Re: Strange Execution-Plan for NOT EXISTS

Jens Hartwig <jhartwig(at)debis(dot)com> writes:
> select (min(id) + 1)
> from t_dummy d1
> where not exists (
> select id
> from t_dummy d2
> where d2.id = (d1.id + 1)
> );

> Aggregate (cost=2924207.88..2924207.88 rows=1 width=12)
-> Seq Scan on t_dummy d1 (cost=0.00..2924207.88 rows=1 width=12)
> SubPlan
> -> Seq Scan on t_dummy d2 (cost=0.00..331.36 rows=1 width=12)

> Why that? Wouldnt it be possible to simple use the primary key index in
> the sub-query

I think that 7.0.* is too stupid to consider an indexscan for a
qualifier that looks like "indexvar = ($1 + 1)", which is what you
effectively have in this case ($1 being the Param passed in from
the outer plan). It only recognizes "indexvar = constant" and
"indexvar = param" as indexable expressions. Current sources do
better.

regards, tom lane