The current version of SVG contains 3 pipelines:
Simple polygons:
path_storage ->
conv_curve ->
conv_transform ->
conv_clip
Strokes:
path_storage ->
conv_curve ->
conv_stroke ->
conv_transform ->
conv_clip
Contoured polygons:
path_storage ->
conv_curve ->
conv_contour ->
conv_transform ->
conv_clip
> And is it doubly calculated for
> filling, then stroking (in
> case both apply)?
It is. Otherwise you would have to store the
intermediate result somewhere. But the good news is
that conv_curve works pretty fast, at least stroking
and clipping are much more time consuming tasks.
The only obvious case when we can approximate the
curves in the last step is drawing text. But again,
when it's without stroking/outlining/contouring.
> Also what exactly does conv_contour do?
Exactly half of the work conv_stroke does. :-) It
dilates or erodes polygons depending on the sign of
the width. I used it to eliminate the defects when
joining anti-aliased polygons:http://www.antigrain.com/img/polygon_join_defects.jpg> I hope you dont mind all these questions :)
Of course I don't. Afterall I myself started cooking
this porridge :-)
> If so, doesnt this give the same result each time,
> as its done before the
> conv_transform?
Yes, it does, at least seems to do :-)
But the reality is more complex. Here's a
contadiction. Yes, we could transform paths before
approximating them with line segments. But conv_stroke
and conv_contour require already decomposed path. So,
the only case we can use this order (conv_transform ->
conv_curve) is a simple filled polygon without a
"border". But the situation is even worse. I use
conv_clip that performs polygonal clipping. Such kind
of a "vectorial" clipping can work with line segments
only, so, the curves must be decomposed before
clipping. Ideally it would be fine to decompose curves
as late as possible, but if the order of the
conversions is different solid and stroked paths will
be inconsistent:http://www.antigrain.com/img/conv_order.gifEventually I decided to set my jaw and to convert the
curves in the first pipeline step. Well, this is the
whole idea of the custom pipelines - in certain cases
you can use different conversion order for the sake of
performance.