Text editing shortcuts involving the arrow keys are no longer handled by RenderEditable's RawKeyboardListener, they use the new Shortcuts setup. First PR in a plan to port all text editing keyboard handling to shortcuts.
This re-lands #49235 with the addition of includeSemantics flag on the Focus widget so that the FocusTraversalGroup can create a Focus widget without affecting the semantics tree.
The FocusTraversalGroup uses the Focus widget to create a grouping of descendants for traversal, but doesn't actually participate in focus (canRequestFocus is always false), so we don't want it to add a Semantics widget in that case, since that can cause semantics changes. The canRequestFocus attribute can also be used when a widget is disabled, so we do sometimes want to include Semantics even if that is false, but not in the case where it is always false, as for FocusTraversalGroup.
- Added a test to make sure that FocusTraversalGroup doesn't add any semantics information.
This change adds a way to provide explicit focus order for a part of the widget tree.
It adds FocusTraversalPolicyGroup, which in many ways is similar to DefaultFocusTraversal, except that it groups a widget subtree together so that those nodes are traversed as a group. DefaultFocusTraversal doesn't work as one would expect: If there is more than one DefaultFocusTraversal inside of a focus scope, the policy can change depending on which node was asked to move "next", which can cause unexpected behavior. The new grouping mechanism doesn't have that problem. I deprecate DefaultFocusTraversal in this PR.
It also adds OrderedFocusTraversalPolicy, which is a policy that can be supplied to FocusTraversalPolicyGroup to set the policy for a sub-tree. It looks for FocusTraversalOrder inherited widgets, which use a FocusOrder to do the sorting. FocusOrder has two subclasses: NumericalFocusOrder (which sorts based on a double), and LexicalFocusOrder, which sorts based on a String.
As part of doing this, I refactored the way FocusTraversalPolicy is implemented so that it has more default implementation methods, and exposes a new protected member: sortDescendants, which makes it easier for developers to make their own policy subclasses: they only need to implement sortDescendants to get a new ordering behavior, but can also still override any of the default implementation behaviors if they need different behavior.
I was able to do this without breaking the API (AFAICT).
