maria-developers team mailing list archive

[Sergei Golubchik <serg@xxxxxxxxxxx>]

Hi, Sanja!

On Feb 25, sanja@xxxxxxxxxxx wrote:
> revision-id: 60c1dc1ab24bfe9b3f4a898fe5e7cb070ee02cf7
> parent(s): d0d6284cab70af148269d95744f615454cac24bb
> committer: Oleksandr Byelkin
> branch nick: work-maria-10.1-MDEV-6997-real
> timestamp: 2015-02-25 19:00:57 +0100
> message:
> 
> MDEV-6997: SET STATEMENT last_insert_id FOR ... does not affect the value written to the binlog
> 
> Problem was in reinitialization of
> first_successful_insert_id_in_prev_stmt_for_binlog after setting
> variables.
> 
> diff --git a/sql/sql_parse.cc b/sql/sql_parse.cc
> index 5af700b..54abfd0 100644
> --- a/sql/sql_parse.cc
> +++ b/sql/sql_parse.cc
> @@ -2728,6 +2728,10 @@ static bool do_execute_sp(THD *thd, sp_head *sp)
>          my_error(ER_WRONG_ARGUMENTS, MYF(0), "SET");
>        goto error;
>      }
> +    if (!thd->in_sub_stmt) /* stored functions and triggers are a special case */
> +    {
> +      thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt= 0;
> +    }
>    }

This seems to be a pretty complicated matter, and your comment isn't
very helpful. Try this:

  The value of last_insert_id is remembered in THD to be written to binlog
  when it's used *the first time* in the statement. But SET STATEMENT
  must read the old value of last_insert_id to be able to restore it at
  the end. This should not count at "reading of last_insert_id" and
  should not remember last_insert_id for binlog. That is, it should set
  stmt_depends_on_first_successful_insert_id_in_prev_stmt flag.

And try the test case where SET STATEMENT last_insert_id is used inside
a trigger or a stored function. I don't think you handle this case
correctly. Try both cases, when stmt_depends... is 0 and is 1 at the
point of this if().

I suspect you should not set stmt_depends_on_first_successful_insert_id_in_prev_stmt
to 0, but you need to make sure it didn't change. That is, you need to
set it to its old value (which may or may not be 0).

Regards,
Sergei