Message-Id: <200307221403.h6ME3Oxj007225@head-cfa.cfa.harvard.edu>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Bruce Momjian <pgman@candle.pha.pa.us>,
	Jean-Luc Lachance <jllachan@nsd.ca>,
	"Michael S. Tibbetts" <mtibbetts@head-cfa.cfa.harvard.edu>,
	pgsql-sql@postgresql.org
Subject: Re: min() and NaN 
In-reply-to: Message from Tom Lane <tgl@sss.pgh.pa.us>
	of "Tue, 22 Jul 2003 02:09:09 -0400." <4074.1058854149@sss.pgh.pa.us>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Date: Tue, 22 Jul 2003 10:03:24 -0400
From: Michael Tibbetts <tibbetts@head-cfa.harvard.edu>

Treating NaN's as larger(or smaller) than all ordinary values seems a fine way 
to go.

It avoids the situation where you request MIN and get an ordinary value which 
is greater than the minimum ordinary value in the table.  If MIN(or MAX given 
the ordering you're suggesting) returns NaN, the user would stand better odds 
of figuring out that something about the query needs to be changed.

Returning an plausible, though possibly incorrect, ordinary value from MIN or 
MAX if there are NaN's in the column can lead users to make some unfortunate 
mistakes(voice of experience?).

Thanks for your help,
Mike


> Bruce Momjian <pgman@candle.pha.pa.us> writes:
> > Is this a TODO?
> 
> It'll only take ten minutes to make it a DONE, once we figure out what
> the behavior ought to be.  So far I think both Stephan and I argued that
> MIN/MAX ought to treat NaN as larger than all ordinary values, for
> consistency with the comparison operators.  That was not the behavior
> Michael wanted, but I don't see that we have much choice given the
> wording of the SQL spec.  Does anyone want to argue against that
> definition?
> 
> 			regards, tom lane