Saturday, July 5, 2008

Re: [HACKERS] introduction of WIP window function patch

On Sat, Jul 05, 2008 at 07:04:29PM +0900, H.Harada wrote:
> Hi,
>
> As I proposed a month before, I am working on window function.

Very nice!

> http://umitanuki.net/pgsql/wfv01/design.html
>
> The problem is, as written in the "Things to discussed" section of the
> document, how you define window functions (e.g. RANK()). My idea is to
> treat them as specialized functions such as SET OF functions and mark
> it in pg_proc. But this doesn't resolve RANK() boundary problem.

Actually, I would make RANK() and ROW_NUMBER() act more like
aggregates. ISTM you have two kinds of window functions:

- aggregation: a result is calculated over a set and the result copied
across all the rows.
- order depenadant: same as above, but the result is different for each
row.

I think you could make the latter work using the current aggregation
setup, just by calling the final_func for each row rather than just
once at the end.

That would make RANK() a normal aggrgate which returns the number of
distinct values seen so far (assuming input is ordered) and
ROW_NUMBER() is just an alias for COUNT().

I hope this is clear, let me know if it doesn't make sense.

Have a nice day,
--
Martijn van Oosterhout <kleptog@svana.org>

http://svana.org/kleptog/
> Please line up in a tree and maintain the heap invariant while
> boarding. Thank you for flying nlogn airlines.

No comments: