Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!mcvax!botter!ark!huisjes
From: huisjes@ark.cs.vu.nl (Michiel Huisjes)
Newsgroups: comp.sys.ibm.pc
Subject: Rounding errors on a PC
Message-ID: <979@ark.cs.vu.nl>
Date: Tue, 21-Apr-87 10:21:08 EST
Article-I.D.: ark.979
Posted: Tue Apr 21 10:21:08 1987
Date-Received: Wed, 22-Apr-87 04:29:21 EST
Reply-To: huisjes@cs.vu.nl (Michiel Huisjes)
Distribution: world
Organization: VU Informatica, Amsterdam
Lines: 27


We are developing an administration package for IBM PC and compatibles,
using the Microsoft C compiler 4.00 and no 8087 co-processor. When we
entered the number "123456.01" and printing it afterwards, we got
"123456.09999999" back. Using this float in calculations leads (very quickly)
to a rounding error of one (or more) cents. Although this amount of money,
isn't that much to loose, it shouldn't happen in an administration package.

Is this a compiler error, or just the inaccuracy of an IBM PC? If the latter
is the case, what can be done about it. I know that the COBOL compiler
uses BCD (Binary Coded Decimals) for its calculations, but implementing
this in C, would mean writing all substraction, addition, multiplication, etc.
routines all over.
Do such packages already exist? Is there an undocumented obscure C compiler 
flag I don't know about? Has anybody out there encountered the same problem
and if so, how did they deal with it?

Answers to these questions or solutions are VERY welcome. If any useful
solutions are presented, or wise remarks are made, I will summarize
to the net.

Please send replies to:

			Michiel Huisjes.  (huisjes@cs.vu.nl)
			{seismo|decvax|philabs}!mcvax!vu44!huisjes

Thanx in advance.