Re: Using indices for UNION.

Hello, I've totally refactored the series of pathes and cut out
the appropriate portion as 'UNION stuff'.

This patch rquires another 'pathkeys expansion using fully-orderd
index' patch to work.

http://www.postgresql.org/message-id/20131119.203516.251520490.horiguchi.kyotaro@lab.ntt.co.jp

1. plan_with_pathkeys_v1_20131119.patch

This is a patch adding pathkeys (and uniqueness) information on
struct Plan to do what the latter part of grouping_planner does
on current_pathkeys in seemingly autonomous way. As in the
previous discussion, this is in a sense a bad direction to
go. But this patch make the path manipulations occured in the
latter of grouping_planner simple and would reduce extra sorting
and uniq'ing.

2. union_uses_idx_v3_20131119.patch

This is made to apply after the first patch above. The core of
"Using indices for UNION". This is - as in the previous
discussion - also not in so smart way to do that. Most of all,
this patch runs subquery_planner additionally for UNION with
some conditions. Nevertheless, the expected gain should not be
small.

regards,

--
Kyotaro Horiguchi
NTT Open Source Software Center