This is a fix for a bug in 3.23 -> 4.0 replication: Exec_master_log_pos is always

too big by 6 bytes. So I add code to substract 6 bytes if the master is 3.23.
This is not perfect (because it won't work if the slave I/O thread has not
noticed yet that the master is 3.23), but as long as the slave I/O thread
starts Exec_master_log_pos will be ok.
It must be merged to 4.1 but not to 5.0 (or it can be, because of #if MYSQL_VERSION_ID),
because 5.0 already works if the master is 3.23 (and in a more natural way:
in 5.0 we store the end_log_pos in the binlog and relay log).
I had to move functions from slave.h to slave.cc to satisfy gcc.


sql/log_event.cc:
  make the event's length 6 bytes shorter if the master is 3.23
sql/slave.cc:
  Moving several st_relay_log_info methods out of the declaration of the struct,
  because gcc complained that 'mi' was not declared.
  Substracting 6 bytes from the event's length in inc_pos() if the master is 3.23.
sql/slave.h:
  moving several methods out of the declaration of st_relay_log_info.
  Removing 'inline', let's have the compiler decide.

3 files changed, 75 insertions, 34 deletions

diff --git a/sql/log_event.cc b/sql/log_event.cc
index a484123b4c7..cd94ee2804d 100644
--- a/sql/log_event.cc
+++ b/sql/log_event.cc
@@ -1805,10 +1805,21 @@ int Query_log_event::exec_event(struct st_relay_log_info* rli)
     position to store is of the END of the current log event.
   */
 #if MYSQL_VERSION_ID < 40100
-  rli->future_master_log_pos= log_pos + get_event_len();
+  /*
+    If the event was converted from a 3.23 format, get_event_len() has grown by
+    6 bytes (at least for most events, except LOAD DATA INFILE which is already
+    a big problem for 3.23->4.0 replication); 6 bytes is the difference between
+    the header's size in 4.0 (LOG_EVENT_HEADER_LEN) and the header's size in
+    3.23 (OLD_HEADER_LEN). Note that using mi->old_format will not help if the
+    I/O thread has not started yet.
+  */
+  rli->future_master_log_pos= log_pos + get_event_len() -
+    (rli->mi->old_format ? (LOG_EVENT_HEADER_LEN - OLD_HEADER_LEN) : 0); 
 #elif MYSQL_VERSION_ID < 50000
-  rli->future_group_master_log_pos= log_pos + get_event_len();
+  rli->future_group_master_log_pos= log_pos + get_event_len() -
+    (rli->mi->old_format ? (LOG_EVENT_HEADER_LEN - OLD_HEADER_LEN) : 0);
 #else
+  /* In 5.0 we store the end_log_pos in the relay log so no problem */
   rli->future_group_master_log_pos= log_pos;
 #endif
   clear_all_errors(thd, rli);
@@ -1960,9 +1971,11 @@ int Load_log_event::exec_event(NET* net, struct st_relay_log_info* rli,
   if (!use_rli_only_for_errors)
   {
 #if MYSQL_VERSION_ID < 40100
-    rli->future_master_log_pos= log_pos + get_event_len();
+    rli->future_master_log_pos= log_pos + get_event_len() -
+      (rli->mi->old_format ? (LOG_EVENT_HEADER_LEN - OLD_HEADER_LEN) : 0);
 #elif MYSQL_VERSION_ID < 50000
-    rli->future_group_master_log_pos= log_pos + get_event_len();
+    rli->future_group_master_log_pos= log_pos + get_event_len() -
+      (rli->mi->old_format ? (LOG_EVENT_HEADER_LEN - OLD_HEADER_LEN) : 0);
 #else
     rli->future_group_master_log_pos= log_pos;
 #endif
@@ -2133,6 +2146,11 @@ Fatal error running LOAD DATA INFILE on table '%s'. Default database: '%s'",
 int Start_log_event::exec_event(struct st_relay_log_info* rli)
 {
 
+  /*
+    If the I/O thread has not started, mi->old_format is BINLOG_FORMAT_CURRENT
+    (that's what the MASTER_INFO constructor does), so the test below is not
+    perfect at all.
+  */
   switch (rli->mi->old_format) {
   case BINLOG_FORMAT_CURRENT : 
     /* 
@@ -2427,9 +2445,11 @@ int Execute_load_log_event::exec_event(struct st_relay_log_info* rli)
   */
 
