990ebb07c31f70d30f89e6723a94bd78bdca 100644 (file)
--- a/communicator.cpp
+++ b/communicator.cpp
@@ -21,13 +21,10 @@ std::string message::to_string()
     return oss.str();
 }
 
-const int communicator::send_count_before_flush = 4;
-
 communicator::communicator()
     : host((hostdata* )MSG_host_get_data(MSG_host_self()))
     , mutex(xbt_mutex_init())
     , cond(xbt_cond_init())
-    , send_counter(0)
     , ctrl_task(NULL)
     , ctrl_comm(NULL)
     , data_task(NULL)
@@ -37,6 +34,7 @@ communicator::communicator()
     receiver_process =
         MSG_process_create("receiver", communicator::receiver_wrapper,
                            this, MSG_host_self());
+    xbt_cond_wait(cond, mutex); // wait for the receiver to be ready
     xbt_mutex_release(mutex);
 }
 
@@ -52,10 +50,11 @@ communicator::~communicator()
     task = MSG_task_create("finalize", 0.0, 0, NULL);
     MSG_task_send(task, get_data_mbox());
 
-    DEBUG0("wait for receiver to terminate");
     xbt_mutex_acquire(mutex);
-    while (receiver_process)
+    while (receiver_process) {
+        DEBUG0("waiting for receiver to terminate");
         xbt_cond_wait(cond, mutex);
+    }
     xbt_mutex_release(mutex);
 
     if (ctrl_comm)
@@ -82,20 +81,29 @@ void communicator::send(const char* dest, message* msg)
     m_task_t task = MSG_task_create("message", 0.0, msg_size, msg);    
     msg_comm_t comm = MSG_task_isend(task, dest);
     sent_comm.push_back(comm);
-
-    if (++send_counter >= send_count_before_flush) {
-        flush(false);
-        send_counter = 0;
-    }
 }
 
-bool communicator::recv(message*& msg, m_host_t& from, bool wait)
+bool communicator::recv(message*& msg, m_host_t& from, double timeout)
 {
-    if (wait) {
-        DEBUG0("suspend main process on recv");
+    if (timeout != 0) {
+        volatile double deadline =
+            timeout > 0 ? MSG_get_clock() + timeout : 0.0;
         xbt_mutex_acquire(mutex);
-        while (received.empty())
-            xbt_cond_wait(cond, mutex);
+        while (received.empty() && (!deadline || deadline > MSG_get_clock())) {
+            xbt_ex_t e;
+            DEBUG0("waiting for a message to come");
+            TRY {
+                if (deadline)
+                    xbt_cond_timedwait(cond, mutex, deadline - MSG_get_clock());
+                else
+                    xbt_cond_wait(cond, mutex);
+            }
+            CATCH (e) {
+                if (e.category != timeout_error)
+                    RETHROW;
+                xbt_ex_free(e);
+            }
+        }
         xbt_mutex_release(mutex);
     }
 
@@ -162,6 +170,11 @@ int communicator::receiver()
 {
     ctrl_comm = MSG_task_irecv(&ctrl_task, get_ctrl_mbox());
     data_comm = MSG_task_irecv(&data_task, get_data_mbox());
+    DEBUG0("receiver ready");
+    xbt_mutex_acquire(mutex);
+    xbt_cond_signal(cond);      // signal master that we are ready
+    xbt_mutex_release(mutex);
+
     xbt_dynar_t comms = xbt_dynar_new(sizeof(msg_comm_t), NULL);
     while (ctrl_comm || data_comm) {
 
@@ -175,7 +188,9 @@ int communicator::receiver()
         if (ctrl_comm && comm_test_n_destroy(ctrl_comm)) {
             if (strcmp(MSG_task_get_name(ctrl_task), "finalize")) {
                 DEBUG0("received message from ctrl");
+                xbt_mutex_acquire(mutex);
                 received.push(ctrl_task);
+                xbt_mutex_release(mutex);
                 ctrl_task = NULL;
                 ctrl_comm = MSG_task_irecv(&ctrl_task, get_ctrl_mbox());
             } else {
@@ -189,7 +204,9 @@ int communicator::receiver()
         if (data_comm && comm_test_n_destroy(data_comm)) {
             if (strcmp(MSG_task_get_name(data_task), "finalize")) {
                 DEBUG0("received message from data");
+                xbt_mutex_acquire(mutex);
                 received.push(data_task);
+                xbt_mutex_release(mutex);
                 data_task = NULL;
                 data_comm = MSG_task_irecv(&data_task, get_data_mbox());
             } else {
@@ -200,7 +217,8 @@ int communicator::receiver()
             }
         }
         xbt_mutex_acquire(mutex);
-        xbt_cond_signal(cond);
+        if (!received.empty())
+            xbt_cond_signal(cond);
         xbt_mutex_release(mutex);
     }
     xbt_dynar_free(&comms);