Actions API revision

Summary

In Flutter an Intent is an object that’s typically bound to a keyboard key combination using the Shortcuts widget. An Intent can be bound to an Action, which can update the application’s state or perform other operations. In the course of using this API, we identified several drawbacks in the design, so we have updated the Actions API to make it easier to use and understand.

In the previous Actions API design, actions were mapped from a LocalKey to an ActionFactory that created a new Action each time the invoke method was called. In the current API, actions are mapped from the type of the Intent to an Action instance (with a Map<Type, Action>), and they are not created anew for each invocation.

Context

The original Actions API design was oriented towards invoking actions from widgets, and having those actions act in the context of the widget. Teams have been using actions, and found several limitations in that design that needed to be addressed:

  1. Actions couldn’t be invoked from outside of the widget hierarchy. Examples of this include processing a script of commands, some undo architectures, and some controller architectures.

  2. The mapping from shortcut key to Intent and then to Action wasn’t always clear, since the data structures mapped LogicalKeySet =>Intent and then LocalKey => ActionFactory. The new mapping is still LogicalKeySet to Intent but then it maps Type (Intent type) to Action, which is more direct and readable, since the type of the intent is written in the mapping.

  3. If the key binding for an action was in another part of the widget hierarchy, it was not always possible for the Intent to have access to the state necessary to decide if the intent/action should be enabled or not.

To address these issues, we made some significant changes to the API. The mapping of actions was made more intuitive, and the enabled interface was moved to the Action class. Some unnecessary arguments were removed from the Action’s invoke method and its constructor, and actions were allowed to return results from their invoke method. Actions were made into generics, accepting the type of Intent they handle, and LocalKeys were no longer used for identifying which action to run, and the type of the Intent is used instead.

The majority of these changes were made in the PRs for Revise Action API and Make Action.enabled be isEnabled(Intent intent) instead, and are described in detail in the design doc.

Description of change

Here are the changes made to address the above problems:

  1. The Map<LocalKey, ActionFactory> that was given to the Actions widget is now a Map<Type, Action<Intent>> (the type is the type of the Intent to be passed to the Action).
  2. The isEnabled method was moved from the Intent class to the Action class.
  3. The FocusNode argument to Action.invoke and Actions.invoke methods was removed.
  4. Invoking an action no longer creates a new instance of the Action.
  5. The LocalKey argument to the Intent constructor was removed.
  6. The LocalKey argument to CallbackAction was removed.
  7. The Action class is now a generic (Action<T extends Intent>) for better type safety.
  8. The OnInvokeCallback used by CallbackAction no longer takes a FocusNode argument.
  9. The ActionDispatcher.invokeAction signature has changed to not accept an optional FocusNode, but instead take an optional BuildContext.
  10. The LocalKey static constants (named key by convention) in Action subclasses have been removed.
  11. The Action.invoke and ActionDispatcher.invokeAction methods now return the result of invoking the action as an Object.
  12. The Action class may now be listened to for state changes.
  13. The ActionFactory typedef has been removed, as it is no longer used.

Example analyzer failures

Here are some example analyzer failures that might be encountered where an outdated use of the Actions API might be the cause of the problem. The specifics of the error might differ, and there may be other failures caused by these changes.

error: MyActionDispatcher.invokeAction' ('bool Function(Action<Intent>, Intent, {FocusNode focusNode})') isn't a valid override of 'ActionDispatcher.invokeAction' ('Object Function(Action<Intent>, Intent, [BuildContext])'). (invalid_override at [main] lib/main.dart:74)

error: MyAction.invoke' ('void Function(FocusNode, Intent)') isn't a valid override of 'Action.invoke' ('Object Function(Intent)'). (invalid_override at [main] lib/main.dart:231)

error: The method 'isEnabled' isn't defined for the type 'Intent'. (undefined_method at [main] lib/main.dart:97)

error: The argument type 'Null Function(FocusNode, Intent)' can't be assigned to the parameter type 'Object Function(Intent)'. (argument_type_not_assignable at [main] lib/main.dart:176)

error: The getter 'key' isn't defined for the type 'NextFocusAction'. (undefined_getter at [main] lib/main.dart:294)

error: The argument type 'Map<LocalKey, dynamic>' can't be assigned to the parameter type 'Map<Type, Action<Intent>>'. (argument_type_not_assignable at [main] lib/main.dart:418)

Migration guide

Significant changes area required to update existing code to the new API.

