Michael Fuhr <mike@fuhr.org> writes:
> Tom Lane says he's found the problem; I expect he'll be committing
> a fix shortly.
The attached patch allows it to generate the expected plan, at least
in the test case I tried.
regards, tom lane
*** src/backend/optimizer/path/indxpath.c.orig Sun Aug 28 18:47:20 2005
--- src/backend/optimizer/path/indxpath.c Thu Sep 22 19:17:41 2005
***************
*** 955,969 ****
/*
* Examine each joinclause in the joininfo list to see if it matches any
* key of any index. If so, add the clause's other rels to the result.
- * (Note: we consider only actual participants, not extraneous rels
- * possibly mentioned in required_relids.)
*/
foreach(l, rel->joininfo)
{
RestrictInfo *joininfo = (RestrictInfo *) lfirst(l);
Relids other_rels;
! other_rels = bms_difference(joininfo->clause_relids, rel->relids);
if (matches_any_index(joininfo, rel, other_rels))
outer_relids = bms_join(outer_relids, other_rels);
else
--- 955,967 ----
/*
* Examine each joinclause in the joininfo list to see if it matches any
* key of any index. If so, add the clause's other rels to the result.
*/
foreach(l, rel->joininfo)
{
RestrictInfo *joininfo = (RestrictInfo *) lfirst(l);
Relids other_rels;
! other_rels = bms_difference(joininfo->required_relids, rel->relids);
if (matches_any_index(joininfo, rel, other_rels))
outer_relids = bms_join(outer_relids, other_rels);
else
