Hi Sven
See below, after done that, the patch is ok to push.
On 2008-01-07 Mon 15:34 +0100，Sven Sandberg wrote：
> Below is the list of changes that have just been committed into a local
> 5.1 repository of sven. When sven does a push these changes will
> be propagated to the main repository and, within 24 hours after the
> push, to the public repository.
> For information on how to access the public repository
> see http://dev.mysql.com/doc/mysql/en/installing-source-tree.html
>
> ChangeSet@stripped, 2008-01-07 15:34:51+01:00, sven@riska.(none) +2 -0
> BUG#27779: Slave cannot read old rows log events.
> Problem: Some previous versions use a different binlog format. In order
> to replicate from old versions to new versions, the new version must be
> aware of all old formats. This patch fixes one of several issues when
> replicating from mysql-5.1-wl2325-5.0-drop6 to mysql-5.1-new-rpl.
> Specifically, in mysql-5.1-wl2325-5.0-drop6, the event type id's were
> different than in mysql-5.1-new-rpl.
> Fix (in mysql-5.1-new-rpl):
> (1) detect that the server that generated the events uses the old
> format, by checking the server version of the format_description_log_event
> This patch recognizes mysql-5.1-wl2325-5.0-drop6 and -drop5.
> (2) if the generating server is old, map old event types to new event
> types using a permutation array.
>
> This fixes part of the bug (specifically, mysqlbinlog can read the
> binlog rpl_bug27779_a.dat attached to the bug report for BUG#27779
> without problems). However, there are remaining problems with replicating
> from mysql-5.1-wl2325-5.0-drop6 to mysql-5.1-new-rpl (specifically, the
> test case for BUG#27779 fails: when the output of
> 'mysqlbinlog rpl_bug27779_a.dat' is piped to a client, the client fails to
> execute an Update_rows_log_event).
>
> sql/log_event.cc@stripped, 2008-01-07 15:34:47+01:00, sven@riska.(none) +73 -2
> Added code to read events generated by mysql-5.1-wl2325-5.0-drop5 and
> ...-drop6. More precisely, the event type id's had different numbers in
> that version. To fix, we add a permutation array which maps old_id to
> new_id when the format_description_log_event indicates that the
> originating server is mysql-5.1-5-a_drop[56]. We also need to permute the
> post_header_len array accordingly.
>
> sql/log_event.h@stripped, 2008-01-07 15:34:47+01:00, sven@riska.(none) +5 -1
> Added declaration needed in log_event.cc. Also, the destructor of
> Format_description_log_event is sometimes called when post_header_len is
> null, so we must pass the MY_ALLOW_ZERO_PTR flag to my_free.
>
> diff -Nrup a/sql/log_event.cc b/sql/log_event.cc
> --- a/sql/log_event.cc 2007-12-17 14:21:19 +01:00
> +++ b/sql/log_event.cc 2008-01-07 15:34:47 +01:00
> @@ -1071,6 +1071,28 @@ Log_event* Log_event::read_log_event(con
> }
> else
> {
> + /*
> + In mysql-5.1-wl2325-5.0-drop6 and -drop5 (and possibly other
> + versions), the event types were assigned different id numbers
> + than in the present version. In order to replicate from such
> + versions to the present version, we must map those event type
> + id's to our event type id's. This is done with the
> + event_type_permutation array, which was set up when the
> + Format_description_log_event was read.
