Floating point non-exactness

[Andrew Koenig]

In article <4958 at memqa.uucp>, r91400 at memqa.uucp (Michael C. Grant) writes:

> For example, assuming a standard which
> requires normalization, this algorithm might work (note this is 
> just a near-equality test, not a complete comparison):

> 	if (exponent_a!=exponent_b) return false
> 	if abs(mantissa_a-mantissa_b)<error_limit return true
> 	   else return false


You can use frexp() to split numbers into mantissa and exponent
(though I forget whether frexp is actually part of the ANSI standard
or not).  However, the algorithm shown here doesn't work because
the two numbers being compared could straddle a power of 2,
which would give them different exponents even though they're
arbitrarily close together.
-- 
				--Andrew Koenig
				  ark at europa.att.com