From: Tor Didriksen
Date: April 8 2011 5:53am
Subject: Re: bzr commit into mysql-5.1 branch (tor.didriksen:3648) Bug#11765713
List-Archive: http://lists.mysql.com/commits/135013
Message-Id: <4D9EA2D5.7040905@oracle.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 8bit

On 2011-04-07 18:05, Guilhem Bichot wrote:
> Hello Tor,
>
> Tor Didriksen a écrit, Le 07.04.2011 13:56:
>> #At file:///export/home/didrik/repo/5.1-foo/ based on 
>> revid:georgi.kodinov@stripped
>>
>>  3648 Tor Didriksen    2011-04-07
>>       Bug#11765713 58705: OPTIMIZER LET ENGINE DEPEND ON 
>> UNINITIALIZED VALUES CREATED BY OPT_SUM_QU
>>      @ mysql-test/r/subselect.result
>>         New test case.
>>      @ mysql-test/t/subselect.test
>>         New test case.
>>      @ sql/opt_sum.cc
>>         Add thd to opt_sum_query/matching_cond/find_key_for_maxmin,
>>         enabling them to test for errors.
>>         Return with error code if thd->is_error() rather than 
>> continuing to read the index.
>>      @ sql/sql_select.cc
>>         Add thd to opt_sum_query, enabling it to test for errors.
>>      @ sql/sql_select.h
>>         Add thd to opt_sum_query, enabling it to test for errors.
>>
>
> Can you explain what the bug's plot was, without the code patch: what 
> function evaluated which subquery (there are several subqueries) and 
> what function set the error flag; what function failed to notice the 
> error flag and used a result which should not have been used, and did 
> what should not have been done...
> I'm surprised that we need to add tests of thd->is_error() to more 
> than one place, hence my questions.
> Thanks!
Hi Guilhem

Run the new testcase with --debug, and you will see what is going on.

You may also want to apply this patch to _db_return_()

--- dbug/dbug.c 2011-02-22 21:03:32 +0000
+++ dbug/dbug.c 2011-04-08 05:40:24 +0000
@@ -1071,7 +1071,7 @@
          pthread_mutex_lock(&THR_LOCK_dbug);
        DoPrefix(cs, _line_);
        Indent(cs, cs->level);
-      (void) fprintf(cs->stack->out_file, "<%s\n", cs->func);
+      (void) fprintf(cs->stack->out_file, "<%s %d\n", cs->func, _line_);


Please also read the discussion between gluh and myself about the first 
version of the patch.

The innermost subquery returns two rows, which makes 'a <=> (subquery)' 
fail.
matching_cond() wasn't prepared for it, hence the new lines checking for 
thd->is_error()

-- didrik