>
> >
> > There is no "position" in a table, there's only a "position" in a
> > resultset that's ordered.
>
> Ok. So i'll rewrite my question:
> Is it possible to select some records from a resultset based
> on their positions?
> I think it's the same idea used by the first/skip clauses. Am I wrong?
>

Yes you are wrong. As I and others have already answered you tables are
unordered sets of rows. There is no position. Postion comes to play on your
client side when you see your data in some nice grid and think, ah this
record comes after the other one so it must be that they have positions in
tables. A couple of years of studying relational theory and you might see
why that kind of thinking is wrong.

But, I am sure that you want some results and not blabing about relations...
So here is something for you...

create table p(
id integer);
commit;
insert into p(id) values (1);
insert into p(id) values (2);
insert into p(id) values (4);
insert into p(id) values (5);
insert into p(id) values (8);
insert into p(id) values (10);
insert into p(id) values (15);
insert into p(id) values (19);
insert into p(id) values (30);
commit;

select p.id, count(*)
from p
join p p2 on p2.id <= p.id
group by p.id
having count(*) in (1, 2, 4)

this query selected 1st, 2nd and 4th row... Isn't that what you wanted?
Do it on a set of more than couple of hundreds of rows and your server will
hate you, but i'll leave it you to understand why... :)

Sasha