Comment 2 for bug 230765

The problem is just that there are a couple of zeros that should be ones. I've attached a patch.

(This patch is essentially equivalent to Alvin Penner's, because

        dx0 = (x1 - x0)/ds # Only works for rectangular coordinates
        dy0 = fp(xstart)*dx0

in his version can be simplified to

        dx0 = 1 # Only works for rectangular coordinates
        dy0 = fp(xstart)

as (x1 - x0)/ds == 1, and similarly for dx1 and dy1.)