On Thu, Feb 24, 2011 at 7:31 PM, Merlin Moncure <mmoncure@gmail.com> wrote:
> the root issue I think here is that the string version of the double
> precision math is approximated:
No, it's simpler than that, all double precision math is approximated.
The root issue is that 2183.67 is not representable in a floating
point binary number. Just like 1/3 isn't representable in base 10
(decimal) numbers many fractions aren't representable in base 2
(binary) numbers. The result are repeated decimals like 0.3333... if you
multiply that by three you get 0.99999 and if you truncate that you
get 0 insted of 1.
It's the trunc() that's exposing the imprecision because like "=" it
depends on the precise value of the number down to the last digit.
Though depending on the arithmetic you can always make the precision
expand beyond the last digit anyways -- when you multiply by 100 you
magnify that imprecision too.
--
greg