From: Patrick Hurley <phurley@...>
Date: 2006-11-15T06:32:10+09:00
Subject: Re: Is 2.0 Integer or Float?

But depending upon how the "integer" was generated, it may still make
more sense to use an epsilon in the test (which certainly has the
"smell" that was mentioned earlier)

Integer(100.0 * 9.95) == 995                      => false
(100.0 * 9.95) - 995.0 <= Float::EPSILON   => true

# So you find yourself tempted to do this
class Float
  def like_an_int
    (self - self.round).abs <= Float::EPSILON
  end
end

(100.0 * 9.95).like_an_int    => false

On my system Float::EPSILON = 2.22044604925031e-016
(100.0 * 9.95) - 995.0   => -1.13686837721616e-013

Which is way bigger than EPSILON, it turns out you need to scale your
"EPSILON" by the relative magnitude of the numbers being compared.
This way leads to madness and general discomfort --> if you find
yourself in this neighborhood, switch to rational numbers (or scaled
integers), you will be happier.

pth