Thanks!  The patch is against the trunk.  (It doesn't quite apply cleanly 
to py3k, but the changes needed to make it do so should be minimal.)

Hmm. Rereading my previous comment, I seem to have a blindness for 
negative signs:

    gamma(x) = -pi/sinpi(x)/x/gamma(x)

should have been

    gamma(-x) = -pi/sinpi(x)/x/gamma(x)

and

    (-256.0, 1/256.0] + [1/256.0, 256.0)

should have been

    (-256.0, -1/256.0] + [1/256.0, 256.0)