From: Roy Lyseng
Date: October 13 2010 1:09pm
Subject: Re: bzr commit into mysql-next-mr-bugfixing branch (jorgen.loland:3262)
 Bug#52329
List-Archive: http://lists.mysql.com/commits/120684
Message-Id: <4CB5AF7E.2090004@oracle.com>
MIME-Version: 1.0
Content-Type: multipart/alternative;
 boundary="------------070104050607020503030904"

--------------070104050607020503030904
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 8bit

  Approved.

I think Øystein's suggestion for test cases is reasonable.

Otherwise, I have one comment to the commit header, see below.

Thanks,
Roy

On 13.10.10 13.38, Jorgen Loland wrote:
> #At file:///export/home/jl208045/mysql/mysql-next-mr-opt-backporting-52329/ based on revid:tor.didriksen@stripped
>
>   3262 Jorgen Loland	2010-10-13
>        Bug#52329: Wrong result: subquery materialization, IN, non-null
>                   field followed by nullable
>
>        Consider a query
>
>        SELECT * FROM t1
>        WHERE (a1, a2) IN (
>        SELECT b1, b2 FROM t2 ...)
>
>        When solved using materialized subselect, we check if a record
>        from t1 is part of the result set by constructing a key from the
>        a1 and a2 values and perform a lookup based on that key. If a1
>        and a2 are CHAR(3) with values foo and bar, the KEY is "foobar".
>        If a2 is NULLable, the null-bit should be the first bit for this
>        field's part of the KEY ("foo<nullbit_a2>bar").

You probably mean null-byte, and not null-bit?
>
>        Before, the null-bit pointer for each key field wrongly pointed
>        to the first bit of the KEY instead of the first bit of the key
>        part. Thus, when setting the null bit for a2 above, the first bit
>        of a1 was wrongly updated: "0oo bar" ('f' in foo replaced
>        with null-bit of a2, and the null-bit of a2 not set)
>
>        This patch sets the null-pointer for each key part to the first
>        bit of that field's part of KEY, not the first bit of KEY.
>
>        The bug was masked if the first field was NOT NULL. The bug also
>        applied to semi-join materialization since the code has been
>        copied (also fixed by this patch).
>
>
>
>


--------------070104050607020503030904--