From: Bil Kleb <W.L.Kleb@...>
Date: 2002-02-01T09:34:15+09:00
Subject: Re: smallest non-zero number (and other limits)

It looks like it has been a year since I first posed
this question [ruby-talk:9948].

Our current answer is to run the following c-program
on for each machine we use:

#include <float.h>
#include <stdio.h>
#include <string.h>

int main(){
  FILE *rubyFloat = fopen("Float.rb","w");

  fprintf(rubyFloat,"class Float\n");
  fprintf(rubyFloat," FLT_EPSILON = %.*e\n", FLT_DIG, FLT_EPSILON);
  fprintf(rubyFloat," DBL_EPSILON = %.*e\n", DBL_DIG, DBL_EPSILON);
  fprintf(rubyFloat,"end\n");
  return 0;
}

which generates an extension to the Float class:

 class Float
  FLT_EPSILON = 1.192093e-07
  DBL_EPSILON = 2.220446049250313e-16
 end

Of course it would be nice if these (and many other similar machine-
dependent limits available via float.h) were part of Ruby proper
(see RCR#37 at http://www.rubygarden.com/index.php?topic=RCR).

These constants come in handy for RubyUnit's tolerence argument
to assert_float_equal and for a host of other numerical analysis
situations.

FWIW,
--
Bil Kleb
NASA Langley Research Center
Hampton, Virginia, USA