gdkinternals: Document our drawing and painting process better

This commit is contained in:
Jasper St. Pierre 2014-11-22 09:07:51 -08:00
parent c01e37a9a5
commit 10a7a89311

View File

@ -216,15 +216,82 @@ struct _GdkWindow
cairo_pattern_t *background;
/* The paint logic here is a bit complex because of our intermingling of
* cairo and GL. Let's first go over the cairo-alone case:
*
* 1) gdk_window_begin_paint_region() is called with an update region. If
* the backend wants it, we redirect drawing to a temporary surface
* sized the same as the update region and set `surface_needs_composite`
* to TRUE. Otherwise, we paint directly onto the real server-side window.
*
* 2) Things paint with cairo using gdk_cairo_create().
*
* 3) When everything is painted, the user calls gdk_window_end_paint().
* If there was a temporary surface, this is composited back onto the
* real backing surface in the appropriate places.
*
* This is similar to double buffering, except we only have partial surfaces
* of undefined contents, and instead of swapping between two buffers, we
* create a new temporary buffer every time.
*
* When we add GL to the mix, we have this instead:
*
* 1) gdk_window_begin_paint_region() is called with an update region like
* before. We always redirect cairo drawing to a temporary surface when
* GL is enabled.
*
* 2) Things paint with cairo using gdk_cairo_create(). Whenever
* something paints, it calls gdk_window_mark_paint_from_clip() to mark
* which regions it has painted in software. We'll learn what this does
* soon.
*
* 3) Something paints with GL and uses gdk_cairo_draw_from_gl() to
* composite back into the scene. We paint this onto the backing
* store for the window *immediately* by using GL, rather than
* painting to the temporary surface, and keep track of the area that
* we've painted in `flushed_region`.
*
* 4) Something paints using software again. It calls
* gdk_window_mark_paint_from_clip(), which subtracts the region it
* has painted from `flushed_region` and adds the region to
* `needs_blended_region`.
*
* 5) Something paints using GL again, using gdk_cairo_draw_from_gl().
* It paints directly to the backing store, removes the region it
* painted from `needs_blended_region`, and adds to `flushed_region`.
*
* 6) gdk_window_end_paint() is called. It composites the temporary surface
* back to the window, using GL, except it doesn't bother copying
* `flushed_region`, and when it paints `needs_blended_region`, it also
* turns on GL blending.
*
* That means that at any point in time, we have three regions:
*
* * `region` - This is the original invalidated region and is never
* touched.
*
* * `flushed_region` - This is the portion of `region` that has GL
* contents that have been painted directly to the window, and
* doesn't have any cairo drawing painted over it.
*
* * `needs_blended_region` - This is the portion of `region` that
* GL contents that have part cairo drawing painted over it.
* gdk_window_end_paint() will draw this region using blending.
*
* `flushed_region` and `needs_blended_region` never intersect, and the
* rest of `region` that isn't covered by either is the "opaque region",
* which is any area of cairo drawing that didn't ever intersect with GL.
* We can paint these from GL without turning on blending.
**/
struct {
cairo_region_t *region;
/* The temporary surface that we're painting to. This will be composited
* back into the window when we call end_paint. This is our poor-man's
* way of doing double buffering. */
cairo_surface_t *surface;
/* Areas of region that have been copied to the back buffer already */
cairo_region_t *region;
cairo_region_t *flushed_region;
/* Areas of region that have been copied to the back buffer but
needs furter blending of surface data. These two regions are
always non-intersecting. */
cairo_region_t *need_blend_region;
gboolean surface_needs_composite;