Fix for BUG#2477 "Slave stop with error after master reboot if use HEAP tables":

when we open the HEAP table for the first time since server restart,
in hp_open(), we set a flag to propagate this info to the handler level
which then writes a DELETE FROM this_heap_table to the binlog.
It is not a perfect solution for the bug, because between the server start and 
the first open of the table, the slave still had old data in his table so
a SELECT on the slave may show wrong content. But if there is a --init-file
to populate the HEAP table on master as startup, then this is a safe fix
(I'll put a note about init-file in the HEAP section of the manual).

3 files changed, 47 insertions, 11 deletions

diff --git a/sql/ha_heap.cc b/sql/ha_heap.cc
index 2edc3b1478e..eb4bf517374 100644
--- a/sql/ha_heap.cc
+++ b/sql/ha_heap.cc
@@ -222,6 +222,7 @@ void ha_heap::info(uint flag)
   index_file_length=info.index_length;
   max_data_file_length= info.max_records* info.reclength;
   delete_length= info.deleted * info.reclength;
+  implicit_emptied= info.implicit_emptied;
 }
 
 int ha_heap::extra(enum ha_extra_function operation)
diff --git a/sql/handler.h b/sql/handler.h
index 8e72267c337..60edf539e2c 100644
--- a/sql/handler.h
+++ b/sql/handler.h
@@ -213,6 +213,7 @@ public:
   uint raid_type,raid_chunks;
   FT_INFO *ft_handler;
   bool  auto_increment_column_changed;
+  bool implicit_emptied;                /* Can be !=0 only if HEAP */
 
   handler(TABLE *table_arg) :table(table_arg),
     ref(0), data_file_length(0), max_data_file_length(0), index_file_length(0),
@@ -221,7 +222,7 @@ public:
     create_time(0), check_time(0), update_time(0),
     key_used_on_scan(MAX_KEY), active_index(MAX_REF_PARTS),
     ref_length(sizeof(my_off_t)), block_size(0),
-    raid_type(0), ft_handler(0)
+    raid_type(0), ft_handler(0), implicit_emptied(0)
     {}
   virtual ~handler(void) {}
   int ha_open(const char *name, int mode, int test_if_locked);
diff --git a/sql/sql_base.cc b/sql/sql_base.cc
index 9937336fb4a..40153ad847e 100644
--- a/sql/sql_base.cc
+++ b/sql/sql_base.cc
@@ -550,6 +550,11 @@ void close_temporary_tables(THD *thd)
   query_buf_size= 50;   // Enough for DROP ... TABLE
 
   for (table=thd->temporary_tables ; table ; table=table->next)
+    /*
+      We are going to add 4 ` around the db/table names, so 1 does not look
+      enough; indeed it is enough, because table->key_length is greater (by 8,
+      because of server_id and thread_id) than db||table.
+    */
     query_buf_size+= table->key_length+1;
 
   if ((query = alloc_root(&thd->mem_root, query_buf_size)))
@@ -566,8 +571,8 @@ void close_temporary_tables(THD *thd)
         Here we assume table_cache_key always starts
         with \0 terminated db name
       */
-      end = strxmov(end,"`",table->table_cache_key,"`",
-                    ".`",table->real_name,"`,", NullS);
+      end = strxmov(end,"`",table->table_cache_key,"`.`",
+                    table->real_name,"`,", NullS);
     }
     next=table->next;
     close_temporary(table);
@@ -1331,8 +1336,7 @@ static int open_unireg_entry(THD *thd, TABLE *entry, const char *db,
       }
     }
     pthread_mutex_unlock(&LOCK_open);
-    thd->net.last_error[0]=0;				// Clear error message
-    thd->net.last_errno=0;
+    thd->clear_error();
     error=0;
     if (openfrm(path,alias,
 		(uint) (HA_OPEN_KEYFILE | HA_OPEN_RNDFILE | HA_GET_INDEX |
@@ -1343,8 +1347,7 @@ static int open_unireg_entry(THD *thd, TABLE *entry, const char *db,
 	(entry->file->is_crashed() && entry->file->check_and_repair(thd)))
     {
       /* Give right error message */
-      thd->net.last_error[0]=0;
-      thd->net.last_errno=0;
+      thd->clear_error();
       my_error(ER_NOT_KEYFILE, MYF(0), name, my_errno);
       sql_print_error("Error: Couldn't repair table: %s.%s",db,name);
       if (entry->file)
@@ -1352,16 +1355,47 @@ static int open_unireg_entry(THD *thd, TABLE *entry, const char *db,
       error=1;
     }
     else
-    {
-      thd->net.last_error[0]=0;			// Clear error message
-      thd->net.last_errno=0;
-    }
+      thd->clear_error();
     pthread_mutex_lock(&LOCK_open);
     unlock_table_name(thd,&table_list);
 
     if (error)
       goto err;
   }
+  /*
+    If we are here, there was no fatal error (but error may be still
+    unitialized).
+  */
+  if (unlikely(entry->file->implicit_emptied))
+  {
+    entry->file->implicit_emptied= 0;
+    if (mysql_bin_log.is_open())
+    {
+      char *query, *end;
+      uint query_buf_size= 20 + 2*NAME_LEN + 1;
+      if ((query= (char*)my_malloc(query_buf_size,MYF(MY_WME))))
+      {
+        end = strxmov(strmov(query, "DELETE FROM `"),
+                      db,"`.`",name,"`", NullS);
+        Query_log_event qinfo(thd, query, (ulong)(end-query), 0);
+        mysql_bin_log.write(&qinfo);
+        my_free(query, MYF(0));
+      }
+      else
+      {
+        /*
+          As replication is maybe going to be corrupted, we need to warn the
+          DBA on top of warning the client (which will automatically be done
+          because of MYF(MY_WME) in my_malloc() above).
+        */
+        sql_print_error("Error: when opening HEAP table, could not allocate \
+memory to write 'DELETE FROM `%s`.`%s`' to the binary log",db,name);
+        if (entry->file)
+          closefrm(entry);
+        goto err;
+      }
+    }
+  }
   DBUG_RETURN(0);
 err:
   DBUG_RETURN(1);