Kristian Nielsen wrote:

Igor Babaev writes:

...

#At lp:maria based on revid:igor@askmonty.org-20090916174301-lzzh1r0khn3dxwvp

2744 Igor Babaev 2009-09-16 Fixed a failure of the merge_innodb test on windows. modified: storage/myisammrg/ha_myisammrg.cc

=== modified file 'storage/myisammrg/ha_myisammrg.cc' --- a/storage/myisammrg/ha_myisammrg.cc 2009-09-07 20:50:10 +0000 +++ b/storage/myisammrg/ha_myisammrg.cc 2009-09-17 03:36:21 +0000 @@ -309,7 +309,7 @@ static MI_INFO *myisammrg_attach_childre TABLE *parent; TABLE *child; TABLE_LIST *child_l; - MI_INFO *myisam; + MI_INFO *myisam= NULL; DBUG_ENTER("myisammrg_attach_children_callback");

my_errno= 0;

Are you sure this fixes the failure?

The reason I ask is that this change does not appear to be able to affect the result of the code in any way... The only code path where myisam is referenced without being assigned is when the table is not MYISAM (child->file->ht->db_type != DB_TYPE_MYISAM), and in this case the only reference is in a DBUG_PRINT(). So this could maybe fix a Valgrind warning, but how could it have any effect on a failure on Windows?

Kristian, The failure is exactly in DBUG_PRINT with debug mode on Windows. Regards, Igor.

The patch looks ok as it is, as it prevents the debug printout of an undefined value. I am just curious how this could fix a failure, maybe it is just hiding the real bug?

- Kristian.