async_redux 2.0.6

  • Readme
  • Changelog
  • Example
  • Installing
  • 91

pub package pub package

async_redux #

Async Redux is a special version of Redux which:

  1. Is easy to learn
  2. Is easy to use
  3. Is easy to test
  4. Has no boilerplate

The below documentation is very detailed. For an overview, go to the Medium story.

Table of Contents #

What is Redux? #

A single store holds all the state, which is immutable. When you need to modify some state you dispatch an action. Then a reducer creates a new copy of the state, with the desired changes. Your widgets are connected to the store (through store-connectors and view-models), so they know that the state changed, and rebuild as needed.

Why use this Redux version over others? #

Plain vanilla Redux is too low-level, which makes it very flexible but results in a lot of boilerplate, and a steep learning curve.

Combining reducers is a manual task, and you have to list them one by one. If you forget to list some reducer, you will not know it until your tests point out that some state is not changing as you expected.

Reducers can't be async, so you need to create middleware, which is also difficult to setup and use. You have to list them one by one, and if you forget one of them you will also not know it until your tests point it out. The redux_thunk package can help with that, but adds some more complexity.

It's difficult to know which actions fire which reducers, and hard to navigate the code in the IDE. In IntelliJ you may press CTRL+B to navigate between a method use and its declaration. However, this is of no use if actions and reducers are independent classes. You have to search for action "usages", which is not so convenient since it also list dispatches.

It's also difficult to list all actions and reducers, and you may end up implementing some reducer just to realize it already exists with another name.

Testing reducers is simple, since they are pure functions, but integration tests are difficult. In the real world you need to test complex middleware that fires other middleware and many reducers, with intermediate state changes that you want to test for. Especially if you are doing BDD or Acceptance Tests you may need to wait for some middleware to finish, and then dispatch some other actions, and test for intermediate states.

Another problem is that vanilla Redux assumes it holds all of the application state, and this is not practical in a real Flutter app. If you add a simple TextField with a TextEditingController, or a ListView with a ScrollController, then you have state outside of the Redux store. Suppose your middleware is downloading some information, and it wishes to scroll a ListView as soon as the info arrives. This would be simple if the list scroll position is saved in the Redux store. However, this state must be in the ScrollController, not the store.

AsyncRedux solves all of these problems and more:

  • It's much easier to learn and use than regular Redux.
  • It comes with its own testing tools that make even complex tests easy to setup and run.
  • You can navigate between action dispatches and their corresponding reducers with a single IDE command or click.
  • You can also use your IDE to list all actions/reducers.
  • You don't need to add or list reducers and middleware anywhere.
  • In fact, reducers can be async, so you don't need middleware.
  • There is no need for generated code (as some Redux versions do).
  • It has the concept of "events", to deal with Flutter state controllers.
  • It helps you show errors thrown by reducers to the user.
  • It's easy to add both logging and store persistence.

Store and State #

Declare your store and state, like this:

var state = AppState.initialState();

var store = Store<AppState>(
  initialState: state,
);

Actions #

If you want to change the store state you must "dispatch" some action. In AsyncRedux all actions extend ReduxAction.

The reducer of an action is simply a method of the action itself, called reduce(). All actions must override this method.

The reducer has direct access to:

  • The store state (which is a getter of the Action class).
  • The action state itself (the class fields, passed to the action when it was instantiated and dispatched).
  • The dispatch method, so that other actions may be dispatched from the reducer.

Sync Reducer #

If you want to do some synchronous work, simply declare the reducer to return AppState, then change the state and return it.

For example, let's start with a simple action to increment a counter by some value:

class IncrementAction extends ReduxAction<AppState> {

  final int amount;

  IncrementAction({this.amount}) : assert(amount != null);

  @override
  AppState reduce() {
	return state.copy(counter: state.counter + amount));
  }
}

This action is dispatched like this:

store.dispatch(IncrementAction(amount: 3));

Note the reducer above has direct access to both the counter state (state.counter) and to the action state (the field amount).

We will show you later how to easily test sync reducers, using the StoreTester.

Try running the: Increment Example.

Async Reducer #

If you want to do some asynchronous work, simply declare the reducer to return Future<AppState> then change the state and return it. There is no need of any "middleware", like for other Redux versions.

Note: In IntelliJ, to convert the reducer from sync to async, press Alt+ENTER and select Convert to async function body.

As an example, suppose you want to increment a counter by a value you get from the database. The database access is async, so you must use an async reducer:

class QueryAndIncrementAction extends ReduxAction<AppState> {

  @override
  Future<AppState> reduce() async {
	int value = await getAmount();
	return state.copy(counter: state.counter + value));
  }
}

This action is dispatched like this:

store.dispatch(QueryAndIncrementAction());

We will show you later how to easily test async reducers, using the StoreTester.

Try running the: Increment Async Example.

One important rule #

When your reducer returns Future<AppState> you must make sure you do not return a completed future. In other words, all execution paths of the reducer must pass through at least one await keyword.

If your reducer has no awaits, you must return AppState instead of Future<AppState>, or simply add something like await Future.sync(() {});.

Example:

// These are right:
AppState reduce() { return state; }
AppState reduce() { someFunc(); return state; }
Future<AppState> reduce() async { await someFuture(); return state; }
Future<AppState> reduce() async { await Future.value(null); return state; }

// But these are wrong:
Future<AppState> reduce() async { return state; }
Future<AppState> reduce() async { someFunc(); return state; }

Dart doesn't let AsyncRedux detect if a future is completed or not (the information is there, but in a private field), so we can't help you with that. One day we'll have an IDE plugin to warn you if you make this mistake, but until then, please pay attention.

If you don't follow this rule, AsyncRedux may seem to work ok, but will eventually misbehave. If you're an advanced user interested in the details, check the sync/async tests.

Changing state is optional #

For both sync and async reducers, returning a new state is optional. If you don't plan on changing the state, simply return null. This is the same as returning the state unchanged.

Why is this useful? Because some actions may simply start other async processes, or dispatch other actions.

For example, suppose you want to have two separate actions, one for querying some value from the database, and another action to change the state:

class QueryAction extends ReduxAction<AppState> {

  @override
  Future<AppState> reduce() async {
    int value = await getAmount();
    dispatch(IncrementAction(amount: value));
    return null;
  }
}

class IncrementAction extends ReduxAction<AppState> {
  
  final int amount;

  IncrementAction({this.amount}) : assert(amount != null);

  @override
  AppState reduce() {
    return state.copy(counter: state.counter + amount));
  }
}

Note the reduce() methods have direct access to state and dispatch. There is no need to write store.state and store.dispatch (although you can, if you want).

Before and After the Reducer #

Sometimes, while an async reducer is running, you want to prevent the user from touching the screen. Also, sometimes you want to check preconditions like the presence of an internet connection, and don't run the reducer if those preconditions are not met.