> + */
> + if (description_event->event_type_permutation)
> + {
> + IF_DBUG({
> + int new_event_type=
> + description_event->event_type_permutation[event_type];
> + DBUG_PRINT("info",
> + ("converting event type %d to %d (%s)",
> + event_type, new_event_type,
> + get_type_str((Log_event_type)new_event_type)));
> + });
> + event_type= description_event->event_type_permutation[event_type];
> + }
> +
> switch(event_type) {
> case QUERY_EVENT:
> ev = new Query_log_event(buf, event_len, description_event, QUERY_EVENT);
> @@ -2771,7 +2793,7 @@ int Start_log_event_v3::do_apply_event(R
>
> Format_description_log_event::
> Format_description_log_event(uint8 binlog_ver, const char* server_ver)
> - :Start_log_event_v3()
> + :Start_log_event_v3(), event_type_permutation(0)
> {
> binlog_version= binlog_ver;
> switch (binlog_ver) {
> @@ -2896,7 +2918,7 @@ Format_description_log_event(const char*
> const
> Format_description_log_event*
> description_event)
> - :Start_log_event_v3(buf, description_event)
> + :Start_log_event_v3(buf, description_event), event_type_permutation(0)
> {
>
> DBUG_ENTER("Format_description_log_event::Format_description_log_event(char*,...)");
> buf+= LOG_EVENT_MINIMAL_HEADER_LEN;
> @@ -2911,6 +2933,55 @@ Format_description_log_event(const char*
> number_of_event_types*
> sizeof(*post_header_len), MYF(0));
> calc_server_version_split();
> +
> + /*
> + In versions 5.1.5-a_drop[56]p20-log, the events were given other
> + event type id numbers than in the present version. When
> + replicating from such a version, we therefore set up an array that
> + maps those id numbers to the id numbers of the present server.
> + (If post_header_len is null, it means malloc failed, and is_valid
> + will fail, so no need to do anything.)
> + */
> + if (post_header_len &&
> + (strncmp(server_version, "5.1.5-a_drop", 12) == 0 &&
> + (server_version[12] == 5 || server_version[12] == 6)))
the server version string for mysql-5.1-wl2325-drop6 is
5.2.2-a_drop6p13-log, should be added here. There might be other
versions should be added too.
> + {
> + if (number_of_event_types != 22)
> + {
> + DBUG_PRINT("info", (" number_of_event_types=%d",
> + number_of_event_types));
> + /* this makes is_valid() return false. */
> + my_free(post_header_len, MYF(MY_ALLOW_ZERO_PTR));
> + post_header_len= NULL;
> + DBUG_VOID_RETURN;
> + }
> + static const uint8 perm[23]=
> + {
> + UNKNOWN_EVENT, START_EVENT_V3, QUERY_EVENT, STOP_EVENT, ROTATE_EVENT,
> + INTVAR_EVENT, LOAD_EVENT, SLAVE_EVENT, CREATE_FILE_EVENT,
> + APPEND_BLOCK_EVENT, EXEC_LOAD_EVENT, DELETE_FILE_EVENT,
> + NEW_LOAD_EVENT,
> + RAND_EVENT, USER_VAR_EVENT,
> + FORMAT_DESCRIPTION_EVENT,
> + TABLE_MAP_EVENT,
> + PRE_GA_WRITE_ROWS_EVENT,
> + PRE_GA_UPDATE_ROWS_EVENT,
> + PRE_GA_DELETE_ROWS_EVENT,
> + XID_EVENT,
> + BEGIN_LOAD_QUERY_EVENT,
> + EXECUTE_LOAD_QUERY_EVENT,
> + };
> + event_type_permutation= perm;
> + /*
> + Since we use (permuted) event id's to index the post_header_len
> + array, we need to permute the post_header_len array too.
> + */
> + uint8 post_header_len_temp[23];
> + for (int i= 0; i < 23; i++)
> + post_header_len_temp[perm[i]]= post_header_len[i];
> + for (int i= 0; i < 23; i++)
> + post_header_len[i] = post_header_len_temp[i];
> + }
> DBUG_VOID_RETURN;
> }
>
> diff -Nrup a/sql/log_event.h b/sql/log_event.h
> --- a/sql/log_event.h 2007-12-14 19:01:59 +01:00
> +++ b/sql/log_event.h 2008-01-07 15:34:47 +01:00
> @@ -2106,12 +2106,16 @@ public:
> /* The list of post-headers' lengthes */
> uint8 *post_header_len;
> uchar server_version_split[3];
> + const uint8 *event_type_permutation;
>
> Format_description_log_event(uint8 binlog_ver, const char* server_ver=0);
> Format_description_log_event(const char* buf, uint event_len,
> const Format_description_log_event
> *description_event);
> - ~Format_description_log_event() { my_free((uchar*)post_header_len, MYF(0)); }
> + ~Format_description_log_event()
> + {
> + my_free((uchar*)post_header_len, MYF(MY_ALLOW_ZERO_PTR));
> + }
> Log_event_type get_type_code() { return FORMAT_DESCRIPTION_EVENT;}
> #ifndef MYSQL_CLIENT
> bool write(IO_CACHE* file);
