Date: Sat, 21 Apr 2007 01:20:46 -0400
From: Lew <lew@nospam.lewscanon.com>
User-Agent: Thunderbird 1.5.0.9 (X11/20070102)
MIME-Version: 1.0
Subject: Re: Floating point type to store numbers
References: <4838.63.118.86.10.1176836034.squirrel@www.88thstreet.com>
	<20070417190443.GA9094@phlogiston.dyndns.org>
	<19968.63.118.86.10.1176842013.squirrel@www.88thstreet.com>
	<20070417205318.GC9094@phlogiston.dyndns.org>
In-Reply-To: <20070417205318.GC9094@phlogiston.dyndns.org>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Message-ID: <BN-dnWNvp_KzArTbnZ2dnUVZ_uninZ2d@comcast.com>
Lines: 20
To: pgsql-sql@postgresql.org

Andrew Sullivan wrote:
> On Tue, Apr 17, 2007 at 04:33:33PM -0400, Radhika Sambamurti wrote:
>> Andrew,
>> This has been quite helpful. My main concern is CPU cost.  Thanks for the
>> input.
> 
> You're welcome.  Are you sure your main concern should be CPU cost?
> It's true that numeric is more costly that float in a lot of cases,
> but I know at least one auditor who will refuse to certify results
> from programs that anywhere use floating-point storage or calculation
> on accounting data.  The problem is really that you can get compound
> errors -- very small rounding errors several times can turn out to be
> a big problem.  (One quick primer that can help you understand this
> is at <http://www2.hursley.ibm.com/decimal/decifaq1.html>.)

Would you rather have wrong answers really, really fast, or right answers fast 
enough?

-- 
Lew