Re: Aggregates with non-commutative transition functions

Emmanuel Charpentier <charpent(at)bacbuc(dot)dyndns(dot)org> writes:
> Tom Lane wrote:
>> Note the lack of an extra sort above the subquery. This provides a
>> general technique for controlling the ordering of inputs to a
>> user-written aggregate function, even when grouping.

> Schn ! I suppose that this has other fringe benefits for planning in
> general ...

Yes, it should save cycles in many scenarios, so I thought it was worth
doing in any case.

> Do you plan incorporation in some forthcoming 7.3.x ? Or push it back
> to 7.4 ?

No, I would not risk back-patching this into 7.3.*. It's not a bug fix.

(But having said that, you could get the diffs from the CVS server and
back-patch to create your own private 7.3 variant, if you can't wait
for 7.4. Offhand I do not think there'd be any great difficulty in
applying the change to 7.3 branch.)

regards, tom lane