[cairo] Avoiding seams (again)

[Robert O'Callahan]

On Thu, 2006-03-23 at 12:49 -0500, MenTaLguY wrote:
> If you do supersampling on a per-group basis, that still doesn't solve the problem of objects in different groups which are adjacent.

I think Vlad was referring to extending the push/popgroup API so it can
be used for global supersampling, i.e. FSAA.

Actually there could be a win for us doing some sort of local
supersampling as one of a mix of strategies. For regular Web pages we
pixel-snap most drawing at rendering time. For certain drawing
operations we can use coverage-based AA when we know edges aren't
intended to be adjacent. Where we can't pixel-snap and we can't
guarantee unintended seams, we could supersample just the troublesome
region of the page. This would save resources in common situations, e.g,
when you have an SVG drawing (possibly containing HTML) as a small-ish
component of a regular HTML page.

Sadly the seams problem hit hard as soon as we started supporting
rotated HTML. Nearly the first thing someone tried was rotating Google
Maps, which is a grid of adjacent images...

Rob