A fix of the fix for BUG#1209 (was: SELECT DATABASE() still shows dropped database).

Don't free the client's selected db (thd->db) when it has been DROPped,
if this is a slave thread; the x_free() was causing a lot
of various bugs in rpl_loaddata_rule_m (garbage characters, segfault, hangs in
other threads).
A small post-merge fix (rli->inside_transaction exists in 4.0 only).


sql/log_event.cc:
  Post-merge fix (rli->inside_transaction existed in 4.0 only)
sql/sql_db.cc:
  Don't free the client's selected db (thd->db) when it has been DROPped,
  if this is a slave thread; the x_free() was causing a lot
  of various bugs in rpl_loaddata_rule_m which does a DROP DATABASE 
  (garbage characters, segfault, hangs in other threads).

sql/log_event.cc

@@ -1876,19 +1876,22 @@ int Rotate_log_event::exec_event(struct st_relay_log_info* rli)
 
   pthread_mutex_lock(&rli->data_lock);
 
-#ifdef TO_BE_CHECKED_BY_GUILHEM
-  if (rli->inside_transaction)
+  if (thd->options & OPTION_BEGIN)
   {
     slave_print_error(rli, 0,
+                      opt_using_transactions ?
                       "\
 There is an unfinished transaction in the relay log (could find neither \
 COMMIT nor ROLLBACK in the relay log); It could be that the master died while \
 writing the transaction to its binary log. Now the slave is rolling back the \
-transaction.");
+transaction." : 
+                      "\
+There is an unfinished transaction in the relay log (could find neither \
+COMMIT nor ROLLBACK in the relay log); It could be that the master died while \
+writing the transaction to its binary log.");
     pthread_mutex_unlock(&rli->data_lock);
     DBUG_RETURN(1);
   }
-#endif
 
   memcpy(log_name, new_log_ident, ident_len+1);
   rli->notify_group_master_log_name_update();

sql/sql_db.cc

@@ -391,8 +391,30 @@ int mysql_rm_db(THD *thd,char *db,bool if_exists, bool silent)
 
 exit:
   start_waiting_global_read_lock(thd);
+  /*
+    If this database was the client's selected database, we silently change the
+    client's selected database to nothing (to have an empty SELECT DATABASE() in
+    the future). For this we free() thd->db and set it to 0. But we don't do
+    free() for the slave thread. Indeed, doing a x_free() on it leads to nasty
+    problems (i.e. long painful debugging) because in this thread, thd->db is
+    the same as data_buf and db of the Query_log_event which is dropping the
+    database. So if you free() thd->db, you're freeing data_buf. You set thd->db
+    to 0 but not data_buf (thd->db and data_buf are two distinct pointers which
+    point to the same place). Then in ~Query_log_event(), we have
+    'if (data_buf) free(data_buf)'
+    data_buf is !=0 so this makes a DOUBLE free().
+    Side effects of this double free() are, randomly (depends on the machine),
+    when the slave is replicating a DROP DATABASE: 
+    - garbage characters in the error message:
+    "Error 'Can't drop database 'test2'; database doesn't exist' on query
+    'h4zI'"
+    - segfault
+    - hang in "free(vio)" (yes!) in the I/O or SQL slave threads (so slave
+    server hangs at shutdown etc).
+  */
   if (thd->db && !strcmp(thd->db, db))
   {
+    if (!(thd->slave_thread)) /* a slave thread will free it itself */
       x_free(thd->db);
     thd->db= 0; 
   }