feat!: Add support for keyboard navigation

[gonfunko]

The basics

• I validated my changes

The details

Proposed Changes

This PR backports the core support for keyboard navigation of the workspace, toolbox and flyout from the blockly-keyboard-experimentation repo. It also refactors things to provide a more cohesive API for interacting with navigation. The general aspirational approach is:

• The various *_navigation_policy.ts files specify, for each kind of focusable/navigable entity in Blockly, what its parent, first child, and previous/next siblings are, from a "just the facts"/AST perspective.
• Navigator provides methods to query the "just the facts" relationships between elements, but also provides methods that incorporate business logic to make navigation order in response to the arrow keys make sense/feel right/pass the vibe test.
• FlyoutNavigator and ToolboxNavigator customize the business logic for navigation in a flyout and toolbox context, respectively, and operate on a set of navigation policies relevant to focusable/navigable entities that exist in those contexts
• Every IFocusableTree provides a Navigator instance which should be used to handle navigation within that tree
• The actual shortcuts for the arrow keys are effectively one-liners that delegate to the Navigator for the active focus tree

This means that keyboard navigation does not have to keep track of the current state/context as in the keyboard-experimentation repo, because that maps precisely to the currently focused tree, which the FocusManager already keeps track of.

I backported the navigation business logic from the add-screen-reader-support-experimental branch, where up/down move between "rows" and left/right navigate within the current "row".

Future Work

• Navigation within the toolbox needs to be adjusted to accomodate horizontal layouts and the various start/end toolbox positions
• Some CSS rules need to be backported to make it clearer where focus is
• The navigation policies should be cleaned up a bit to remove some vestiges of business logic
• Additional tests for navigation behavior should be added; the suite from the keyboard-experimentation repo was backported with LLM assistance and manual review, but additional cases should be exercised, particularly with looping enabled
• Enter/space to act on the focused element still needs to be backported

Breaking Changes

• MarkerManager, Marker, and LineCursor have all been removed, along with their accessors. If you were interacting with one of these classes, Navigator should be close to a drop-in replacement.
• IFocusableTree must implement getNavigator(), which should return a Navigator instance or subclass appropriate for navigating the tree
• Flyout no longer implements IFocusableNode or IFocusableTree. The flyout's workspace continues to implement those interfaces, and should be used in place of the flyout itself.
• Changes have been made to several toolbox-related interfaces. If you have custom toolbox-related classes that do not inherit from those provided in Blockly, you may need to adjust your implementation to ensure that:
  • IToolbox provides a getToolboxItems() method that returns an array of all items
  • IToolboxItem provides a getParentToolbox() method that returns a reference to its containing toolbox

[gonfunko]

This fixes RaspberryPiFoundation/blockly-keyboard-experimentation#241

[maribethb]

should this change be here? it's different than all the others

[maribethb]

The jsdoc no longer matches the implementation and I can't tell which one should be correct.

Or is the behavior the same but you moved the responsibility of returning null to the policy? I would still probably clarify the jsdoc in that case.

[maribethb]

curious why this is needed - what scenarios are coming up where a flyout item isn't included in the flyout contents?

[gonfunko]

The thing being checked isn't necessarily a flyout item; this prevents things that aren't top-level flyout items (e.g. icons, fields, connections) from becoming focused by indicating that they aren't navigable.

[maribethb]

this comment doesn't seem to match reality - it returns null, not the workspace. i'm not sure why it would return the workspace?

[maribethb]

can you add jsdoc for this please?

[maribethb]

what's the reason for the cast here and is there a way we can eliminate it?

[maribethb]

There is a LOT of logic in this function. I was expecting more of this logic to live in the navigation policies themselves rather than have so much business logic in the navigator. It's confusing to me that you have some logic here but some similar-seeming business logic about allowable possible navigation stops in the block navigation policy for example.

Possibly we should discuss this over video call because of the complexity

If this logic has to stay here I think it's worth investing in it to make this as readable as possible. Maybe start by splitting up the in/out vs next/prev into different named functions to remove one layer of logic?

[maribethb]

This needs more comments explaining, or maybe it would be more readable if you split up the conditions into separate statements.
yes it's more redundant but the super nested conditions are difficult to parse and probably need individual comments with reasoning

[maribethb]

worth adding a if typoeof node.getSourceBlock === function return node.getSourceBlock() to sort of make this work as if we had a IBelongsToBlock interface or something? Or worth formalizing that as an interface? potentially as a followup.

[maribethb]

I think we probably need to remove this constant? It's not used anywhere that I can tell? We shouldn't need both it and the keyboardNavigationController.