Jakub Ouhrabka <kuba@comgate.cz> writes:
> preparing the test case was easier than I expected. It's attached. Fast
> planning on 8.2.4, very slow on 8.2.5.
Hmm. I think there are two different bugs involved here. One is fixed
by the attached patch, and it masks the performance problem on your test
case, but I wonder whether it will be enough for your real application.
Can you try this and see if it fixes 8.2.5 for you?
regards, tom lane
Index: initsplan.c
===================================================================
RCS file: /cvsroot/pgsql/src/backend/optimizer/plan/initsplan.c,v
retrieving revision 1.123.2.7
diff -c -r1.123.2.7 initsplan.c
*** initsplan.c 4 Oct 2007 20:44:55 -0000 1.123.2.7
--- initsplan.c 24 Oct 2007 20:34:08 -0000
***************
*** 625,630 ****
--- 625,631 ----
* rel in the lower OJ's min_righthand, not its whole syn_righthand.
*/
if (bms_overlap(left_rels, otherinfo->syn_righthand) &&
+ bms_overlap(clause_relids, otherinfo->syn_righthand) &&
!bms_overlap(strict_relids, otherinfo->min_righthand))
{
min_lefthand = bms_add_members(min_lefthand,
