[cairo] Re: Avoiding seams

[mental at rydia.net]

Quoting Timothée Lecomte <timothee.lecomte at ens.fr>:

> So SATURATE is not the best solution. I think that drawing at a
> higher resolution has the same last problem, unless I draw
> everything on the higher-resolution context.

You'd need to draw everything which needs to meet cleanly on that
same context, yes.

> I have followed with great interest the discussion between
> MenTaLguY and Bill Spitzak about a possible new entry in the API.
> I am not convinced at all by "aliased_lineto"
> or "aliased_curveto". It implies to detect which sides of the
> polygons need to be aliased to avoid seams, and it's not simple
> at all !

It's actually a rehashing of a very old discussion.  It isn't just a
problem for drawing e.g. adjoining triagles for
perspective-projected surfaces, but really a general issue which
affects any renderer which tries to approximate coverage via the
alpha channel for intermediate rendering steps.

Google for:

 "star and ring" cairo

and you'll find the original thread, along with the star-and-ring
test case I devised to show off the issue.

> Moreover, I want antialiased separations between my polygons, as
> they are of different colors.

Ahh, good catch.  I'd overlooked that shortcoming.

> Do you have any other ideas ?

Maybe two-sided fills.  But that's not really any easier to work
with than the aliased_lineto business, even if it did allow
antialiasing.

The only apparent solutions involve FSAA, which as far as I know
involves one of two families of techniques:

 - oversampling (render at higher resolution w/o antialiasing)

 - geometric (e.g. build a BSP tree of the entire scene and
rasterize in a single pass)

There are a couple other interesting ways of "cheating" -- for
example, the AGG renderer dialates paths slightly so they overlap. 
However, that approach only manages to avoid seams if the shapes
involved are all fully opaque.

As far as I can tell, the Macromedia Flash renderer uses
oversampling.  I don't know what Xara uses, but it'll be
open-sourced soon.  I suspect it also oversamples.

For what it's worth, I originally became aware of the issue with
Inkscape's current renderer (livarot).  Designers complaining
loudly about seams between their precisely aligned shapes.  Livarot
isn't cairo-based, but uses the same approach to partial coverage
approximation.  libart too.

I've got a nasty feeling that OSS renderers have been pursuing a
dead end in this regard.  It's a 90% solution that doesn't seem to
admit a 10% fix.

-mental