[PATCH weston] compositor: add a masking mechanism to weston_view

[Pekka Paalanen]

On Tue,  3 Dec 2013 14:29:21 +0100
Giulio Camuffo <giuliocamuffo at gmail.com> wrote:

> This adds a 'mask' pixman_region32 to weston_view and to
> weston_view::geometry. A user, a shell probably, can set a view's
> mask to some region (e.g. the workspace area) and the view, and all
> its transform children, will be clipped to that region.
> weston_view::geometry.mask is the intersection of the view's mask and
> of the parent's geometry.mask.
> ---
>  src/compositor.c | 25 +++++++++++++++++++++++--
>  src/compositor.h |  2 ++
>  2 files changed, 25 insertions(+), 2 deletions(-)
> 
> diff --git a/src/compositor.c b/src/compositor.c
> index 32e72b1..8316653 100644
> --- a/src/compositor.c
> +++ b/src/compositor.c
> @@ -366,6 +366,8 @@ weston_view_create(struct weston_surface *surface)
>  	view->plane = NULL;
>  
>  	pixman_region32_init(&view->clip);
> +	region_init_infinite(&view->mask);
> +	pixman_region32_init(&view->geometry.mask);
>  
>  	view->alpha = 1.0;
>  	pixman_region32_init(&view->transform.opaque);
> @@ -926,6 +928,11 @@ weston_view_update_transform(struct weston_view
> *view) 
>  	weston_view_damage_below(view);
>  
> +	if (parent)
> +		pixman_region32_intersect(&view->geometry.mask,
> &view->mask, &parent->geometry.mask);
> +	else
> +		pixman_region32_copy(&view->geometry.mask,
> &view->mask); +
>  	pixman_region32_fini(&view->transform.boundingbox);
>  	pixman_region32_fini(&view->transform.opaque);
>  	pixman_region32_init(&view->transform.opaque);
> @@ -1185,10 +1192,14 @@ weston_compositor_pick_view(struct
> weston_compositor *compositor, wl_fixed_t *vx, wl_fixed_t *vy)
>  {
>  	struct weston_view *view;
> +	int ix = wl_fixed_to_int(x);
> +	int iy = wl_fixed_to_int(y);
>  
>  	wl_list_for_each(view, &compositor->view_list, link) {
>  		weston_view_from_global_fixed(view, x, y, vx, vy);
> -		if
> (pixman_region32_contains_point(&view->surface->input,
> +		if
> (pixman_region32_contains_point(&view->geometry.mask,
> +						   ix, iy, NULL) &&
> +
> pixman_region32_contains_point(&view->surface->input,
> wl_fixed_to_int(*vx), wl_fixed_to_int(*vy),
>  						   NULL))
> @@ -1274,8 +1285,10 @@ weston_view_destroy(struct weston_view *view)
>  	wl_list_remove(&view->link);
>  	wl_list_remove(&view->layer_link);
>  
> +	pixman_region32_fini(&view->mask);
>  	pixman_region32_fini(&view->clip);
>  	pixman_region32_fini(&view->transform.boundingbox);
> +	pixman_region32_fini(&view->geometry.mask);
>  
>  	weston_view_set_transform_parent(view, NULL);
>  
> @@ -1481,7 +1494,15 @@ view_accumulate_damage(struct weston_view
> *view, pixman_region32_union(&view->plane->damage,
>  			      &view->plane->damage, &damage);
>  	pixman_region32_fini(&damage);
> -	pixman_region32_copy(&view->clip, opaque);
> +
> +	/* set view->clip with the part of
> view->transform.boundingbox
> +	 * that doesn't fit into view->geometry.mask, then add the
> opaque region
> +	 * to it. We don't do the opposite, adding view->clip to
> opaque,
> +	 * because opaque is then passed to the next views, which
> may have
> +	 * a different mask.
> +	 */
> +	pixman_region32_subtract(&view->clip,
> &view->transform.boundingbox, &view->geometry.mask);
> +	pixman_region32_union(&view->clip, &view->clip, opaque);
>  	pixman_region32_union(opaque, opaque,
> &view->transform.opaque); }
>  
> diff --git a/src/compositor.h b/src/compositor.h
> index 3e878c8..e0b3895 100644
> --- a/src/compositor.h
> +++ b/src/compositor.h
> @@ -750,6 +750,7 @@ struct weston_view {
>  	struct wl_list layer_link;
>  	struct weston_plane *plane;
>  
> +	pixman_region32_t mask;
>  	pixman_region32_t clip;
>  	float alpha;                     /* part of geometry, see
> below */ 
> @@ -762,6 +763,7 @@ struct weston_view {
>  	struct {
>  		float x, y; /* surface translation on display */
>  		int32_t width, height;
> +		pixman_region32_t mask;
>  
>  		/* struct weston_transform */
>  		struct wl_list transformation_list;

Hi Giulio,

these new fields are structured a little unexpectedly IMO. Originally,
everything under 'geometry' is modifiable, and modifying them requires
one to also call weston_view_geometry_dirty(), so that all
transform-children will be updated too. The state under 'transform' are
read-only, derived from 'geometry' state as needed (marked by the dirty
flag).

Unfortunately, partly due to hysterical raisins(*), there are bits of
'geometry' also elsewhere, like the 'alpha' seen above, and
weston_surface::width,height.

Would it not be more consistent to have geometry.mask as the settable
field, and transform.mask as the used field that also integrates the
parent state?


Thanks,
pq

* Oh how I've been waiting to get to use that phrase. ;-)