NetNews Usenet Archive 1992 #19/NN_1992_19.iso/spool/comp/unix/bsd/4986

Text File | 1992-09-01 | 1.7 KB | 64 lines

Newsgroups: comp.unix.bsd
Path: sparky!uunet!mcsun!Germany.EU.net!rrz.uni-koeln.de!aix370!a0045
From: [e-mail] (Jochen Roderburg)
Subject: 386BSD FP problems, or: why is 0.0 == -1.0 ??
Message-ID: <[e-mail]>
Sender: [e-mail] (Usenet News System)
Organization: Regional Computing Center, University of Cologne, F. R. Germany
Distribution: comp
Date: 2 Sep 92 12:05:30
Lines: 52

The C system on 386BSD seems to have some strange problems with
floating-point constants.

The following test program justs prints out a few constants:

main () {
printf ("0.  %+f\n", 0. );
printf ("0.0 %+f\n", 0.0);
printf ("0.1 %+f\n", 0.1);
printf ("0.5 %+f\n", 0.5);
printf ("0.6 %+f\n", 0.6);
printf ("1.0 %+f\n", 1.0);
printf ("1.1 %+f\n", 1.1);
printf ("1.5 %+f\n", 1.5);
printf ("1.6 %+f\n", 1.6);
}

And these are the suprising results:

0.  +0.000000
0.0 -1.000000
0.1 +0.000000
0.5 +0.000000
0.6 +1.000000
1.0 +1.000000
1.1 +1.000000
1.5 +2.000000
1.6 +2.000000

Especially entertaining is the wonderful transformation from 0.0 to -1.0.
You may understand that a small utility which uses a few of these constants
brought not very useful results.

The errors seem to happen already in the compiler, not during the printout.
When you use the constants in expressions you get results that fit to the
printout, for example  2+0.0 -> 1.0.

I have observed this on PCs without mathematical coprocessor, if this is of
any importance.

Has anybody a clue what is going on here ?

Jochen Roderburg
Regional Computing Center
University of Cologne
Robert-Koch-Str. 10                    Tel.:   [phone]
D-5000 Koeln 41                        E-Mail: Ro @ RRZ.Uni-Koeln.DE
Germany