Handle LIMIT/OFFSET before select clause (was: Feature request: optimizer improvement)

On Oct 31, 2013, at 11:04 AM, Joe Love <joe(at)primoweb(dot)com> wrote:
> In postgres 9.2 I have a function that is relatively expensive. When I write a query such as:
>
> select expensive_function(o.id),o.* from offeirng o where valid='Y' order by name limit 1;
>
> the query runs slow and appears to be running the function on each ID, which in this case should be totally unnecessary as it really only needs to run on 1 row.
>
> When I rewrite the query like so:
>
> select expensive_function(o.id), o.*
> from (select *offering where valid='Y' order by name limit 1) o;
>
> the expensive function only runs once and thus, much faster. I would think that the optimizer could handle this situation, especially when limit or offset is used and the expensive function is not used in a group by, order by or where.

Does anyone know what the SQL standard says about this, if anything? I can't see any way that this would change the result set, but of course if the function has external effects this would make a difference...