floating point test suite



On 8/7/09, Barton Willis <willisb at unk.edu> wrote:

> ;; Return true if $float converts both a and b to floats and
> ;; |a - b| <= float_approx_equal_tolerance * min(|a|, |b|).
> ;; In all other cases, return false.

Incidentally this definition is, I believe, unnecessarily strict for
quantities near zero. There are some tests which fail for some
Lisp implementations because e.g. 1e-16 isn't close enough to
1.1e-16 where these numbers are residual errors of some kind
(i.e. computed by subtracting much larger numbers which are
nearly equal).

FWIW

Robert Dodier