handler.cc:

  If the autocommit is on, let handler.cc commit or rollback the whole transaction at an updating SQL statement end. This probably fixes bug number 578. The problem was that when explicit LOCK TABLES is used, then the lock count method in autocommit does not work.

sql/handler.cc

@@ -208,23 +208,45 @@ void ha_close_connection(THD* thd)
 }
 
 /*
-  This is used to commit or rollback a single statement depending
-  on the value of error
+  This is used to commit or rollback a single statement depending on the value
+  of error. If the autocommit is on, then we will commit or rollback the whole
+  transaction (= the statement). The autocommit mechanism built into handlers
+  is based on counting locks, but if the user has used LOCK TABLES then that
+  mechanism does not know to do the commit.
 */
 
 int ha_autocommit_or_rollback(THD *thd, int error)
 {
+  bool do_autocommit=FALSE;
+
   DBUG_ENTER("ha_autocommit_or_rollback");
 #ifdef USING_TRANSACTIONS
+
+  if (!(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)))
+    do_autocommit=TRUE;  /* We can commit or rollback the whole transaction */
+
   if (opt_using_transactions)
   {
     if (!error)
+    {
+      if (do_autocommit)
+      {
+	if (ha_commit(thd))
+	  error=1;
+      }
+      else
       {
         if (ha_commit_stmt(thd))
 	  error=1;
       }
+    }
+    else
+    {
+      if (do_autocommit)
+	(void) ha_rollback(thd);
       else
         (void) ha_rollback_stmt(thd);
+    }
 
     thd->variables.tx_isolation=thd->session_tx_isolation;
   }