From: "Roberto Tomás Collins McCarthy" <redmine@...>
Date: 2010-05-30T18:41:44+09:00
Subject: [ruby-core:30513] [Bug #3365] floats revisited (see bug 1841)

Bug #3365: floats revisited (see bug 1841)
http://redmine.ruby-lang.org/issues/show/3365

Author: Roberto Tom��s Collins McCarthy
Status: Open, Priority: Normal
ruby -v: ruby 1.9.1p378 (2010-01-10 revision 26273) [powerpc-linux]

someone posted that the float errors were just fine by them, referring to IEEE 754. I suggest visiting that standard yourself before closing the bug. If you look at the bug as reported, there is a rounding error on the 16th digit. IEEE 754 provides precision through 16 digits on doubles (binary64), the first rounding (eg, error) can only occur at the 17th digit.

I can confirm on my system that C does not produce the same results as ruby in this regard:
bash$ gcc -o fp fp.c && ./fp && ruby -e 'puts "\nfp.c\n������������\n"' && cat fp.c && ruby -e 'puts "\nruby #{RUBY_VERSION}: #{100.0 - 91.6}"'
System double representaiton: 8.400000e+00
8.40000000000000568434188608080148696899414062500000e+00
8.40000000000000568434188608080148696899414062500000e+00
sizeof doubles 8
sizeof long doubles 16

fp.c
������������
#include <stdio.h>
#include <math.h>

int main(){
 double cien = 100.0;
 double frac = 91.6;
 long double ciento = 100.0;
 long double fract = 91.6;
 printf("System double representaiton: %e\n", cien - frac);
 printf("%.50e\n", cien - frac);
 printf("%.50Le\n", ciento - fract);

 printf("sizeof doubles %d\n", sizeof(cien));
 printf("sizeof long doubles %d\n", sizeof(ciento));
 return 0;
}
ruby 1.9.1: 8.40000000000001


independently of whether ruby would or would not conform to IEEE754 the fact that C does not see doubles the same as ruby can be seen as a bug. as can the fact that right now elementary math any 2nd grader could expect to use with ruby can fail (100.0 - 91.6 == 8.4).


----------------------------------------
http://redmine.ruby-lang.org