Re: GdkSpan

[Owen Taylor <otaylor redhat com>]

Alexander Larsson <alla lysator liu se> writes:

> On 20 Nov 2000, Owen Taylor wrote:
> 
> > 
> > Alexander Larsson <alla lysator liu se> writes:
> > 
> > > For my next checkin to GtkFB i need to do some generic Gdk changes. I
> > > added a new type GdkSpan, and two functions that handle intersection these
> > > with regions.
> > > 
> > > Is this patch ok? I'd also be very glad if someone (owen?) reviewed the
> > > intersection functions, since these kind of functions are very sensitive
> > > to boundary condition errors, and quite hard to test.
> > 
> > Hmmm, I'm basically OK with the addition, but some questions about the interface
> > details:
> > 
> >  - How are you going to use these functions?
> > 
> >    It worries me a bit since the typical case is taking a graphics
> >    primitive and breaking it into a set of spans. Taking those spans
> >    and intersecting them one-by-one with a clip region is 
> >    going to be _slow_ ...
> 
> They are used by the GtkFB rasterization code. Currently
> gdk_region_intersect() is used instead.

Which may well be faster...
 
> How would you do it then? Consider a wide arc with dashes. Rasterizing
> this to a Drawable using a GC with a clip region in GtkFB currently
> generates the spans from the arc, then clips each span to the GC region.
> 
> Clipping the dashed arc to the clipping region first and then directly
> rendering each span is just undoable. The intersection of the clip region
> and the arc is far to complex.
> 
> Note that there can be "shortcuts" for certain primitives when the GC has 
> a "simple" state.

I agree that generating spans and clipping them is the right thing to
do. My doubts come from the fact that there the possibility of very
significant optimization from having a routine that takes an array of
spans in banded order and does the clipping on all of them at once,
rather than clipping them one by one.

> A foreach func/data would match the usage pattern much better and make
> this much cleaner. Then the need for the _simple case is removed too.
> 
> gdk_region_span_intersection_foreach(GdkRegion *region,
> 				     GdkSpan *span,
> 				     void (*func) (GdkSpan *span, 
>                                                    gpointer data),
>                                      gpointer data);
> 
> How about that?

Sounds good - though use a typedef for the function.

                                        Owen