On Thu, Mar 02, 2006 at 11:19:32AM -0500, Chaz. wrote:
> I am trying to understand something I have seen happen. I had a select
> that looked like:
>
> select f(A) from A, B, C where g(A)
>
> Where f(A) is the select that only depends on table A;
> g(A) is the where part that only depends on table A.
>
> What I saw happen was the optimizer will waste a lot of time (seconds!)
> bringing in table B and C. I was wondering why doesn't the optimizer
> drop references to tables B and C since they aren't used any where?
The above query does a cross join. Even though you're not using
values from B and C they're still contributing rows to the result
set.
test=> SELECT * FROM a;aid
-----a1a2
(2 rows)
test=> SELECT * FROM b;bid
-----b1b2
(2 rows)
test=> SELECT * FROM c;cid
-----c1c2
(2 rows)
test=> SELECT a.*, b.*, c.* FROM a, b, c WHERE a.aid = 'a1';aid | bid | cid
-----+-----+-----a1 | b1 | c1a1 | b2 | c1a1 | b1 | c2a1 | b2 | c2
(4 rows)
test=> SELECT a.*, b.* FROM a, b, c WHERE a.aid = 'a1';aid | bid
-----+-----a1 | b1a1 | b2a1 | b1a1 | b2
(4 rows)
test=> SELECT a.* FROM a, b, c WHERE a.aid = 'a1';aid
-----a1a1a1a1
(4 rows)
--
Michael Fuhr
