Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!uwm.edu!zaphod.mps.ohio-state.edu!brutus.cs.uiuc.edu!apple!sun-barr!newstop!sun!urbina!nano
From: nano%urbina@Sun.COM (Nano)
Newsgroups: comp.sys.mac
Subject: Re: Testing for NaN in Think C (help!)
Message-ID: <133544@sun.Eng.Sun.COM>
Date: 27 Mar 90 21:40:48 GMT
References: <10538@portia.Stanford.EDU>
Sender: news@sun.Eng.Sun.COM
Reply-To: nano@sun.UUCP (Nano)
Organization: Sun Microsystems, Mountain View
Lines: 34

Melissa Hines writes:

>    if((test/test != 1) && (test != 0))
>        ; /* Here if NaN */
>      else
>        ; /* Here otherwise */
>Is there a simpler/ more elegant way of doing this?  I think part of the
>problem is that the IEEE have decreed that the result of any operation on
>a NaN is a Nan.  Also, I think Think C may have more than one type of NaN. I
>couldn't care less about the type, I just want to know if it is a NaN.

There are quiet NaN's and signalling NaN's.  They differ by whether the
MSBit of the fraction part of the number is 1 or 0.  (Fraction as in IEEE
fraction).  You could always look at the actual representation of the
number and determine whether it's a NaN, but the simpler way would be
to compare the number to itself.  The IEEE spec defines a comparison involving
a NaN as "unordered", thus, if you compare test to itself, if it's 
a NaN, the comparison will fail:

	if (test != test)
		; /* Here if NaN */
	else
		; /* Here otherwise */

I haven't tried it, but it should work this way if everything is
IEEE compatible.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 
Fernando A. Urbina              INTERNET:  nano@sun.com
                                       nano@mcimail.com
                                MCI:           408-0187

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%