 #if MYSQL_VERSION_ID < 40100
-    rli->future_master_log_pos= log_pos + get_event_len();
+    rli->future_master_log_pos= log_pos + get_event_len() -
+      (rli->mi->old_format ? (LOG_EVENT_HEADER_LEN - OLD_HEADER_LEN) : 0);
 #elif MYSQL_VERSION_ID < 50000
-    rli->future_group_master_log_pos= log_pos + get_event_len();
+    rli->future_group_master_log_pos= log_pos + get_event_len() -
+      (rli->mi->old_format ? (LOG_EVENT_HEADER_LEN - OLD_HEADER_LEN) : 0);
 #else
     rli->future_group_master_log_pos= log_pos;
 #endif
diff --git a/sql/slave.cc b/sql/slave.cc
index d6dda473be0..9e82a521cb3 100644
--- a/sql/slave.cc
+++ b/sql/slave.cc
@@ -329,6 +329,52 @@ void init_slave_skip_errors(const char* arg)
   }
 }
 
+void st_relay_log_info::inc_pending(ulonglong val)
+{
+  pending += val;
+}
+
+/* TODO: this probably needs to be fixed */
+void st_relay_log_info::inc_pos(ulonglong val, ulonglong log_pos, bool skip_lock)
+{
+  if (!skip_lock)
+    pthread_mutex_lock(&data_lock);
+  relay_log_pos += val+pending;
+  pending = 0;
+  if (log_pos)
+#if MYSQL_VERSION_ID < 50000
+    /*
+      If the event was converted from a 3.23 format, get_event_len() has
+      grown by 6 bytes (at least for most events, except LOAD DATA INFILE
+      which is already a big problem for 3.23->4.0 replication); 6 bytes is
+      the difference between the header's size in 4.0 (LOG_EVENT_HEADER_LEN)
+      and the header's size in 3.23 (OLD_HEADER_LEN). Note that using
+      mi->old_format will not help if the I/O thread has not started yet.
+      Yes this is a hack but it's just to make 3.23->4.x replication work;
+      3.23->5.0 replication is working much better.
+      
+      The line "mi->old_format ? : " below should NOT BE MERGED to 5.0 which
+      already works. But it SHOULD be merged to 4.1.
+    */
+    master_log_pos= log_pos + val -
+      (mi->old_format ? (LOG_EVENT_HEADER_LEN - OLD_HEADER_LEN) : 0);
+#endif
+  pthread_cond_broadcast(&data_cond);
+  if (!skip_lock)
+    pthread_mutex_unlock(&data_lock);
+}
+
+/*
+  thread safe read of position - not needed if we are in the slave thread,
+  but required otherwise as var is a longlong
+*/
+void st_relay_log_info::read_pos(ulonglong& var)
+{
+  pthread_mutex_lock(&data_lock);
+  var = relay_log_pos;
+  pthread_mutex_unlock(&data_lock);
+}
+
 void st_relay_log_info::close_temporary_tables()
 {
   TABLE *table,*next;
diff --git a/sql/slave.h b/sql/slave.h
index 4a6389e9f87..979c63af201 100644
--- a/sql/slave.h
+++ b/sql/slave.h
@@ -210,34 +210,9 @@ typedef struct st_relay_log_info
 
   st_relay_log_info();
   ~st_relay_log_info();
-  inline void inc_pending(ulonglong val)
-  {
-    pending += val;
-  }
-  /* TODO: this probably needs to be fixed */
-  inline void inc_pos(ulonglong val, ulonglong log_pos, bool skip_lock=0)
-  {
-    if (!skip_lock)
-      pthread_mutex_lock(&data_lock);
-    relay_log_pos += val+pending;
-    pending = 0;
-    if (log_pos)
-      master_log_pos = log_pos+ val;
-    pthread_cond_broadcast(&data_cond);
-    if (!skip_lock)
-      pthread_mutex_unlock(&data_lock);
-  }
-  /*
-    thread safe read of position - not needed if we are in the slave thread,
-    but required otherwise as var is a longlong
-  */
-  inline void read_pos(ulonglong& var)
-  {
-    pthread_mutex_lock(&data_lock);
-    var = relay_log_pos;
-    pthread_mutex_unlock(&data_lock);
-  }
-
+  void inc_pending(ulonglong val);
+  void inc_pos(ulonglong val, ulonglong log_pos, bool skip_lock=0);
+  void read_pos(ulonglong& var);
   int wait_for_pos(THD* thd, String* log_name, longlong log_pos, 
 		   longlong timeout);
   void close_temporary_tables();