Re: Path question

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> FIXME #1 and FIXME #2 were much harder to trigger. In fact, barring
> significant further lobotimization of the code, I couldn't.

It's not that hard if you just tweak equivclass.c to make the order of
equivalence-class lists different, viz

diff --git a/src/backend/optimizer/path/equivclass.c b/src/backend/optimizer/path/equivclass.c
index a20ed5f..9528d0b 100644
--- a/src/backend/optimizer/path/equivclass.c
+++ b/src/backend/optimizer/path/equivclass.c
@@ -353,7 +353,7 @@ add_eq_member(EquivalenceClass *ec, Expr *expr, Relids relids,
{
ec->ec_relids = bms_add_members(ec->ec_relids, relids);
}
- ec->ec_members = lappend(ec->ec_members, em);
+ ec->ec_members = lcons(em, ec->ec_members);

return em;
}

Then for instance:

regression=# create table t1 (f1 int);
CREATE TABLE
regression=# create table t2 () inherits (t1);
CREATE TABLE
regression=# explain select * from t1 a join t1 b using (f1);
WARNING: FIXME #1
WARNING: FIXME #1
WARNING: FIXME #1
WARNING: FIXME #1
WARNING: FIXME #1
WARNING: FIXME #1
WARNING: FIXME #1
WARNING: FIXME #1

Since the order of equivalence-class member lists isn't supposed to be
semantically significant, I claim that the code in createplan has to
be able to deal with this.

Note that what this is triggering is the em_is_child condition. I think
it may indeed be impossible to get a hit on the em_is_const case as the
system currently stands; the reason being that an EC containing a const
won't normally show up as a pathkey. It can only do so if it's
below_outer_join, as the comment notes. Now the calls to
make_sort_from_pathkeys in createplan.c are only used for constructing
subsidiary sorts for a mergejoin, and we won't consider building a
mergejoin with an EC that contains a const (see
eclass_useful_for_merging). There are some calls in planner.c that are
associated with doing a final sort or distinct, but I suspect they'd
never be applied with a below_outer_join EC. So given the current usage
of make_sort_from_pathkeys it might be pretty hard to get it applied to
an EC containing a constant. That's not a reason for it not to handle
the case, though.

regards, tom lane