NV/IV Bug ?

MZSanford has asked for the wisdom of the Perl Monks concerning the following question:

The following code does not work as i would expect. I was wondering if anyone knows why 1.15*100 acts diffrent under int(). My original code is below :

use strict;

my $key = 12345;
my @amn = (3.95,1.15);
my ($s,$s2);

for (@amn) {
  $s += int($_*100);
  $s2 += $_*100;
}

print "Scalar 1 : $s\n";
print "Scalar 2 : $s2\n";
[download]

This prints :

Scalar 1 : 509
Scalar 2 : 510
[download]

This was tested on

Windows 2000 running ActiveState perl 5.6.1
Solaris (sparc) 2.8 running perl 5.8.0 and 5.5.3
RedHat 7.4 running perl 5.6.1

A shorter version of this would be :

perl -e 'print 1.15*100,"\n",int(1.15*100),"\n"'
[download]

I know NV conversion is always open to a fudge-factor, but this seems pretty tame to be effected by that. Any help is appriciated.

from the frivolous to the serious

Comment on NV/IV Bug ? Select or Download Code

Replies are listed 'Best First'.
Re: NV/IV Bug ? by Elian (Parson) on Feb 27, 2003 at 18:42 UTC
Why are you thinking that it's surprising? Floats can't represent anything that isn't a fraction of two (1/2, 1/4, 1/8, and so on) so since you've chosen a number that's unrepresentable, you're going to get bad results. Just one more good reason to not use IEEE floating point numbers to represent money...	[reply]
Re: Re: NV/IV Bug ? by MZSanford (Curate) on Feb 27, 2003 at 18:48 UTC
good point. Hence the `100` being used on the string which was read from a file. from the frivolous to the serious*	[reply] [d/l]
Re: NV/IV Bug ? (a fix) by tye (Sage) on Feb 27, 2003 at 19:35 UTC
Use `int( $_100 + 0.5 )` instead of `int($_100)`. (I saw lots of explanations posted but no solutions.) - tye	[reply] [d/l] [select]
Re: NV/IV Bug ? by IlyaM (Parson) on Feb 27, 2003 at 18:47 UTC
Another variation of this "bug": `$x = 114.99999999999999; print $x, "\n"; print int($x), "\n";` [download] Prints `115 114` [download] on my computer. -- Ilya Martynov, ilya@iponweb.net CTO IPonWEB (UK) Ltd Quality Perl Programming and Unix Support UK managed @ offshore prices - http://www.iponweb.net Personal website - http://martynov.org	[reply] [d/l] [select]
Re: NV/IV Bug ? by extremely (Priest) on Feb 27, 2003 at 18:48 UTC
1.15 in binary is... ooops, an infinite series. There is special magic in place for what is basically 1.1499999999999 when multiplied and printed. But when you you force to int(), you dodge that magic and get the proper truncated value. -- $you = new YOU; honk() if $you->love(perl)	[reply]
Re: Re: NV/IV Bug ? by MZSanford (Curate) on Feb 27, 2003 at 18:52 UTC
Thanks, that explains. Inconsistant (by design) Magic application. If neither worked, i would not have posted ... i do understand floating point numbers, but the inconsistancy was bothering me. Thanks again. from the frivolous to the serious	[reply]
Re: NV/IV Bug ? by Steve_p (Priest) on Feb 27, 2003 at 19:18 UTC
`int()` always rounds down. In cases where you are working with floating-points, the precision may not be what you expect. So, to paraphase `perldoc -f int`, your 395 and 115 may look more like 394.99999943242 or 114.999999523523. In either case, the decimal portion is truncated off. Instead, `sprintf, printf, POSIX::floor, or POSIX::ceil` may be better options for what you want to do.	[reply] [d/l] [select]
Re: NV/IV Bug ? by Anonymous Monk on Feb 27, 2003 at 18:40 UTC
int make int from number int (5)=5 int (5.9)=5 thats why	[reply]