On Wed, Jun 3, 2020 at 10:48 PM Anuj Verma <an...@iitbhilai.ac.in> wrote: > > Hello Behdad, Jim, > > > I see that in your implementation you are converting cubics to quadratic > > spline. That makes a lot of sense and I support that. > > Is it better to split the cubic bezier into quadratic one? why? Currently I > am using Newton-Raphson's iteration to find the > closest point on cubic curves, which is a bit slow but works well and > produces accurate results. (https://imgur.com/QSGCoIj)
Indeed, iterations or bisections might be better if some analytical solution for quadratic curves has a square root, which is not allowed in FreeType. Behdad will send me to my cave again, of course... > > > I think we do want a 8bit representation at least, and possible a higher > > one. > > Currently I am using 32bit floats, but since freetype doesn't use floats I am > thinking of using 32bit 16.16 fixed point instead. The problem with 16.16 is > that > most of the bits in the integer part will not be used because I am > normalizing the > values between [-1.0, 1.0] before outputting the data. Is 1/65563 not enough accuracy for you? Do not be greedy. What would you gain from extra accuracy in the world of discrete pixels? Alexei
