MetaShapedTexture is the class responsible for drawing the contents of a client window, applying transformations and masks and etcetera. Right now, it is implemented as a ClutterActor subclass that also happens to semi-implement MetaCullable.
This MR turns MetaShapedTexture into a GObject subclass that implements ClutterContent, and is attached to MetaSurfaceActor.
Depends on: !405 (merged)
Performance
Turns out, there is a theoretical performance gain from this MR. Before, each window would have 1 + 2 * number of surfaces actors, which would generally mean 3: a MetaWindowActor, a MetaSurfaceActor, and a MetaShapedTexture. With ClutterContent, each window has 1 + number of surfaces, which generally means 2: a MetaWindowActor, a MetaSurfaceActor. So at the very least, it reduces the number of window actors by 33%.
If you consider subsurfaces, e.g. a window with an open popover, well, the picture starts to change: before that would mean 5 actors, now that's 3. I guess you can see where this is going.
"That is not a lot", one might say. But it turns out to be significant due to an easily ignored detail: all these windows are children of MetaWindowGroup. And MetaWindowGroup traverses the actor tree 3 times on each repaint.
To Do:
-
Figure out broken input coordinates -
Measure impact on painting times