Tom Lane wrote:
Chris Kratz <chris.kratz@xxxxxxxxxxxxxx> writes:
Thanks for the reply. Yes, subselects would work very well and in some ways
are more elegant then the hand waving we had to do to get the multi-column
aggregates to work.
BTW, there is not any fundamental reason why we can't support aggregate
functions with multiple inputs. I looked at this not long ago and
determined that the major stumbling blocks would be
(1) trying to keep the code in nodeAgg.c from getting a lot more
complicated and hence slower for the single-input case;
(2) figuring out how to change the syntax for CREATE AGGREGATE.
It certainly seems doable if someone wants to spend time on it.
I did some work a couple months ago (and posted here
"http://archives.postgresql.org/pgsql-general/2005-11/msg01214.php";) on
developing composite aggregates for linear-regression, and while at first I was a bit dissatisfied with having to cast my data pairs as POINT data type, once I got confortable with that it made a lot of sense.
Regards,
Berend Tober
