>>>>> "Robert" == Robert Dodier <robert.dodier at gmail.com> writes:
Robert> I'm not in favor of providing a signed zero for floating
Robert> point implementations which lack it. I think that implies
Robert> the bigfloat implementation shouldn't have one either.
This is acceptable to me. Having signed zeroes certainly helps me to
think about branch cuts, though. Kahan's article does show that they
are useful.
Now we need to examine all of special functions and their
implementations to make sure that we are consistent with an unsigned
zero.
Ray