-----Richard Fateman wrote: -----
> I believe the commercial Macsyma has a sophisticated program
> involved in this, courtesy of Bill Gosper, which allows the system
> to decide how much precision is required to determine the
> sign of (or,especially, the zeroness of ..) an expression.
Too bad we don't have the source -- with Macysma 2.2, it's
easy to break 'floor':
(c1) p : sqrt(5) * (sqrt(5) - 1/sqrt(5))$ /* p = 4 */
(c2) floor(p);
(d2) 3 <--- bogus
(c3) floor(expand(p));
(d3) 4 <--- OK
Maybe this was fixed in later versions.
Barton