2013/2/24 Werner LEMBERG
<address@hidden>
> I wanted to know your opinion on the rounding errors when splitting
> Bezier segments. [...]
>
> the alternative method would overflow one bit quicker for conic and
> 3 bits quicker for cubic segments.
I think this enhanced precision is only worth the effort if it
improves rendering. So: Do you have an example which gets enhanced?
In particular, which gets *visibly* enhanced? Otherwise, speed is
more important.
I agree. If I recall correctly this code was left as is intentionally, i.e.:
- for "small" glyph rendering, the rasterizer uses units of 1/1024 pixels.
- for "large" glyph rendering, it uses units of 1/64 pixels.
Due to the numerical stability of the computations, the total error won't exceed one unit in the current implementation. It's hardly something that is going to be noticeable, so simplicity / speed of implementation is more important.
Of course, I could be wrong, but it'd be nice to have a counter-example then.
David
Werner