To help you with these use cases, you may override methods ReduxAction.before() and ReduxAction.after(), which run respectively before and after the reducer.

The before() method runs before the reducer. If you want it to run synchronously, it should return void:

void before() { ... }

To run it asynchronously, return Future<void>:

Future<void> before() async { ... }

If it throws an error, then reduce() will NOT run. This means you can use it to check any preconditions and throw an error if you want to prevent the reducer from running. For example:

Future<void> before() async => await checkInternetConnection();

This method is also capable of dispatching actions, so it can be used to turn on a modal barrier:

void before() => dispatch(WaitAction(true));

Note: If this method runs asynchronously, then reduce() will also be async, since it must wait for this one to finish.

The after() method runs after reduce(), even if an error was thrown by before() or reduce() (akin to a "finally" block). If the after() method itself throws an error, then this error will be "swallowed" and ignored. Avoid after() methods which can throw errors.

This method can also dispatch actions, so it can be used to turn off some modal barrier when the reducer ends, even if there was some error in the process:

void after() => dispatch(WaitAction(false));

Complete example:

// This action increments a counter by 1, and then gets some description text.
class IncrementAndGetDescriptionAction extends ReduxAction<AppState> {

  @override
  Future<AppState> reduce() async {
	dispatch(IncrementAction());
	String description = await read("http://numbersapi.com/${state.counter}");
	return state.copy(description: description);
  }

  void before() => dispatch(WaitAction(true));

  void after() => dispatch(WaitAction(false));
}

Try running the: Before and After Example.

What's the order of execution of sync and async reducers? #

A reducer is only sync if both reducer() return AppState AND before() return void. If you any of them return a Future, then the reducer is async.

When you dispatch sync reducers, they are executed synchronously, in the order they are called. For example, this code will wait until MyAction1 is finished and only then run MyAction2, assuming both are sync reducers:

dispatch(MyAction1()); 
dispatch(MyAction2());

When you dispatch an async reducer with dispatch it starts immediately, but there is really no way of knowing when it finishes, because it depends on how long the async code takes to run, which depends on the async code itself, not AsyncRedux.

Dispatching an async reducer with dispatch is like starting any other async processes without writing await. The process will start, but you are not waiting for it to finish. For example, this code will start both MyAsyncAction1 and MyAsyncAction2, but is says nothing about how long they will take or which one finishes first:

dispatch(MyAsyncAction1()); 
dispatch(MyAsyncAction2());

If you want to wait for some async action to finish, you must dispatch it with dispatchFuture instead of dispatch. Then you can actually wait for it to finish. For example, this code will wait until MyAsyncAction1 is finished, and only then run MyAsyncAction2:

await dispatchFuture(MyAsyncAction1()); 
await dispatchFuture(MyAsyncAction2());

Connector #

As usual, in Redux you generally have two widgets, one called the "dumb-widget", which knows nothing about Redux and the store, and another one to "wire" the store with that dumb-widget. Vanilla Redux calls these wiring widgets "containers", but we consider this bad since Flutter's most common widget is already called a Container. So we call them "connectors", and they do their magic by using a StoreConnector and a ViewModel.

For example:

class MyHomePageConnector extends StatelessWidget {
  @override
  Widget build(BuildContext context) {
	return StoreConnector<AppState, ViewModel>(
	  model: ViewModel(),
	  builder: (BuildContext context, ViewModel vm) => MyHomePage(
		counter: vm.counter,
		description: vm.description,
		onIncrement: vm.onIncrement,
	  ));
  }
}

// Helper class to the connector widget. Holds the part of the State the widget needs,
// and may perform conversions to the type of data the widget can conveniently work with.
class ViewModel extends BaseModel<AppState> {
  ViewModel();

  int counter;
  String description;
  VoidCallback onIncrement;

  ViewModel.build({
	@required this.counter,
	@required this.description,
	@required this.onIncrement,
  }) : super(equals: [counter, description]);

  @override
  ViewModel fromStore() => ViewModel.build(
		counter: state.counter,
		description: state.description,
		onIncrement: () => dispatch(IncrementAndGetDescriptionAction()),
	  );
}

The StoreConnector has a distinct parameter. As a performance optimization, distinct:true allows the widget to be rebuilt only when the ViewModel changes. If this is not done, then the widget will be rebuilt every time any state in the store is changed.

This distinct parameter is true by default, but this can be changed when creating the store, by passing it defaultDistinct:false.

If distinct is true, you must implement equals and hashcode for the ViewModel, otherwise there is no way to know if the ViewModel changed.

This can be done in three ways:

  • By typing ALT+INSERT in IntelliJ IDEA and choosing ==() and hashcode. You can't forget to update this whenever new parameters are added to the model.

  • You can use the built_value package to ensure they are kept correct, without you having to update them manually.

  • Just add all the fields you want to the equals parameter to the ViewModel's build constructor. This will allow the ViewModel to automatically create its own operator == and hashcode implicitly. For example:

ViewModel.build({
	  @required this.field1,
	  @required this.field2,
}) : super(equals: [field1, field2]);

How to provide the ViewModel to the StoreConnector #

The StoreConnector actually accepts two parameters for the ViewModel, of which one but only one should be provided in the StoreConnector constructor: model or converter.

  1. the model parameter

    It expects a ViewModel that extends BaseModel. This allows your model class to use the equals parameter, as already explained above, so that you don't need to implement operator == and hashcode by hand.

    Also, AsyncRedux will automatically inject state and dispatch into your model instance, so that boilerplate is reduced in your fromStore method. For example:

class ViewModel extends BaseModel<Store<AppState>> {
   ViewModel();
 
   String name;
   VoidCallback onSave;

   ViewModel.build({
     @required this.name,
     @required this.onSave,
   }) : super(equals: [name]);

   @override
   ViewModel fromStore() => ViewModel.build(
       name: state.user.name,
       onSave: () => dispatch(SaveUserAction()),
   );
}

With this architecture you may also create separate methods for helping construct your model, without having to pass the store around. For example:

@override
ViewModel fromStore() => ViewModel.build(
    name: _name(),
    onSave: _onSave,
);

String _name() => state.user.name;

VoidCallback _onSave: () => dispatch(SaveUserAction()),

Another idea is to subclass BaseModel to provide additional features to your model. For example, you could add extra getters to help you access state:

User user => state.user;

@override
ViewModel fromStore() => ViewModel.build(
   name: user.name,
   ...
);

Most examples in the example tab use the model parameter.

  1. The converter parameter

    This is the good old one from flutter_redux. It expects a static factory function that gets a store and returns the ViewModel. You probably should use this one if you are migrating from flutter_redux.

class ViewModel {
   String name;
   VoidCallback onSave;

   ViewModel({
      @required this.name,
      @required this.onSave,
   });

   static ViewModel fromStore(Store<AppState> store) {
      return ViewModel(
         name: store.state,
         onSave: () => store.dispatch(IncrementAction(amount: 1)),
      );
   }

   @override
   bool operator ==(Object other) =>
       identical(this, other) ||
       other is ViewModel && runtimeType == other.runtimeType && name == other.name;

   @override
   int get hashCode => name.hashCode;
}

With this architecture it's a bit more difficult to create separate methods for helping construct your model:

static ViewModel fromStore(Store<AppState> store) {
   return ViewModel(
      name: _name(store),
      onSave: _onSave(store),
   );
}

static String _name(Store<AppState>) => store.state.user.name;

static VoidCallback _onSave(Store<AppState>) { 
   return () => store.dispatch(SaveUserAction());
} 

To see the converter parameter in action, please run this example.

Alternatives to the Connector #

The StoreConnector forces you to cleanly separate the widgets from the way they get their data. This is better for clean code and will help a lot when you are writing tests.

However, if you want and you know what you are doing, here is how to access the store directly from inside of your widgets (for example in the build method):

/// Dispatch an action without a StoreConnector.
StoreProvider.dispatch<AppState>(context, MyAction());

/// Dispatch an action without a StoreConnector,
/// and get a `Future<void>` which completes when the action is done.
StoreProvider.dispatchFuture<AppState>(context, MyAction());

/// Get the state, without a StoreConnector.
AppState state = StoreProvider.state<AppState>(context); 

Provider #

Another good alternative to the StoreConnector is using the Provider package.

Both the StoreConnector (from async_redux) and ReduxSelector (from provider_for_redux) let you deal with widget rebuilds when the state changes.

You may use StoreConnector when you want to have two widgets, one to access the store and prepare the state to use, and the second as a dumb widget. You may use ReduxSelector when you want less boilerplate, and want to access the store directly from inside a single widget.

Please visit the provider_for_redux package for in-depth explanation and examples on how to use AsyncRedux and Provider together.

Processing errors thrown by Actions #

AsyncRedux has special provisions for dealing with errors, including observing errors, showing errors to users, and wrapping errors into more meaningful descriptions.

Let's see an example. Suppose a logout action that checks if there is an internet connection, and then deletes the database and sets the store to its initial state:

class LogoutAction extends ReduxAction<AppState> {
  @override
  Future<AppState> reduce() async {
	await checkInternetConnection();
	await deleteDatabase();
	dispatch(NavigateToLoginScreenAction();
	return AppState.initialState();
  }
}

In the above code, the checkInternetConnection() function checks if there is an internet connection, and if there isn't it throws an error:

Future<void> checkInternetConnection() async {
	if (await Connectivity().checkConnectivity() == ConnectivityResult.none)
		throw NoInternetConnectionException();
}

All errors thrown by action reducers are sent to the ErrorObserver, which you may define during store creation. For example:

var store = Store<AppState>(
  initialState: AppState.initialState(),
  errorObserver: errorObserver,
);

bool errorObserver(Object error, ReduxAction action, Store store, Object state, int dispatchCount) {
  print("Error thrown during $action: $error);
  return true;
}

If your error observer returns true, the error will be rethrown after the errorObserver finishes. If it returns false, the error is considered dealt with, and will be "swallowed" (not rethrown).

Giving better error messages #

If your reducer throws some error you probably want to collect as much information as possible. In the above code, if checkInternetConnection() throws an error, you want to know that you have a connection problem, but you also want to know this happened during the logout action. In fact, you want all errors thrown by this action to reflect that.

The solution is implementing the optional wrapError(error) method:

class LogoutAction extends ReduxAction<AppState> {

  @override
  Future<AppState> reduce() async { ... }

  @override
  Object wrapError(error)
	  => LogoutError("Logout failed.", cause: error);
}

Note the LogoutError above gets the original error as cause, so no information is lost.

In other words, the wrapError(error) method acts as the "catch" statement of the action.

User exceptions #

To show error messages to the user, make your actions throw an UserException, and then wrap your home-page with UserExceptionDialog, below StoreProvider and MaterialApp:

class MyApp extends StatelessWidget {
  @override
  Widget build(BuildContext context)
	  => StoreProvider<AppState>(
		  store: store,
		  child: MaterialApp(
			home: UserExceptionDialog<AppState>(
			  child: MyHomePage(),
			)));
}

Try running the: Show Error Dialog Example.

In more detail:

Sometimes, actions fail because the user provided invalid information. These failings don't represent errors in the code, so you usually don't want to log them as errors. What you want, instead, is just warn the user by opening a dialog with some corrective information. For example, suppose you want to save the user's name, and you only accept names with at least 4 characters:

class SaveUserAction extends ReduxAction<AppState> {
   final String name;
   SaveUserAction(this.name);

   @override
   Future<AppState> reduce() async {
	 if (name.length < 4) dispatch(ShowDialogAction("Name must have at least 4 letters."));
	 else await saveUser(name);
	 return null;
   }
}

Clearly, there is no need to log as an error the user's attempt to save a 3-char name. The above code dispatches a ShowDialogAction, which you would have to wire into a Flutter error dialog somehow.

However, there's an easier approach. Just throw AsyncRedux's built-in UserException:

class SaveUserAction extends ReduxAction<AppState> {
   final String name;
   SaveUserAction(this.name);

   @override
   Future<AppState> reduce() async {
	 if (name.length < 4) throw UserException("Name must have at least 4 letters.");
	 await saveName(name);
	 return null;
   }
}

The special UserException error class represents "user errors" which are meant as warnings to the user, and not as code errors to be logged. By default (if you don't define your own errorObserver) only errors which are not UserException are thrown. And if you do define an errorObserver, you'd probably want to replicate this behavior.

In any case, UserExceptions are put into a special error queue, from where they may be shown to the user, one by one. You may use UserException as is, or subclass it, returning title and message for the alert dialog shown to the user.

As explained in the beginning of this section, if you use the build-in error handling you must wrap your home-page with UserExceptionDialog. There, you may pass the onShowUserExceptionDialog parameter to change the default dialog, show a toast, or some other suitable widget:

UserExceptionDialog<AppState>(
	  child: MyHomePage(),
	  onShowUserExceptionDialog:
		  (BuildContext context, UserException userException) => showDialog(...),
);

UserExceptionAction #

If you want the UserExceptionDialog to display some UserException, you must throw the exception from inside an action's before() or reduce() methods.

However, sometimes you need to create some callback that throws an UserException. If this callback is called outside of an action, the dialog will not display the exception. To solve this, the callback should not throw an exception, but instead call the provided UserExceptionAction, which will then simply throw the exception in its own reduce() method.

The UserExceptionAction is also useful even inside of actions, when you want to display an error dialog to the user but you don't want to interrupt the action by throwing an exception.

Testing #

It's often said that vanilla Redux reducers are easy to test because they're pure functions. While this is true, real-world applications are composed not only of sync reducers, but also of middleware async code, which is not easy to test at all.

AsyncRedux provides the StoreTester class that makes it easy to test both sync and async reducers.

Start by creating the store-tester from a store:

var store = Store<AppState>(initialState: AppState.initialState());
var storeTester = StoreTester.from(store);

Or else, creating it directly from AppState:

var storeTester = StoreTester<AppState>(initialState: AppState.initialState());

Then, dispatch some action, wait for it to finish, and check the resulting state:

storeTester.dispatch(SaveNameAction("Mark"));
TestInfo<AppState> info = await storeTester.wait(SaveNameAction);
expect(info.state.name, "Mark");

The variable info above will contain information about after the action reducer finishes executing, no matter if the reducer is sync or async.

The TestInfo instance contains the following:

  • state: The store state.
  • action: The dispatched Action that resulted in that state.
  • ini: A boolean which indicates true if this info represents the "initial" state right before the action is dispatched, or false it represents the "end" state right after the action finishes executing.
  • dispatchCount: The number of dispatched actions so far.
  • reduceCount: The number of reduced states so far.
  • errors: The UserExceptions the store was holding when the information was gathered.

While the above example demonstrates the testing of a simple action, real-world apps have actions that dispatch other actions. You may use different StoreTester methods to check if the expected actions are dispatched, and test their intermediary states.

Let's see all the available methods of the StoreTester:

  1. Future<TestInfo> wait(Type actionType)

    Expects one action of the given type to be dispatched, and waits until it finishes. Returns the info after the action finishes. Will fail with an exception if an unexpected action is seen.

  2. Future<TestInfo> waitUntil(Type actionType)

    Runs until an action of the given type is dispatched, and then waits until it finishes. Returns the info after the action finishes. Ignores other actions types.

  3. Future<TestInfo> waitUntilAction(ReduxAction action)

    Runs until the exact given action is dispatched, and then waits until it finishes. Returns the info after the action finishes. Ignores other actions.

  4. Future<TestInfo> waitAllGetLast(List<Type> actionTypes, {List<Type> ignore})

    Runs until all given actions types are dispatched, in order. Waits until all of them are finished. Returns the info after all actions finish. Will fail with an exception if an unexpected action is seen, or if any of the expected actions are dispatched in the wrong order. To ignore some actions, pass them to the ignore list.

  5. Future<TestInfo> waitAllUnorderedGetLast(List<Type> actionTypes, {List<Type> ignore})

    Runs until all given actions types are dispatched, in any order. Waits until all of them are finished. Returns the info after all actions finish. Will fail with an exception if an unexpected action is seen. To ignore some actions, pass them to the ignore list.

  6. Future<TestInfoList> waitAll(List<Type> actionTypes, {List<Type> ignore})

    The same as waitAllGetLast, but instead of returning just the last info, it returns a list with the end info for each action. To ignore some actions, pass them to the ignore list.

  7. Future<TestInfoList> waitAllUnordered(List<Type> actionTypes, {List<Type> ignore})

    The same as waitAllUnorderedGetLast, but instead of returning just the last info, it returns a list with the end info for each action. To ignore some actions, pass them to the ignore list.

  8. Future<TestInfoList<St>> waitCondition(StateCondition<St> condition, {bool ignoreIni = true})

    Runs until the predicate function condition returns true. This function will receive each testInfo, from where it can access the state, action, errors etc. Only END states will be received, unless you pass ignoreIni as false. Returns a list with all info until the condition is met.

  9. Future<TestInfo<St>> waitConditionGetLast(StateCondition<St> condition, {bool ignoreIni = true})

    Runs until the predicate function condition returns true. This function will receive each testInfo, from where it can access the state, action, errors etc. Only END states will be received, unless you pass ignoreIni as false. Returns the info after the condition is met.

  10. Future<TestInfo> waitUntilError({Object error, Object processedError})

    Runs until after an action throws an error of this exact type, or this exact error (using equals). You can also, instead, define processedError, which is the error after wrapped by the action's wrapError() method. Returns a list with all info until the error condition is met.

  11. Future<TestInfo> waitUntilErrorGetLast({Object error, Object processedError})

    Runs until after an action throws an error of this exact type, or this exact error (using equals). You can also, instead, define processedError, which is the error after wrapped by the action's wrapError() method. Returns the info after the condition is met.

Some of the methods above return a list of type TestInfoList, which contains the step by step information of all the actions. You can then query for the actions you want to inspect. For example, suppose an action named IncrementAndGetDescriptionAction calls another 3 actions. You can assert that all actions are called in order, and then get the state after each one of them have finished, all at once:

var storeTester = StoreTester<AppState>(initialState: AppState.initialState());
expect(storeTester.state.counter, 0);
expect(storeTester.state.description, isEmpty);

storeTester.dispatch(IncrementAndGetDescriptionAction());

TestInfoList<AppState> infos = await storeTester.waitAll([
  IncrementAndGetDescriptionAction,
  WaitAction,
  IncrementAction,
  WaitAction,
]);

// Modal barrier is turned on (first time WaitAction is dispatched).
expect(infos.get(WaitAction, 1).state.waiting, true);

// While the counter was incremented the barrier was on.
expect(infos[IncrementAction].waiting, true);

// Then the modal barrier is dismissed (second time WaitAction is dispatched).
expect(infos.get(WaitAction, 2).state.waiting, false);

// In the end, counter is incremented, description is created, and barrier is dismissed.
var info = infos[IncrementAndGetDescriptionAction];
expect(info.state.waiting, false);
expect(info.state.description, isNotEmpty);
expect(info.state.counter, 1);

Try running the: Testing with the Store Listener.

Also, the tests of the StoreTester can also serve as examples.

Testing UserExceptions #

Since UserExceptions don't represent bugs in the code, AsyncRedux put them into the store's errors queue, and then swallows them. This is usually what you want during production, where errors from this queue are shown in a dialog to the user. But it may or may not be what you want during tests.

In tests there are two possibilities:

  1. You are testing that some UserException is thrown. For example, you want to test that users are warned if they typed letters in some field that only accepts numbers. To that end, your test would dispatch the appropriate action, and then check if the errors queue now contains an UserException with some specific error message.

  2. You are testing some code that should not throw any exceptions. If the test has thrown an exception it means the test has failed, and the exception should show up in the console, for debugging. However, this won't work if when test throws an UserException it simply go to the errors queue. If this happens, the test will continue running, and may even pass. The only way to make sure no errors were thrown would be asserting that the errors queue is still empty at the end of the test. This is even more problematic if the unexpected UserException is thrown inside of a before() method. In this case it will prevent the reducer to run, and the test will probably fail with wrong state but no errors in the console.

The solution is to use the shouldThrowUserExceptions parameter in the StoreTester constructor.

Pass shouldThrowUserExceptions as true, and all errors will be thrown and not swallowed, including UserExceptions. Use this in all tests that should throw no errors:

var storeTester = StoreTester<AppState>(
                     initialState: AppState.initialState(), 
                     shouldThrowUserExceptions: true);

Pass shouldThrowUserExceptions as false (the default) when you are testing code that should indeed throw UserExceptions. These exceptions will then silently go to the errors queue, where you can assert they exist and have the right error messages:

storeTester.dispatch(MyAction());
TestInfo info = await storeTester.waitAllGetLast([MyAction]);
expect(info.errors.removeFirst().msg, "You can't do this.");

Test files #

If you want your tests to be comprehensive you should probably have 3 different types of test for each widget:

  1. State Tests — Test the state of the app, including actions/reducers. This type of tests make use of the StoreTester described above.

  2. Connector Tests — Test the connection between the store and the "dumb-widget". In other words it tests the "connector-widget" and the "view-model".

  3. Presentation Tests — Test the UI. In other words it tests the "dumb-widget", making sure that the widget displays correctly depending on the parameters you use in its constructor. You pass in the data the widget requires in each test for rendering, and then writes assertions against the rendered output. Think of these tests as "pure function tests" of our UI. It also tests that the callbacks are called when necessary.

For example, suppose you have the counter app shown here. Then:

  • The state test could create a store with count 0 and description empty, and then dispatch IncrementAction and expect the count to become 1. Then it could test dispatching IncrementAndGetDescriptionAction alters the count to 2 and the description to some non-empty string.

  • The connector test would create a store and a page with the MyHomePageConnector widget. It would then access the MyHomePage and make sure it gets the expected info from the store, and also that the expected IncrementAndGetDescriptionAction is dispatched when the "+" button is tapped.

  • The presentation test would create the MyHomePage widget, pass counter:0 and description:"abc" parameters in its constructor, and make sure they appear in the screen as expected. It would also test that the callback is called when the "+" button is tapped.

Since each widget will have a bunch of related files, you should have some consistent naming convention. For example, if some dumb-widget is called MyWidget, its file could be my_widget.dart. Then the corresponding connector-widget could be MyWidgetConnector in my_widget_CONNECTOR.dart. The three corresponding test files could be named my_widget_STATE_test.dart, my_widget_CONNECTOR_test.dart and my_widget_PRESENTATION_test.dart. If you don't like this convention use your own, but just choose one early and stick to it.

Route Navigation #

AsyncRedux comes with a NavigateAction which you can dispatch to navigate your Flutter app. For this to work, during app initialization you must create a navigator key and then inject it into the action:

final navigatorKey = GlobalKey<NavigatorState>();

void main() async {
  NavigateAction.setNavigatorKey(navigatorKey);
  ...
}

You must also use this same navigator key in your MaterialApp:

return StoreProvider<AppState>(
  store: store,
  child: MaterialApp(
	  ...
	  navigatorKey: navigatorKey,
	  initialRoute: '/',
	  onGenerateRoute: ...
	  ),
);

Then, use the action as needed:

dispatch(NavigateAction.pop());
dispatch(NavigateAction.pushNamed("myRoute"));
dispatch(NavigateAction.pushReplacementNamed("myRoute"));
dispatch(NavigateAction.pushNamedAndRemoveAll("myRoute"));
dispatch(NavigateAction.popUntil("myRoute"));

Note: Don't ever save the current route in the store. This will create all sorts of problems. If you need to know the route you're in, you may use this static method provided by NavigateAction:

String routeName = NavigateAction.getCurrentNavigatorRouteName(context);

Try running the: Navigate Example.

Events #

In a real Flutter app it's not practical to assume that a Redux store can hold all of the application state. Widgets like TextField and ListView make use of controllers, which hold state, and the store must be able to work alongside these. For example, in response to the dispatching of some action you may want to clear the text-field, or you may want to scroll the list-view to the top. Even when no controllers are involved, you may want to execute some one-off processes, like opening a dialog or closing the keyboard, and it's not obvious how to do that in vanilla Redux.

AsyncRedux solves these problems by introducing the concept of "events". The naming convention is that Events are named with the Evt suffix.

Boolean events can be created like this:

var clearTextEvt = Event();

But you can have events with payloads of any other data type. For example:

var changeTextEvt = Event<String>("Hello");
var myEvt = Event<int>(42);

Events may be put into the store state in their "spent" state, by calling its spent() constructor. For example, while creating the store initial-state:

static AppState initialState() {
  return AppState(
	clearTextEvt: Event.spent(),
	changeTextEvt: Event<String>.spent(),
}

And then events may be passed down by the StoreConnector to some StatefulWidget, just like any other state:

class MyConnector extends StatelessWidget {
  @override
  Widget build(BuildContext context) {
	return StoreConnector<AppState, ViewModel>(
	  model: ViewModel(),
	  builder: (BuildContext context, ViewModel vm) => MyWidget(
		initialText: vm.initialText,
		clearTextEvt: vm.clearTextEvt,
		changeTextEvt: vm.changeTextEvt,
		onClear: vm.onClear,
	  ));
  }
}

class ViewModel extends BaseModel<AppState> {
  ViewModel();

  String initialText;
  Event clearTextEvt;
  Event<String> changeTextEvt;

  ViewModel.build({
	@required this.initialText,
	@required this.clearTextEvt,
	@required this.changeTextEvt,
  }) : super(equals: [initialText, clearTextEvt, changeTextEvt]);

  @override
  ViewModel fromStore() => ViewModel.build(
		initialText: state.initialText,
		clearTextEvt: state.clearTextEvt,
		changeTextEvt: state.changeTextEvt,
		onClear: () => dispatch(ClearTextAction()),
	  );
}

class ClearTextAction extends ReduxAction<AppState> {
  @override
  AppState reduce() => state.copy(changeTextEvt: Event());
}

class ChangeTextAction extends ReduxAction<AppState> {
  String newText;
  ChangeTextAction(this.newText);

  @override
  AppState reduce() => state.copy(changeTextEvt: Event<String>(newText));
}

This is how it differs: The dumb-widget will "consume" the event in its didUpdateWidget method, and do something with the event payload:

@override
void didUpdateWidget(MyWidget oldWidget) {
  super.didUpdateWidget(oldWidget);
  consumeEvents();
}

void consumeEvents() {
  if (widget.clearTextEvt.consume()) { // Do something }

  var payload = widget.changeTextEvt.consume();
  if (payload != null) { // Do something }
}

The evt.consume() will return the payload once, and then that event is considered "spent".

In more detail, if the event has no value and no generic type, then it's a boolean event. This means evt.consume() returns true once, and then false for subsequent calls. However, if the event has value or some generic type, then Event.consume() returns the value once, and then null for subsequent calls.

So, for example, if you use a controller to hold the text in a TextField:

void consumeEvents() {

	if (widget.clearTextEvt.consume())
	  WidgetsBinding.instance.addPostFrameCallback((_) {
		if (mounted) controller.clear();
	  });

	String newText = widget.changeTextEvt.consume();
	if (newText != null)
	  WidgetsBinding.instance.addPostFrameCallback((_) {
		if (mounted) controller.value = controller.value.copyWith(text: newText);
	  });
  }

Try running the: Event Example.

Can I put mutable events into the store state? #

Events are mutable, and store state is supposed to be immutable. Won't this create problems? No! Don't worry, events are used in a contained way, and were crafted to play well with the Redux infrastructure. In special, their equals() and hashcode() methods make sure no unnecessary widget rebuilds happen when they are used as prescribed.

You can think of events as piggybacking in the Redux infrastructure, and not belonging to the store state. You should just remember not to persist them when you persist the store state.

When should I use events? #

The short answer is that you'll know it when you see it. When you want to do something and it's not obvious how to do it by changing regular store state, it's probably easy to solve it if you try using events instead.

However, we can also give these guidelines:

  1. You may use regular store state to pass constructor parameters to both stateless and stateful widgets.
  2. You may use events to change the internal state of stateful widgets, after they are built.
  3. You may use events to make one-off changes in controllers.
  4. You may use events to make one-off changes in other implicit state like the open state of dialogs or the keyboard.

Advanced event features #

There are some advanced event features you probably won't need, but you should know they exist:

  1. Methods isSpent, isNotSpent and state

    Methods isSpent and isNotSpent tell you if an event is spent or not, without consuming the event. Method state returns the event payload, without consuming the event.

  2. Method Event.from(Event<T> evt1, Event<T> evt2)

    This is a convenience factory method to create EventMultiple, a special type of event which consumes from more than one event. If the first event is not spent, it will be consumed, and the second will not. If the first event is spent, the second one will be consumed. So, if both events are NOT spent, the method will have to be called twice to consume both. If both are spent, returns null.

  3. Method static T consumeFrom<T>(Event<T> evt1, Event<T> evt2)

    This is a convenience static method to consume from more than one event. If the first event is not spent, it will be consumed, and the second will not. If the first event is spent, the second one will be consumed. So, if both events are NOT spent, the method will have to be called twice to consume both. If both are spent, returns null. For example:

String getMessageEvt() {
   return Event.consumeFrom(firstMsgEvt, secondMsgEvt);
 }

Waiting until an Action is finished #

In a real Flutter app it's also the case that some Widgets ask for futures that complete when some async process is done.

If instead of dispatch() you use dispatchFuture(), it will return you a Future<void> that completes as soon as the action is done.

This is an example using the RefreshIndicator widget:

Future<void> downloadStuff() => dispatchFuture(DownloadStuffAction());

return RefreshIndicator(
    onRefresh: downloadStuff;
    child: ListView(...),

Try running the: Dispatch Future Example.

State Declaration #

While your main state class, usually called AppState, may be simple and contain all of the state directly, in a real world application you will probably want to create many state classes and add them to the main state class. For example, if you have some state for the login, some user related state, and some todos in a To-Do app, you can organize it like this:

class AppState {

  final LoginState loginState;
  final UserState userState;
  final TodoState todoState;

 AppState({
	this.loginState,
	this.userState,
	this.todoState,
  });

  AppState copy({
	LoginState loginState,
	UserState userState,
	TodoState todoState,
  }) {
	return AppState(
	  login: loginState ?? this.loginState,
	  user: userState ?? this.userState,
	  todo: todoState ?? this.todoState,
	);
  }

  static AppState initialState() =>
	AppState(
	  loginState: LoginState.initialState(),
	  userState: UserState.initialState(),
	  todoState: TodoState.initialState());

  @override
  bool operator ==(Object other) =>
	identical(this, other) || other is AppState && runtimeType == other.runtimeType &&
	  loginState == other.loginState && userState == other.userState && todoState == other.todoState;

  @override
  int get hashCode => loginState.hashCode ^ userState.hashCode ^ todoState.hashCode;
}

All of your state classes may follow this pattern. For example, the TodoState could be like this:

class TodoState {    
  
  final List<Todo> todos;            

  TodoState({this.todos});

  TodoState copy({List<Todo> todos}) {
    return TodoState(          
      todos: todos ?? this.todos);
  }             
  
  static TodoState initialState() => TodoState(todos: const []);               
  
  @override
  bool operator ==(Object other) {          
    return identical(this, other) || other is TodoState && runtimeType == other.runtimeType && 
      listEquals(todos, other.todos);
  }    
  
  @override
  int get hashCode => const ListEquality.hash(todos);
}

Selectors #

Your connector-widgets usually have a view-model that goes into the store and selects the part of the store the widget needs. If you have some "selecting logic" that you use in different places, you may create a "selector". Selectors may be put in separate files, or they may be put into state classes, as static methods. For example, the TodoState class above could contain a selector to filter out some todos:

static List<Todo> selectTodosForUser(AppState state, User user)
   => state.todoState.todos.where((todo) => (todo.user == user)).toList();

Action Subclassing #

Suppose you have the following AddTodoAction for the To-Do app:


class AddTodoAction extends ReduxAction<AppState> {
  final Todo todo;
  AddTodoAction(this.todo);

  @override
  AppState reduce() {
	if (todo == null) return null;
	else return state.copy(todoState: List.of(state.todoState.todos)..add(todo)));
  }
}

// You would use it like this:
store.dispatch(AddTodoAction(Todo("Buy some beer.")));

Since all actions extend ReduxAction, you may further use object oriented principles to reduce boilerplate. Start by creating an abstract action base class to allow easier access to the sub-states of your store. For example:

abstract class BaseAction extends ReduxAction<AppState> {
  LoginState get loginState => state.loginState;
  UserState get userState => state.userState;
  TodoState get todoState => state.todoState;
  List<Todo> get todos => todoState.todos;
}

And then your actions have an easier time accessing the store state:

class AddTodoAction extends BaseAction {
  final Todo todo;
  AddTodoAction(this.todo);

  @override
  AppState reduce() {
	if (todo == null) return null;
	else return state.copy(todoState: List.of(todos)..add(todo)));
  }
}

As you can see above, instead of writing List.of(state.todoState.todos) you can simply write List.of(todos). It may seem a small reduction of boilerplate, but it adds up.

Another thing you may do is creating more specialized abstract actions, that modify only some part of the state. For example:

abstract class TodoAction extends ReduxAction<AppState> {
  
  TodoState reduceTodoState();
      
  @override
  AppState reduce() {
    TodoState todoState = reduceTodoState();  
    return (todoState == null) ? null : state.copy(todoState: todoState);
  }
}

If you declare those specialized abstract actions, you can have specialized reducers that only need to return that part of the state that changed:

class AddTodoAction extends TodoAction {
  final Todo todo;
  AddTodoAction(this.todo);

  @override
  TodoState reduceTodoState() {
    if (todo == null) return null;
    else return List.of(todos)..add(todo);
  }
}

Abstract Before and After #

Other useful abstract classes you may create provide already overridden before() and after() methods. For example, this abstract class turns on a modal barrier when the action starts, and removes it when the action finishes:

abstract class BarrierAction extends ReduxAction<AppState> {
  void before() => dispatch(WaitAction(true));
  void after() => dispatch(WaitAction(false));
}

Then you could use it like this:

class ChangeTextAction extends BarrierAction {

  @override
  Future<AppState> reduce() async {
	String newText = await read("http://numbersapi.com/${state.counter}");
	return state.copy(
	  counter: state.counter + 1,
	  changeTextEvt: Event<String>(newText));
  }
}

The above BarrierAction is demonstrated in this example.

IDE Navigation #

How does AsyncRedux solve the IDE navigation problem?

During development, if you need to see what some action does, you just tell your IDE to navigate to the action itself (CTRL+B in IntelliJ/Windows, for example) and you have the reducer right there.

If you need to list all of your actions, you just go to the ReduxAction class declaration and ask the IDE to list all of its subclasses.

Logging and Persistence #

Your store optionally accepts lists of actionObservers and stateObservers. The first one may be used for logging, and the second for persistence:

var store = Store<AppState>(
  initialState: state,
  actionObservers: [Log.printer(formatter: Log.verySimpleFormatter)],
  stateObservers: persistor.createStateObservers(),
);

The ActionObserver is an abstract class with an observe method which you can implement to be notified of action dispatching:

abstract class ActionObserver<St> {

   void observe(
      ReduxAction<St> action, 
      int dispatchCount, {
      @required bool ini,
      }
   );
}

The above observer will actually be called twice, one with ini==true for the INITIAL action observation, and one with ini==false for the END action observation,

Meanwhile, the StateObserver is an abstract class which you can implement to be notified of state changes:

abstract class StateObserver<St> {

   void observe(
      ReduxAction<St> action, 
      St stateIni, 
      St stateEnd, 
      int dispatchCount,
      );
}

In more detail:

  1. The INI action observation means the action was just dispatched and haven't changed anything yet. After that, it may do sync stuff, and it may or may not start async processes, depending if its reducer is sync or async.

  2. The END action observation means the action reducer has just finished returning a new state, thus changing the store state. Only after getting END states you may see store changes.

  3. The state observation is therefore called as soon as possible after the store change has taken place, i.e., right after the END action observation. However, it contains a copy of both the state before the action INI and the state after the action END, in case you need to compare them.

Please note, unless the action reducer is synchronous, getting an END action observation doesn't mean that all of the action effects have finished, because the action may have started async processes that may well last into the future. And these processes may later dispatch other actions that will change the store state. However, it does mean that the action can no longer change the state directly.

Observing rebuilds #

Your store optionally accepts a modelObserver, which lets you visualize rebuilds.

The ModelObserver is an abstract class with an observe method which you can implement to be notified, by each StoreConnector currently in the widget tree, whenever there is a state change. You can create your own ModelObserver, but the provided DefaultModelObserver can be used out of the box to print to the console and do basic testing:

var store = Store<AppState>(
  initialState: state,
  modelObserver: DefaultModelObserver(),  
);

This is an example output to the console, showing how MyWidgetConnector responded to 3 state changes:

Model D:1 R:1 = Rebuid:true, Connector:MyWidgetConnector, Model:MyViewModel{B}.
Model D:2 R:2 = Rebuid:false, Connector:MyWidgetConnector, Model:MyViewModel{B}.
Model D:3 R:3 = Rebuid:true, Connector:MyWidgetConnector, Model:MyViewModel{C}.

You can see above that the first and third state changes resulted in a rebuild (Rebuid:true), but the second one did not, probably because the part of the state that changed was not part of MyViewModel.

This example also shows the ModelObserver in action.

Note: You must pass debug:this as a StoreConnector constructor parameter, if you want the ModelObserver to be able to print the StoreConnector type to the output. You can also override your ViewModel.toString() to print out any extra info you need.

The ModelObserver is also useful when you want to create tests to assert that rebuilds happen when and only when the appropriate parts of the state change. For an example, see the Model Observer Test.

How to interact with the database #

The following advice works for any Redux version, including AsyncRedux.

Pretend the user presses a button in the dumb-widget, running a callback which was passed in its constructor. This callback, which was created by the Connector widget, will dispatch an action.

This action's async reducer will connect to the database and get the desired information. You can directly connect to the database from the async reducer, or have a DAO to abstract the database implementation details.

This would be your reducer:

@override
Future<AppState> reduce() async {
	var something = await myDao.loadSomething();
	return state.copy(something: something);
}

This rebuilds your widgets that depend on something, with its new value. The state now holds the new something, and the local store persistor may persist this value to the local file system, if that's what you want.

How to deal with Streams #

The following advice works for any Redux version, including AsyncRedux.

AsyncRedux plays well with Streams, as long as you know how to use them:

  • Don't send the streams down to the dumb-widget, and not even to the Connector. If you are declaring, subscribing to, or unsubscribing from streams inside of widgets, it means you are mixing Redux with some other architecture. You can do that, but it's not recommended and not necessary.
  • Don't put streams into the store state. They are not app state, and they should not be persisted to the local filesystem. Instead, they are something that "generates state".

So, how do you use streams? #

Let's pretend you want to listen to changes to the user name, in a Firestore database. First, create an action to start listening, and another action to cancel. We could name them StartListenUserNameAction and CancelListenUserNameAction.

  • If the stream should run all the time, you may dispatch the start action as soon as the app starts, right after you create the store, possibly in main. And cancel it when the app finishes.

  • If the stream should run only when the user is viewing some screen, you may dispatch the action from the initState method of the screen widget, and cancel it from the dispose method. Note: More precisely, these things are done by the callbacks that the Connectors create and send down to the stateful dumb-widgets.

  • If the stream should run only when some actions demand it, their reducers may dispatch the actions to start and cancel as needed.

Where the stream subscriptions themselves are stored #

As discussed above, you should NOT put them in the store state. Instead save them in some convenient place elsewhere, where your reducers may access them. Remember you only need to access them from the reducers. If you have separate business and client layers, put them into the business layer.

Some ideas:

  • Put them as static variables of the specific actions that start them. For example, userNameStream could be a static field of the StartListenUserNameAction class.

  • Put them in the state classes that most relate to them, but as static variables, not instance variables (which would be store state). For example, if your AppState contains some UserState, then userNameStream could be a static field of the UserState class.

  • Save them in global static variables.

  • Use a service locator, like get_it.

Or put them wherever you think makes sense. In all cases above, you can still inject them with mocks, for tests.

