Message-ID: <06dc01c24756$9008f1a0$0500a8c0@Wilbur>
From: "Julian Scarfe" <julian.scarfe@ntlworld.com>
To: "Tom Lane" <tgl@sss.pgh.pa.us>
Cc: <pgsql-sql@postgresql.org>
References: <046401c246ad$48dbdc10$0500a8c0@Wilbur>
	<1115.1029700538@sss.pgh.pa.us>
Subject: Re: Ordering with GROUPs 
Date: Mon, 19 Aug 2002 09:01:02 +0100
MIME-Version: 1.0
Content-Type: text/plain;
	charset="Windows-1252"
Content-Transfer-Encoding: 7bit

From: "Tom Lane" <tgl@sss.pgh.pa.us>

> ISTM the problem here is the lack of any ordering operator for POINT,
> which defeats GROUP BY, *plus* the lack of any aggregate you might use
> for an aggregate-based solution.  This is not really a language failing
> but a problem with an impoverished datatype.

Yes, I agree completely.  If I were doing this again from scratch I'd be
using PostGIS, but I've got a lot of data that depends on POINT.

> So, if you don't like
> Bruno's subselect-based workaround, the dummy aggregate seems the way
> to go.

I've actually implemented the dummy aggregate now, and it works fine.  So
does Bruno's subselect (thank you Bruno), and the efficiency seems to be
similar in each case.

> SQL99 contains a whole bunch of verbiage whose intent seems to be that
> if you GROUP BY a unique or primary-key column, you can reference the
> other columns of that table without aggregation (essentially, the
> system treats them as implicitly GROUP BY'd).  Sooner or later we'll
> probably get around to implementing that, and that would solve your
> problem as long as you declare location.ident properly.

That makes a lot of sense, though I imagine there are higher priorities.

Thanks for your help.

Julian Scarfe