Replication: various small fixes specific to the new binlog format of 5.0

(including one which may explain autobuild's failure of yesterday)


client/mysqlbinlog.cc:
  - In mysqlbinlog, we should not reset the Format event when we see Rotate. If a binlog started with a Format event, it is not going to switch later to 4.0 format.
  I had already did the same fix in Rotate_log_event::exec_event() in replication.
  - Fix for a merge bug.
sql/log_event.cc:
  An event with an uninited catalog (read from a 4.x server) is not the same as an event with a NULL catalog
  (5.0 server which did not specify catalog), the difference is that they are not in the same format;
  so I introduce a way to know if the catalog has been inited or not. This fixes a rpl_trunc_binlog failure
  I had.
  When we leave Load_log_event::exec_event(), we must reset thd->catalog to 0, like we already do
  in Query_log_event::exec_event(). This fixes a Valgrind error which popped in rpl_charset (which may
  be what caused autobuild to crash yesterday).
  And a fix for event's parsing (the position was always computed right because start_dup==end is always true
  and will until we add new string members to Query_log_event.
sql/log_event.h:
  catalog_len changed from uint to int to allow -1, which means "not inited"
  (I preferred to do it like this rather than create a new bool var Query_log_event::catalog_inited
  like we have in Query_log_event::sql_mode_inited; that's because catalog will not use the whole range of int,
  so it's allowed to pick -1 as a special value and have only one var.
sql/slave.cc:
  comments

1 files changed, 1 insertions, 1 deletions

diff --git a/sql/log_event.h b/sql/log_event.h
index 5c1a143f407..c9cce1d40ea 100644
--- a/sql/log_event.h
+++ b/sql/log_event.h
@@ -604,7 +604,7 @@ public:
      concerned) from here.
   */
 
-  uint catalog_len;				/* <= 255 char */
+  int catalog_len;				// <= 255 char; -1 means uninited
 
   /*
     We want to be able to store a variable number of N-bit status vars: