Subject: Re: SQL challenge--top 10 for each key value?
From: Rod Taylor <pg@rbt.ca>
To: Greg Stark <gsstark@mit.edu>
Cc: Josh Berkus <josh@agliodbs.com>, Jeff Boes <jboes@nexcerpt.com>,
	pgsql-sql@postgresql.org
In-Reply-To: <87ad1kg4c2.fsf@stark.xeocode.com>
References: <33af4fa4e4ec230574cefcff737ea40a@news.teranews.com>
	<87r7uyf3ki.fsf@stark.xeocode.com> <1081481466.56361.808.camel@jester>
	<200404090906.39541.josh@agliodbs.com>
	<87ad1kg4c2.fsf@stark.xeocode.com>
Content-Type: text/plain
Message-Id: <1081560594.56361.818.camel@jester>
Mime-Version: 1.0
Date: Fri, 09 Apr 2004 21:29:55 -0400
Content-Transfer-Encoding: 7bit

On Fri, 2004-04-09 at 18:43, Greg Stark wrote:
> Josh Berkus <josh@agliodbs.com> writes:
> 
> > Rod,
> > 
> > > Something along the lines of the below would accomplish what you want
> > > according to spec. ROW_NUMBER() is a spec defined function. (6.10 of
> > > SQL200N)
> > 
> > Great leaping little gods!   They added something called "row number" to the 
> > spec? 
> > 
> > Boy howdy, folks were right ... the ANSI committee really has completly blown 
> > off the relational model completely.   
> 
> If it's like Oracle's rownum then it's the row number of the *output*, not the
> position on disk. So it's not entirely blowing off the relational model any
> more than ORDER BY does.
> 
> The weird thing is the number of cases where you want ORDER BY or rownum
> inside subselects. Which the solution to the original question needed.

It's not really like Oracles row num at all, though I suppose you can
emulate rownum using it. The intention is that you will use it for
"aggregates" like running totals, moving averages, counting, etc.

http://www.devx.com/getHelpOn/10MinuteSolution/16573/1954?pf=true