and then we can remove few *.include files that do just that — switch connection with a message

serg, please review patch for this task. It is 95% complete, available in bb-10.2-MDEV-6720. I expect there're still tests to be stabilized (to be discovered by bb). I'll commit further changes on top of it and squash before pushing.

You seem to be getting tasks that remove lots of code

This one: 64 insertions, 3414 deletions (not counting result files).

Ok to push, thanks!
One comment only: the change in rpl_begin_commit_rollback.test looks like a genuine bug in the test. Please, fix it, in a separate commit.

diff --git a/mysql-test/suite/rpl/t/rpl_begin_commit_rollback.test b/mysql-test/suite/rpl/t/rpl_begin_commit_rollback.test

--- a/mysql-test/suite/rpl/t/rpl_begin_commit_rollback.test

+++ b/mysql-test/suite/rpl/t/rpl_begin_commit_rollback.test

@@ -133,8 +129,7 @@ source include/wait_for_slave_sql_to_start.inc;

 --echo # SAVEPOINT and ROLLBACK TO have the same problem in BUG#43263

 --echo # This was reported by BUG#50407

 connection master;

-echo [on master]

-SET SESSION AUTOCOMMIT=0;

+echo SET SESSION AUTOCOMMIT=0;

 let $binlog_start=query_get_value(SHOW MASTER STATUS, Position, 1);

 BEGIN;

Compared to previous tasks that removed lots of code, this one was disaster.

I also came across the same line in rpl_begin_commit_rollback.test, but I couldn't immediately find out if this test was actually supposed to set autocommit or not. I'll double check it.

Thanks!