Currently, property animators use lambda captures (std::function<>) to push
values to adapters and then to the scene graph. Some downsides:
* complex lambda captures are expensive in terms of object code size
* adapters with multiple animated properties don't synchronize/quiesce: each individual property tick triggers a SG
synchronization, possibly with inconsistent state (as animator running
order is unspecified)
* there is no enforced scoping, resulting in fragile constructs when SG
fragments are discarded
This CL introduces a simplified and more robust animator pattern:
* property animators are scoped to explicit containers
* instead of capturing arbitrary value functors, animators only capture
a pointer to the target value
Some implementation details:
* keyframe/interpolation logic is pretty much unchanged (just relocated)
* introduced AnimatablePropertyContainer - a base class for animatable
adapters
* legacy binding functions are refactored based on the new mechanism
(they now/transitionally inject adapter objects)
* converted a handful of effects, to exercise trivial refactoring patterns
* converted the text animator goo, to exercise non-trivial refactoring:
- detecting value changes is now trickier (no more lambda magic)
- value adjustments must be hoisted into adapter logic (no more lambda magic)
- all dependent animated values (selectors, etc) must be scoped to the
text adapter to avoid lifetime issues
TBR=
Change-Id: Ia5821982f251de0de58fd3f87812219ff7fcc726
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/263938
Commit-Queue: Florin Malita <fmalita@chromium.org>
Reviewed-by: Mike Reed <reed@google.com>
d5c42c8c03