Re: Float numbers comparison

> a=1.9
> b= 1.8 +0.1
> puts a==b  #false
The Test::Unit library provides a special method to deal with that  
technical issue, called
assert_in_delta:

   Passes if +expected_float+ and +actual_float+ are equal
   within +delta+ tolerance.

   Example:
     assert_in_delta 0.05, (50000.0 / 10**6), 0.00001

Inside, it's using this formula:

   (expected_float.to_f - actual_float.to_f).abs <= delta.to_f

You can easily make a new Float#about_equal method out of this, or  
you use Float#===,
or even overwrite Float#==.

Or, if you prefer calculating with fractions, you can use the  
Rational library.

However, I wonder what the Float::EPSILON constant is about. Can it  
be used as a sensible
default tolerance delta?

1.0 / (1.0 / 0.0) == 0.0   # ^^
[murphy]