Backpedal on node/reval-time-determined damage: nodes cannot control
the invalidation order, and shared descendants may be revalidated before
a particular ancestor gets to query their state - thus making any
decisions based on that invalid.
Instead, apply damage suppression at invalidation time, based on node
type/traits. Node types which don't generate direct damage are marked
as such, and the invalidation logic bubbles damage past them, until it
finds a valid damage receiver.
Nodes which currently suppress damage:
- PaintNode (and subclasses)
- GeometryNode (and subclasses)
- Matrix
TBR=
Change-Id: I843e683e64cb6253d8c26d8397c44d02a7d6026f
Reviewed-on: https://skia-review.googlesource.com/91421
Reviewed-by: Florin Malita <fmalita@chromium.org>
Commit-Queue: Florin Malita <fmalita@chromium.org>
Node subclasses can now control whether their bounds (changes)
contribute to damage.
Tristate:
* Default: The node bounds contribute to damage if the node itself was
invalidated, observing hasSelfInval(). This is the default
behavior.
* ForceSelf: The node bounds contribute to damage, regardless of
hasSelfInval(). Used for domain-boundary nodes (e.g. Draw),
which gate blocked fragments (e.g. geometry, paint nodes).
* BlockSelf: The node bounds do not contribute to damage, regardless of
hasSelfInval(). Used for nodes which do not contribute
damage directly (e.g. paints, geometry).
TBR=
Change-Id: I7c941c7ea12e14b008d846ec13108e66e34dbc73
Reviewed-on: https://skia-review.googlesource.com/91104
Reviewed-by: Florin Malita <fmalita@chromium.org>
Commit-Queue: Florin Malita <fmalita@chromium.org>
We need to discriminate between nodes whose bounds updates contribute to the dirty
region, and nodes whose bounds changes do not.
E.g. animated shape in a group: the animated shape node bounds should yield damage,
but the ancestor group bounds should not.
To accomplish this, we refine the invalidation state:
1) self invalidation == the node itself was invalidated, and its bounds updates
yield damage.
2) descendant invalidation == the node has some (self-)invalidated descendant,
but its own bounds are not contributing damage.
Also:
* hoist the bounding box invalidation logic into the base class (Node::revalidate)
and update to respect the states described above.
* remove (now-redundant) GeometryNode bbox logic.
* update revalidation methods to return the node bbox instead of void
TBR=
Change-Id: I8023d1793fb501c945a53f2dc2d2983e5b620ade
Reviewed-on: https://skia-review.googlesource.com/90581
Reviewed-by: Florin Malita <fmalita@chromium.org>
Commit-Queue: Florin Malita <fmalita@chromium.org>
