Since floor uses bfloat internally, this seems like a very roundabout,
computationally expensive, and potentially incorrect way of calculating the
continued fraction of a power. It would be much more straightforward and
computationally efficient (though possibly just as likely to be incorrect)
to do cf(bfloat(...)),fpprec:xxx with some sort of loop increasing the
fpprec as needed. And that would work for *arbitrary* expressions, not just
a^b.
-s