[cairo] Clip region problems

[Owen Taylor]

 * For exact rectangular clips, we change the clip region on the
   destination surface, but we don't necessarily reset it
   when we are done drawing to the surface.

   We also don't do anything to guard against the same surface
   being used as the target of two cairo_t, or the target of
   a cairo_t and a source for drawing to another cairo_t.

 * In libpixman and Render, a clip region set on a source affects
   rendering... clipped pixels are treated as fully transparent.
   This isn't particularly useful behavior, but because of the
   above problems, it will leak out and affect Cairo rendering
   in some cases.

So, how do we fix these problems? Ideas:

 1. Pass the clip region as an explicit argument to all drawing
    functions. If the surface needs to set a clip region, it 
    sets it, then unsets it afterwards.

    Pro: simple
    Con: expensive

 1b. Pass the clip region as an explicit argument to all drawing
    functions, modify pixman_region_t to make it efficient.

    If you made pixman_region_t a copy-on-write wrapper for
    a refcounted object, then you could easily provide cheap

     void pixman_region_copy(pixman_region_t *region)
     int  pixman_region_definitely_equal(pixman_region_t *a,
                                         pixman_region_t *b);

    So, it would be easy for the backend to avoid changing the
    clip when it didn't need to.

    Pro: fairly simple, 
    Con: major pixman_region_t changes
    
 2. Have a "drawing state" for a surface. So, something like:

    cairo_surface_draw_state_t *cairo_surface_begin_draw (cairo_surface_t *surface);
    void cairo_surface_end_draw (cairo_surface_t            *surface,
                                 cairo_surface_draw_state_t *state);
 
    And maybe some copy or save/restore functionality to integrate with
    the gstate stack.

    The idea here is that this would allow the target surface to realize
    that two cairo_t's were targetting it and use different Picture
    or pixman_image_t objects for each use.

     Pro: allows a lot of backend flexibility, extensible to other
          similar situations.
     Con: complex, makes writing new backends hard, changes all over
          the place

 3. Kill the rectangular clip optimization, but expose device
    space region clips for particular backends. GTK+, would prefer
    
     cairo_xlib_surface_set_clip_region ()

    To the current clip API. Maybe users would also be happy with
    a surface-based clip. 

     Pro: Simple
     Con: API ugliness, still possible problems using the same surface
          as source and dest.

 4. Only allow a surface to be used for a single thing at a same time

     - two cairo_t's cannot be active for the same target surface
     - a surface that is the target of a cairo_t cannot be used
       as a source.

    Pro: Simple
    Con: Fairly big restriction

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://lists.freedesktop.org/archives/cairo/attachments/20050311/9a3301ca/attachment.pgp