RE: [ft-devel] more thoughts on cubic spline flattening

[David Bevan]

I've added some trace code to gray_render_cubic to report the average number of 
line segments created for a Bezier arc, and to calculate for each segment the 
(exact) deviation of the arc segment it replaces.

I've still only tested this with the pre-2.4.0 code. Here are the results with 
my test data: 

For screen resolutions, an average of about 9 segments are generated per arc, 
with an average deviation of about 1.5 (x 1/64 pel) and a maximum deviation of 
about 11 (x 1/64 pel).

For printer resolutions, an average of about 12 segments are generated per arc, 
with an average deviation of about 2.5 (x 1/64 pel) and a maximum deviation of 
about 20 (x 1/64 pel).

In both cases, about 20% of the segments have a deviation less that 0.5 (x 1/64 
pel).


Assuming I have the time, next week I'll attempt to implement something based 
on Hain's paper. I will also do some profiling (timing) of the code, so we can 
compare performance.


David %^>