2010/10/2 Tom Lane <tgl@sss.pgh.pa.us>:
> Hitoshi Harada <umi.tanuki@gmail.com> writes:
>> 2010/10/2 Tom Lane <tgl@sss.pgh.pa.us>:
>>> The implementation I would've expected to see is to do the sort and then
>>> have two code paths for retrieving the median, depending on whether the
>>> sort result is all in memory or not.
>
>> Hm? The problem we encountered in the middle of the patch is there is
>> no chance to call tuplesort_end if median is called in moving frame
>> window aggregate because final function is called multiple times
>> during moving.
>
> Well, if you haven't got a solution for that, then this patch isn't
> ready for prime time.
>
> It's entirely possible that median as a window function is intractable.
> I'd rather have it throwing error than offer an implementation that will
> fall over as soon as the window gets large.
Well, that sounds like the conclusion. It is a shame, but we have to
throw an error from median() in the window aggregate, if Pavel does
not have any better solution. And as an aggregate function only, the
patch is ready if the window-related parts are removed.
Regards,
--
Hitoshi Harada