🚀 go-pugleaf

roedy@mindprod.com (Roedy Green) writes:

(regarding decimal floating point)

>Mike Cowlishaw (The inventor of Rexx and NetRexx) would agree with
>you.  He has been lobbying for floating point hardware to be
>redesigned to work in base 10 rather than base 2 to avoid these
>anomalies.  People are used to .3333... + .66666 = .9999 but they
>don't expect 0.1 to behave the same way.

I tried to discuss this on comp.arch, mostly saying that chip density
is high enough to add decimal floating point, along with binary.

>I disagree.  Binary arithmetic is faster and simpler.  The repeaters
>just move to different places.  You can't get rid of them.  

With appropriate hardware, decimal floating point would be fast enough.

>If you
>want absolute precision, use scaled integer arithmetic which is many
>times faster that floating point (e.g. integer pennies for currency,
>with a decorative decimal point inserted for display).

I do believe that fixed point arithmetic should be used for all financial
calculations, and, like D.E.Knuth, for all typesetting calculations.

-- glen