Discrepancies in `arc_to` paths

[corranwebster]

When drawing a rounded rect using methods like the following:

def gen_rounded_rect_ccw(obj, x, y, w, h, r):
    obj.move_to(x + w/2, y)
    obj.arc_to(x + w, y, x + w, y + h/2, r)
    obj.arc_to(x + w, y + h, x + w / 2, y + h, r)
    obj.arc_to(x, y + h, x, y + h/2, r)
    obj.arc_to(x, y, x + w / 2, y, r)


def gen_rounded_rect_cw(obj, x, y, w, h, r):
    obj.move_to(x + w/2, y)
    obj.arc_to(x, y, x, y + h / 2, r)
    obj.arc_to(x, y + h, x + w / 2, y + h, r)
    obj.arc_to(x + w, y + h, x + w, y + h / 2, r)
    obj.arc_to(x + w, y, x + w / 2, y, r)

sometimes Kiva draws the entire boundary (eg. celiagg, qpainter, pdf backends):

and sometimes it does not (eg. agg, opengl, quartz backends):

I think given that the end point of the paths is the same as the start point, I think the first case is the correct one.

Also the PS and SVG backends fail completely to render (because basecore2d.GraphicsContext doesn't implement arc_to).

[corranwebster]

From AbstractGraphicsContext:

   If the tangent point on the line from the current point to ``(x1, y1)``
   is not equal to the current point, a line is drawn to it. Depending on
   the supplied ``radius``, the tangent point on the line from
   ``(x1, y1)`` to ``(x2, y2)`` may or may not be ``(x2, y2)``. In either
   case, the arc is drawn to the point of tangency, which is also the new
   current point.

That implies that the second case is what is expected: the path goes to the tangent point, not (x2, y2).