Re: proposal: tuplestore, tuplesort aggregate functions

On Thu, Aug 19, 2010 at 5:53 PM, David Fetter <david(at)fetter(dot)org> wrote:
> On Thu, Aug 19, 2010 at 05:40:46PM -0400, Tom Lane wrote:
>> "Kevin Grittner" <Kevin(dot)Grittner(at)wicourts(dot)gov> writes:
>> > Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
>> >> I suppose there could also be a bit of an ambiguity if you're working
>> >> with a type like int4 where the values are discrete steps.  Like, what
>> >> do you do with {1, 2}?
>>
>> Hmm, good point.
>>
>> > The same thing you do with the avg function?
>>
>> avg's approach is not at all datatype-independent though.  If you're
>> willing to give up the idea of a polymorphic median() function, that
>> would probably be the thing to do.  If not, you need to take the left
>> or right one of the two central elements.
>
> Whether the median needs to be in the sample is one question that
> doesn't really have a universal right answer.  A separate question,
> also without a universal right answer, is whether the median needs to
> be in the same domain as the sample, int4 being the case above.
>
> We could just go with "whatever Oracle, DB2 and MS-SQL Server have,"
> assuming it's the same thing, until something appears in the SQL
> standard.

That's usually a sensible default, when in doubt. If nothing else, it
improves compatibility.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise Postgres Company