Greg Stark <gsstark(at)mit(dot)edu> writes:
> slo=> explain
> select *
> from (select foo_id,count(*) as n from foo_bar group by foo_id) as x
> join (select * from foo limit 1) as foo using (foo_id)
> ;
Why not put the subselect in the output list, if that's the kind of plan
you want?
regression=# explain select foo.*, (select count(*) from foo_bar where foo_id = foo.foo_id) as n from foo;
QUERY PLAN
------------------------------------------------------------------------------------
Seq Scan on foo (cost=0.00..17102.50 rows=1000 width=8)
SubPlan
-> Aggregate (cost=17.08..17.08 rows=1 width=0)
-> Index Scan using foobi on foo_bar (cost=0.00..17.07 rows=5 width=0)
Index Cond: (foo_id = $0)
(5 rows)
regards, tom lane