On Thu, Jul 2, 2020 at 8:52 PM Etsuro Fujita <etsuro.fujita@gmail.com> wrote:
> I'll look at the patch more closely tomorrow.
I spent some time reviewing the patch, and noticed that the changes to
gen_prune_steps_from_opexps() break the logic in
get_matching_range_bounds(), causing another issue. Here is an
example:
postgres=# create table prefix_test (a int, b varchar) partition by
range (a, b);
postgres=# create table prefix_test1 partition of prefix_test for
values from (1, 'a') to (1, 'b');
postgres=# create table prefix_test2 partition of prefix_test for
values from (2, 'a') to (2, 'b');
postgres=# set enable_partition_pruning to on;
postgres=# explain select * from prefix_test where a <= 2 and b = 'a';
QUERY PLAN
--------------------------------------------------------------------------
Seq Scan on prefix_test2 prefix_test (cost=0.00..29.05 rows=2 width=36)
Filter: ((a <= 2) AND ((b)::text = 'a'::text))
(2 rows)
Will do a bit more investigation about this.
Best regards,
Etsuro Fujita