Actions mapping for pre-defined actions

To update the action maps in the Actions widget for predefined actions in Flutter, like ActivateAction and SelectAction, do the following:

  • Update the argument type of the actions argument
  • Use an instance of a specific Intent class in the Shortcuts mapping, rather than an Intent(TheAction.key) instance.

Code before migration:

class MyWidget extends StatelessWidget {
  // ...
  @override
  Widget build(BuildContext context) {
    return Shortcuts(
      shortcuts: <LogicalKeySet, Intent> {
        LogicalKeySet(LogicalKeyboardKey.enter): Intent(ActivateAction.key),
      },
      child: Actions(
        actions: <LocalKey, ActionFactory>{
          Activate.key: () => ActivateAction(),
        },
        child: Container(),
      )
    );
  }
}

Code after migration:

class MyWidget extends StatelessWidget {
  // ...
  @override
  Widget build(BuildContext context) {
    return Shortcuts(
      shortcuts: <LogicalKeySet, Intent> {
        LogicalKeySet(LogicalKeyboardKey.enter): ActivateIntent,
      },
      child: Actions(
        actions: <Type, Action<Intent>>{
          ActivateIntent: ActivateAction(),
        },
        child: Container(),
      )
    );
  }
}

Custom actions

To migrate your custom actions, eliminate the LocalKeys you’ve defined, and replace them with Intent subclasses, as well as changing the type of the argument to the actions argument of the Actions widget.

Code before migration:

class MyAction extends Action {
  MyAction() : super(key);

  /// The [LocalKey] that uniquely identifies this action to an [Intent].
  static const LocalKey key = ValueKey<Type>(RequestFocusAction);

  @override
  void invoke(FocusNode node, MyIntent intent) {
    // ...
  }
}

class MyWidget extends StatelessWidget {
  // ...
  @override
  Widget build(BuildContext context) {
    return Shortcuts(
      shortcuts: <LogicalKeySet, Intent> {
        LogicalKeySet(LogicalKeyboardKey.enter): Intent(MyAction.key),
      },
      child: Actions(
        actions: <LocalKey, ActionFactory>{
          MyAction.key: () => MyAction(),
        },
        child: Container(),
      )
    );
  }
}

Code after migration:

// You may need to create new Intent subclasses if you used
// a bare LocalKey before.
class MyIntent extends Intent {
  const MyIntent();
}

class MyAction extends Action<MyIntent> {
  @override
  Object invoke(MyIntent intent) {
    // ...
  }
}

class MyWidget extends StatelessWidget {
  // ...
  @override
  Widget build(BuildContext context) {
    return Shortcuts(
      shortcuts: <LogicalKeySet, Intent> {
        LogicalKeySet(LogicalKeyboardKey.enter): MyIntent,
      },
      child: Actions(
        actions: <Type, Action<Intent>>{
          MyIntent: MyAction(),
        },
        child: Container(),
      )
    );
  }
}

Custom Actions and Intents with arguments

To update actions that use intent arguments or hold state, you need to modify the arguments to the invoke method. In the example below, the code keeps the value of the argument in the intent as part of the action instance. This is because in the old design there is a new instance of the action created each time it’s executed, and the resulting action could be kept by the ActionDispatcher to record the state.

In the example of post migration code below, the new MyAction returns the state as the result of calling invoke, since a new instance isn’t created for each invocation. This state is returned to the caller of Actions.invoke, or ActionDispatcher.invokeAction, depending on how the action is invoked.

Code before migration:

class MyIntent extends Intent {
  const MyIntent({this.argument});

  final int argument;
}

class MyAction extends Action {
  MyAction() : super(key);

  /// The [LocalKey] that uniquely identifies this action to an [Intent].
  static const LocalKey key = ValueKey<Type>(RequestFocusAction);

  int state;

  @override
  void invoke(FocusNode node, MyIntent intent) {
    // ...
    state = intent.argument;
  }
}

Code after migration:

class MyIntent extends Intent {
  const MyIntent({this.argument});

  final int argument;
}

class MyAction extends Action<MyIntent> {
  @override
  int invoke(Intent intent) {
    // ...
    return intent.argument;
  }
}

Timeline

Landed in version: 1.18
In stable release: 1.20

References

API documentation:

Relevant issue:

Relevant PRs: