Re: Reordering DISTINCT keys to match input path's pathkeys

Thanks for pointing it out.  I looked at the patch there and noticed
several problems with it.

* That patch is incomplete and does not work as expected.  It at least
needs to modify truncate_useless_pathkeys() to account for DISTINCT
clause (I think this has been mentioned in that thread).

* That patch would not consider the origin DISTINCT pathkeys if it could
do some reordering, which is not great and can generate inefficient
plans.  For instance (after fixing the first problem)

create table t (a int, b int);
create index on t(a);

set enable_hashagg to off;
set enable_incremental_sort to off;
set enable_seqscan to off;

explain (costs off) select distinct b, a from t order by b, a;
                   QUERY PLAN
-------------------------------------------------
 Sort
   Sort Key: b, a
   ->  Unique
         ->  Sort
               Sort Key: a, b
               ->  Index Scan using t_a_idx on t
(6 rows)

Using DISTINCT pathkeys {b, a} is more efficient for this plan, because
only one Sort would be required.  But that patch is not able to do that,
because it does not consider the origin DISTINCT pathkeys after
reordering.

Thanks
Richard