Also display the pid on backtraces

[simgrid.git] / src / gras / Msg / msg.c

diff --git a/src/gras/Msg/msg.c b/src/gras/Msg/msg.c
index 5b02e5c..61a0701 100644 (file)
--- a/src/gras/Msg/msg.c
+++ b/src/gras/Msg/msg.c
@@ -8,6 +8,7 @@
  * under the terms of the license (GNU LGPL) which comes with this package. */
 
 #include "xbt/ex.h"
+#include "xbt/ex_interface.h"
 #include "gras/Msg/msg_private.h"
 #include "gras/Virtu/virtu_interface.h"
 #include "gras/DataDesc/datadesc_interface.h"
@@ -292,7 +293,7 @@ gras_msg_wait_ext(double           timeout,
   while (1) {
     memset(&msg,sizeof(msg),0);
 
-    msg.expe = gras_trp_select(timeout - now + start);
+    msg.expe = gras_trp_select(timeout ? timeout - now + start : 0);
     gras_msg_recv(msg.expe, &msg);
     DEBUG0("Got a message from the socket");
 
@@ -373,7 +374,7 @@ gras_msg_handle(double timeOut) {
   
   double          untiltimer;
    
-  int             cpt;
+  int             cpt, ran_ok;
 
   s_gras_msg_t    msg;
 
@@ -388,7 +389,7 @@ gras_msg_handle(double timeOut) {
   VERB1("Handling message within the next %.2fs",timeOut);
   
   untiltimer = gras_msg_timer_handle();
-  DEBUG2("[%.0f] Next timer in %f sec", gras_os_time(), untiltimer);
+  DEBUG1("Next timer in %f sec", untiltimer);
   if (untiltimer == 0.0) {
      /* A timer was already elapsed and handled */
      return;
@@ -419,7 +420,8 @@ gras_msg_handle(double timeOut) {
       TRY {
        /* FIXME: if not the right kind, queue it and recall ourself or goto >:-) */
        gras_msg_recv(msg.expe, &msg);
-       DEBUG0("Received a msg from the socket");
+       DEBUG1("Received a msg from the socket kind:%s",
+              e_gras_msg_kind_names[msg.kind]);
     
       } CATCH(e) {
        RETHROW1("Error caught while receiving a message on select()ed socket %p: %s",
@@ -447,7 +449,8 @@ gras_msg_handle(double timeOut) {
        
      } else {
        /* select timeouted, and no timer elapsed. Nothing to do */
-       THROW0(timeout_error, 0, "No new message or timer");
+       THROW1(timeout_error, 0, "No new message or timer (delay was %f)",
+             timeOut);
      }
      
   }
@@ -471,38 +474,47 @@ gras_msg_handle(double timeOut) {
   ctx.ID = msg.ID;
   ctx.msgtype = msg.type;
 
-  switch (msg.type->kind) {
+  switch (msg.kind) {
   case e_gras_msg_kind_oneway:
   case e_gras_msg_kind_rpccall:
+    ran_ok=0;
     TRY {
       xbt_dynar_foreach(list->cbs,cpt,cb) { 
-       VERB3("Use the callback #%d (@%p) for incomming msg %s",
-             cpt+1,cb,msg.type->name);
-       if ((*cb)(&ctx,msg.payl)) {
-         /* cb handled the message */
-         free(msg.payl);
-         return;
+       if (!ran_ok) {
+         VERB3("Use the callback #%d (@%p) for incomming msg %s",
+               cpt+1,cb,msg.type->name);
+         if ((*cb)(&ctx,msg.payl)) {
+           /* cb handled the message */
+           free(msg.payl);
+           ran_ok = 1;
+         }
        }
       }
     } CATCH(e) {
+      free(msg.payl);
       if (msg.type->kind == e_gras_msg_kind_rpccall) {
        /* The callback raised an exception, propagate it on the network */
-       e.host = (char*)gras_os_myname();
-#ifdef HAVE_EXECINFO_H
-       e.bt_strings = backtrace_symbols (e.bt, e.used);
-#endif
-       gras_msg_send_ext(msg.expe, e_gras_msg_kind_rpcerror , msg.ID, msg.type, &e);
-       e.host = NULL;
-       INFO2("RPC callback raised an exception, which were propagated back to %s:%d",
-             gras_socket_peer_name(msg.expe),  gras_socket_peer_port(msg.expe));
+       if (!e.remote) { /* the exception is born on this machine */
+         e.host = (char*)gras_os_myname();
+         xbt_ex_setup_backtrace(&e);
+       } 
+       VERB4("Propagate %s exception from '%s' RPC cb back to %s:%d",
+             (e.remote ? "remote" : "local"),
+             msg.type->name,
+             gras_socket_peer_name(msg.expe),
+             gras_socket_peer_port(msg.expe));
+       gras_msg_send_ext(msg.expe, e_gras_msg_kind_rpcerror,
+                         msg.ID, msg.type, &e);
        xbt_ex_free(e);
-       return;
+       ran_ok=1;
+      } else {
+       RETHROW0("Callback raised an exception: %s");
       }
-      RETHROW;
     }
+    if (!ran_ok)
+      THROW1(mismatch_error,0,
+            "Message '%s' refused by all registered callbacks", msg.type->name);
     /* FIXME: gras_datadesc_free not implemented => leaking the payload */
-    THROW1(mismatch_error,0,
-          "Message '%s' refused by all registered callbacks", msg.type->name);
     break;
 
 
@@ -518,7 +530,7 @@ gras_msg_handle(double timeOut) {
 
   default:
     THROW1(unknown_error,0,
-          "Cannot handle messages of kind %d yet",msg.type->kind);
+          "Cannot handle messages of kind %d yet",msg.type->kind);
   }
 
 }