fixed mysterious offset confusion bug

added a test case for it - took some creative work to figure out
how to make it happen at will
updated the manual


Docs/manual.texi:
  fixed wrong info on SLAVE_SKIP_COUNTER
  fixed wrong info in BitKeeper tree build instructions
  updated change history about bug fix
mysql-test/t/rpl_sporadic_master.test:
  tried hard to get slave confused, but failed. nevertheless, a more
  exhaustive test case does not hurt
sql/slave.cc:
  fixed mysterious offset confusion bug

5 files changed, 68 insertions, 6 deletions

diff --git a/Docs/manual.texi b/Docs/manual.texi
index 3d355bd0410..1d3fae2f785 100644
--- a/Docs/manual.texi
+++ b/Docs/manual.texi
@@ -6543,8 +6543,8 @@ shell> make
 
 A collection of our standard configure scripts is located in the
 @file{BUILD/} subdirectory.  If you are lazy, you can use
-@file{BUILD/compile-pentium-debug}.  It will actually work on a lot of
-non-x86 machines despite its name.
+@file{BUILD/compile-pentium-debug}. To compile on a different architecture,
+modify the script removing flags that are Pentium-specific.
 
 @item
 When the build is done, run @code{make install}.  Be careful with this
@@ -29828,8 +29828,11 @@ query from the master.
 @item
 If you have decided you can skip the next query, do
 @code{SET SQL_SLAVE_SKIP_COUNTER=1; SLAVE START;} to skip a query that
-does not use auto_increment, last_insert_id or timestamp, or
-@code{SET SQL_SLAVE_SKIP_COUNTER=2; SLAVE START;} otherwise
+does not use auto_increment, or last_insert_id  or
+@code{SET SQL_SLAVE_SKIP_COUNTER=2; SLAVE START;} otherwise. The reason
+auto_increment/last_insert_id queries are different is that they take
+two events in the binary log of the master.
+
 @item
 If you are sure the slave started out perfectly in sync with the master,
 and no one has updated  the tables involved outside of slave thread,
@@ -45728,6 +45731,9 @@ not yet 100% confident in this code.
 @appendixsubsec Changes in release 3.23.40
 @itemize @bullet
 @item
+Fixed bug in slave thread when under some rare circumstances it could
+get 22 bytes ahead on the offset in the master
+@item
 Added @code{slave_wait_timeout} for replication.
 @item
 Fixed problem with @code{UPDATE} and BDB tables.
diff --git a/mysql-test/r/rpl_mystery22.result b/mysql-test/r/rpl_mystery22.result
new file mode 100644
index 00000000000..f85b057eefa
--- /dev/null
+++ b/mysql-test/r/rpl_mystery22.result
@@ -0,0 +1,4 @@
+n
+1
+2
+3
diff --git a/mysql-test/t/rpl_mystery22.test b/mysql-test/t/rpl_mystery22.test
new file mode 100644
index 00000000000..3a48ef84dc1
--- /dev/null
+++ b/mysql-test/t/rpl_mystery22.test
@@ -0,0 +1,43 @@
+# test case to make slave thread get ahead by 22 bytes
+
+source include/master-slave.inc;
+connection master;
+# first, cause a duplicate key problem on the slave
+create table t1(n int auto_increment primary key);
+save_master_pos;
+connection slave;
+sync_with_master;
+insert into t1 values (2);
+connection master;
+insert into t1 values(NULL);
+insert into t1 values(NULL);
+save_master_pos;
+connection slave;
+sleep 1; # there is no way around this sleep - we have to wait until
+# the slave tries to run the query, fails and aborts slave thread
+delete from t1 where n = 2;
+slave start;
+sync_with_master;
+#now the buggy slave would be confused on the offset but it can replicate
+#in order to make it break, we need to stop/start the slave one more time
+slave stop;
+connection master;
+# to be able to really confuse the slave, we need some non-auto-increment
+# events in the log
+create table t2(n int);
+drop table t2;
+insert into t1 values(NULL);
+save_master_pos;
+connection slave;
+slave start;
+#now the truth comes out - if the slave is buggy, it will never sync because
+#the slave thread is not able to read events
+sync_with_master;
+select * from t1;
+#clean up
+connection master;
+drop table t1;
+save_master_pos;
+connection slave;
+sync_with_master;
+
diff --git a/mysql-test/t/rpl_sporadic_master.test b/mysql-test/t/rpl_sporadic_master.test
index a3217f9dd55..c1a47781ddf 100644
--- a/mysql-test/t/rpl_sporadic_master.test
+++ b/mysql-test/t/rpl_sporadic_master.test
@@ -3,12 +3,19 @@
 
 source include/master-slave.inc;
 connection master;
-drop table if exists t1;
+drop table if exists t1,t2;
+create table t2(n int);
 create table t1(n int not null auto_increment primary key);
 insert into t1 values (NULL),(NULL);
 delete from t1;
 # We have to use 4 in the following to make this test work with all table types
 insert into t1 values (4),(NULL);
+save_master_pos;
+connection slave;
+sync_with_master;
+slave stop;
+slave start;
+connection master;
 insert into t1 values (NULL),(NULL);
 flush logs;
 delete from t1;
@@ -20,7 +27,7 @@ connection slave;
 sync_with_master;
 select * from t1;
 connection master;
-drop table t1;
+drop table t1,t2;
 save_master_pos;
 connection slave;
 sync_with_master;
diff --git a/sql/slave.cc b/sql/slave.cc
index ab1a880a46e..7cb9bdfd3fc 100644
--- a/sql/slave.cc
+++ b/sql/slave.cc
@@ -1222,6 +1222,8 @@ pthread_handler_decl(handle_slave,arg __attribute__((unused)))
   thd->thread_stack = (char*)&thd; // remember where our stack is
   thd->temporary_tables = save_temporary_tables; // restore temp tables
   threads.append(thd);
+  glob_mi.pending = 0;  //this should always be set to 0 when the slave thread
+  // is started
   
   DBUG_PRINT("info",("master info: log_file_name=%s, position=%s",
 		     glob_mi.log_file_name, llstr(glob_mi.pos,llbuff)));