Re: [MacPerl] Math::BigFloat Question

[Ronald J Kimball <rjk@linguist.dartmouth.edu>]

On Sat, Dec 11, 1999 at 06:55:55PM +0000, Na'im Tyson wrote:
> 
> I'm having some problems using Math::BigFloat.  I was using it for a
> research project, but I had to come up with another quick fix.  In any
> event, my problem is that the following subroutine returns 0 and position
> blank.  For each comparison that's made, 0 continues to be the largest,
> which shouldn't be the case.
> 
> Now, I'm sure that the problem lies w/ myself, however, I don't know that
> it is.  If anyone has any suggestions, please let me know.

I'm not so sure, actually.  This appears to be a bug in Math::BigFloat.

Here are two of the values you're comparing:

  0.0009 is +9E-4.  0 is +0E+0.

Here's Math::BigFloat::fcmp() :

  # compare 2 values returns one of undef, <0, =0, >0
  #   returns undef if either or both input value are not numbers
  sub fcmp #(fnum_str, fnum_str) return cond_code
  {
      local($x, $y) = (fnorm($_[$[]),fnorm($_[$[+1]));
      if ($x eq "NaN" || $y eq "NaN") {
          undef;
      } else {
          ord($y) <=> ord($x)
          ||
          (  local($xm,$xe,$ym,$ye) = split('E', $x."E$y"),
               (($xe <=> $ye) * (substr($x,$[,1).'1')
                || Math::BigInt::cmp($xm,$ym))
          );
      }
  }

The line ($xe <=> $ye) * (substr($x,$[,1).'1') compares the exponents of
the two values; because 0 is greater than -4, fcmp determines incorrectly
that +0E+0 is greater than +9E-4.

Math::BigFloat's documentation is rather intriguing:

  BUGS

  The current version of this module is a preliminary version of the
  real thing that is currently (as of perl5.002) under development.

even in the Math::BigFloat bundled with 5.005_03.  Apparently
Math::BigFloat was never fully developed.


Ronald

# ===== Want to unsubscribe from this list?
# ===== Send mail with body "unsubscribe" to macperl-request@macperl.org