On 3/21/11 10:25 PM, Don wrote:
Andrei Alexandrescu wrote:
I think by and large failure to define rational for BigInt in a way
that has many commonalities with rational for built-in integrals
reflects a failure of BigInt. By its very design, BigInt is supposed
to transparently replace integers.If this is largely the case, the design
has succeeded. If the design of anything must be reworked essentially
from scratch to work with BigInt instead of int et al, then the
abstraction may be too porous to be useful.
There are a few approaches we can take from here. One is to define
certain traits that differentiate BigInt from other integrals (e.g.
preferAdditionToMultiplication or whatnot), and then design Rational to
use those traits. Another is of course to specialize the entire
Rational on BigInt. Third would be to specialize certain core routines
(gcd and friends) for BigInt and keep Rational agnostic.
You're missing something important here. Rational big integers are a
very well defined area of research, with its own set of algorithms.
Although you can pretend a BigInt behaves as an int, and make a rational
type, the performance will be pathetic, and everyone will laugh at you.
Interesting. Thanks for the insight (missed this post for a while).
Andrei
