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

1 files changed, 43 insertions, 0 deletions

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;
+