Message-ID: <039401c3335d$f94d1760$0600a8c0@Wilbur>
From: "Julian Scarfe" <julian@avbrief.com>
To: "Tom Lane" <tgl@sss.pgh.pa.us>
Cc: <pgsql-sql@postgresql.org>
References: <036301c33350$a4bd5e60$0600a8c0@Wilbur>
	<13056.1055692197@sss.pgh.pa.us>
Subject: Re: Ordering in an aggregate -- points to paths 
Date: Sun, 15 Jun 2003 17:48:40 +0100
MIME-Version: 1.0
Content-Type: text/plain;
	charset="Windows-1252"
Content-Transfer-Encoding: 7bit

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

> Yeah, this is a fairly obvious thing to want to do with a user-written
> aggregate.  It does not work in released versions, because the planner
> does not notice that the inner SELECT's output ordering matches what
> the GROUP BY needs, and so it inserts an additional Sort plan step
> above the sub-select (you can see this if you look at EXPLAIN output).
> Unfortunately, on most platforms qsort() isn't stable and will not
> preserve the ordering of its input for equal keys.  So you lose the
> minor ordering by seq_no in the re-sort.

Most grateful for the rapid response Tom. Knowing that, I can work around by
iterating through the firs at the application level.

Regards

Julian Scarfe

PS: you shouldn't be working on a Sunday, it's bad for you ;-)