> > As for bezier curves, there is almost always an overflow because > > while solving polynomials some large numbers are involved. I even > > tried using Newton's approximation, but it's too slow and still not > > as accurate as floating points.
Not a good sign of numerical stability... > Perhaps you have to subdivide the Bezier curves, using the Bezier > envelope as a guide? I thought about it as well: subdivide Bezier curves until the segments are flat enough. The code can be borrowed from FreeType smooth renderer stopping short of calculating the coverages. > Or maybe you can do something similar to > Behdad's `GLyphy` implementation: He uses circular arcs to approximate > Bezier curves. ... if it is easy to work with constant curvature, then subdivide Bezier curves until the segments have fairly constant curvature.
