Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!pasteur!ames!sgi!key!perry
From: perry@key.COM (Peter Kiehtreiber)
Newsgroups: comp.sys.mac.programmer
Subject: Re: IBM (Intel) Real Numbers
Summary: Be careful with ASCII representations of floating-point numbers
Keywords: IEEE floating-point, 8087, 68881
Message-ID: <361@key.COM>
Date: 14 Feb 89 21:40:02 GMT
References: <14486@cup.portal.com> <12194@dartvax.Dartmouth.EDU> <655@internal.Apple.COM>
Reply-To: perry@arkon.key.COM (Perry The Cynic)
Organization: Key Computer Laboratories, Fremont
Lines: 30

In article <655@internal.Apple.COM> SIAC@applelink.apple.com (Eric Ulevik) writes:
> In article <14486@cup.portal.com> Greg_Mark_Finnegan@cup.portal.com 
> writes:
> >Has anyone ever tried to read a data file from an IBM PC
> >that consisted of REAL (in the Pascal sense) numbers.> >
> >Any help or code would be appreciated.
> The simplest method (_if_ you can afford the disk space and CPU time) is
> to re-read the pc data file into a pc program, and write it out as
> ascii-format numbers (eg. "0.123E4"). This makes it a lot easier to
> write a Mac program to read the data.

Be careful with (decimal) text representations for floating-point numbers.
Both the conversion to and from decimal text involves rounding and therefore
inaccuracies, and the changes are (generally) not reversible. Unless you
take special precautions, the numbers coming out at the other end are
marginally different from the original.

Both the 80[23]87 and 6888[12] coprocessors purport to conform to the IEEE
floating-point standard. As such, the separation in exponent/mantissa,
exponent biasing, NaN recognition etc. are the same. The only difference
lies in bit (and byte) ordering, and certain details like silent/screaming
NaN distinction. The text transmission method Eric mentions above is OK if
the programs separate the numbers into exponent and mantissa, and read and
write them as hexadecimal (or octal or whatever) text. (Decimal is of course
also OK, but might not be as easy unless your compilers support (very) long
integers).
-- 
------------------------------------------------------------------------
Perry The Cynic (Peter Kiehtreiber)		     perry@arkon.key.com
** What good signature isn't taken yet? **	   ...!pacbell!key!perry