Tom noted that create_unique_path did not check for set returning functions.
> I notice that create_unique_path is not paying attention to the question
> of whether the subselect's tlist contains SRFs or volatile functions.
> It's possible that that's a pre-existing bug.
I looked at this a bit and I can confirm that it does not behave as it should do. Take the following as an example:
create table x (id int primary key);
create table y (n int not null);
insert into x values(1);
insert into y values(1);
select * from x where (id,id) in(select n,generate_series(1,2) / 10 + 1 g from y);
id
----
1
(1 row)
select * from x where (id,id) in(select n,generate_series(1,2) / 10 + 1 g from y group by n);
id
----
1
1
(2 rows)
The 2nd query does group by n, so query_is_distinct_for returns true, therefore the outer query think's it's ok to perform an INNER JOIN rather than a SEMI join, which is this case produces an extra record.
I think we should probably include the logic to test for set returning functions into query_is_distinct_for.
The attached fixes the problem.