[cairo] cairo anti aliasing

[Bill Spitzak]

Tim Janik wrote:

> right, which is why i'm interested in point sampling a gaussian integral 
> with
> varying radius, perpendicular to the object edge. not blurring it after 
> point
> sampling the object (by nearest neighbour, box filter or whatever).
> i.e. i *do* have the corner artefacts you call "mach-banded points" ;)

Okay, I see. We are describing the same thing. When I said "point 
sampling a very thin rectangle along the edge", my rectangle is colored 
in such a way that a line perpendicular to the edge is your filter, thus 
producing the same result.

I suspect at the dimensions we are talking about, a linear ramp will 
work just as well as a gaussian or bicubic curve, this has the advangage 
of being doable on OpenGL hardware.

My original statement that this is not the best antialiasing algorithim 
still stands. I personally use this algorithim in my own code and I am 
quite familiar with each and every failing of it. In particular it 
cannot correctly antialias any details that are smaller than 2 pixels 
thick. I do not think Cairo should be defined in any way to require this 
algorithim.