How do streams pass their information to the store and ultimately to the widgets? #

When you create the stream, define its callback so that it dispatches an appropriate action. Each time the stream gets some data it will pass it to this action's constructor. The action's reducer will put the data into the store state, from where it will be automatically sent down to the widgets that observe them (through their Connector/ViewModel).

For example:

Stream<QuerySnapshot> stream = query.snapshots();

streamSub = stream.listen((QuerySnapshot querySnapshot) {
  dispatch(DoSomethingAction(querySnapshot.documentChanges));
  }, onError: ...);

To sum up: #

  1. Put your stream subscriptions where they can be accessed by the reducers, but NOT inside of the store state.

  2. Don't use streams directly in widgets (not in the Connector widget, and not in the dumb-widget).

  3. Create actions to start and cancel streams, and call them when necessary.

  4. The stream callback should dispatch actions to put the snapshot data into the store state.

You probably have your own way of organizing your directory structure, but if you want some recommendation, here it goes.

First, separate your directory structure by client and business. The client directory holds Flutter stuff like widgets, including your connector and dumb widgets. The business directory holds the business layer stuff, including the store, state, and code to access the database and to persist the state to disk.

├── business
│   ├── lib
│   ├── test
│   └── pubspec.yaml
└── client
    ├── lib
    ├── test
    └── pubspec.yaml

Edit the client/pubspec.yaml file to contain this:

dependencies:
  business:
    path: ../business/

However, business/pubspec.yaml should contain no references to the client. This guarantees the client code can use the business code, but the business code can't access the client code.

In business/lib create separate directories for your main features, and only then create directories like actions, models, dao or other.

Note that AsyncRedux has no separate reducers nor middleware, so this simplifies the directory structure in relation to vanilla Redux.

Your final directory structure would then look something like this:

├── business
│   ├── lib
│   │   ├── login
│   │   │   ├── actions
│   │   │   │   ├── login_action.dart
│   │   │   │   ├── logout_action.dart
│   │   │   │   └── ...
│   │   │   └── models
│   │   │       └── login_state.dart
│   │   ├── todos
│   │   │   ├── actions
│   │   │   │   └── ...
│   │   │   └── models
│   │   │       ├── todos_state.dart
│   │   │       └── todo.dart
│   │   └── users
│   │       ├── actions
│   │       │   ├── create_user_action.dart
│   │       │   ├── change_user_action.dart
│   │       │   ├── delete_user_action.dart
│   │       │   └── ...
│   │       └── models
│   │           └── user.dart
│   ├── test
│   │   ├── login
│   │   │   ├── login_STATE_test.dart
│   │   │   ├── login_action_test.dart
│   │   │   ├── logout_action_test.dart
│   │   │   └── ...
│   │   ├── todos
│   │   │   ├── todos_STATE_test.dart
│   │   │   └── todo_test.dart
│   │   └── users
│   │       └── user_test.dart
│   ├── pubspec.yaml
│   └── ...
└── client
    ├── lib
    │   ├── login
    │   │   ├── login_connector_widget.dart
    │   │   └── login_widget.dart
    │   └── todos
    │       ├── todos_connector_widget.dart
    │       └── todos_widget.dart
    ├── test
    │   ├── login
    │   │   ├── login_CONNECTOR_test.dart
    │   │   └── login_PRESENTATION.dart
    │   └── todos
    │       ├── todos_CONNECTOR_test.dart
    │       └── todos_PRESENTATION.dart
    └── pubspec.yaml

Architectural discussion #

Reading the following text is not important for the practical use of AsyncRedux, and is meant only for those interested in architectural discussions:

Is AsyncRedux really Redux? #

According to redux.js.org there are three principles to Redux:

  1. The state of your whole application is stored in an object tree within a single store.

    That’s true for AsyncRedux.

  2. The only way to change the state is to emit an action, an object describing what happened.

    That’s also true for AsyncRedux.

  3. To specify how the state tree is transformed by actions, you write pure reducers.

    Ok, so how about middleware? It's not possible to create real world applications without async calls and external databases access. So, even in vanilla Redux, actions start async processes that yield results that only then will be put into the store state, through reducers. So it's not true that the state tree depends only on pure functions. You can't separate the pure part and call it a reducer, and then conveniently forget about the impure/async part. In other words, you have A and B. A is simple and pure, but we can't call it a reducer and say that's part of our principles, and then forget about B. Async Redux acknowledges that B is also part of the solution, and then creates tools to deal with it as easily as possible. The litmus test here, to prove that AsyncRedux is Redux, is that you can have a 1 to 1 mapping from vanilla Redux reducers+middleware code to the AsyncRedux sync+async reducers code. The same async code will call the same pure code. You just organize it differently to avoid boilerplate. Another way to look at it is that at first glance the AsyncRedux reducer doesn't appear to be a pure function. Pure function reducers are the wall of sanity against the side-effects managed by middleware via thunks, sagas, observables, etc. But when you take a second look, return state.copy(...) is the pure reducer, and everything else in reduce() is essentially middleware.

Besides the reduction of boilerplate, what are the main advantages of the AsyncRedux architecture? #

In vanilla Redux it's easy to reason about the code at first, when it's just pure function reducers, but it gets difficult to understand the whole picture as soon as you have to add complex Middleware. When you see a side by side comparison of code written for vanilla Redux and for AsyncRedux, the code is easier to understand with AsyncRedux.

Also, vanilla Redux makes it easy to test its pure functions reducers, but it doesn't help at all with testing the middleware. In contrast, since AsyncRedux natively takes async code into consideration, its testing capabilities (the StoreTester) make it easy to test the code as a whole.

AsyncRedux also helps with code errors, by simply letting your reducers throw errors. AsyncRedux will catch them and deal with them appropriately, while vanilla Redux forces your middleware to catch errors and maybe even dispatch actions do deal with them.

Is AsyncRedux a minimalist or lightweight Redux version? #

No. AsyncRedux is concerned with being "easy to use", not with being lightweight. In terms of library code size it's larger than the original Redux implementation. However it's still very small, and will make the total application code smaller than with the vanilla implementation, because of the boilerplate reduction. In terms of speed/performance there should be no differences in respect to the vanilla implementation.

Is the AsyncRedux architecture useful for small projects? #

It's usually said that you should not use Redux for small projects, because of the extra boilerplate and limitations. Maybe it's not worth the effort. However, since AsyncRedux is easier than vanilla Redux and has far less boilerplate, the limit of code complexity where a robust architecture starts making sense is much lower.


The AsyncRedux code is based upon packages redux by Brian Egan, and flutter_redux by Brian Egan and John Ryan. Also uses code from package equatable by Felix Angelov. Special thanks: Eduardo Yamauchi and Hugo Passos helped me with the async code, checking the documentation, testing everything and making suggestions. This work started after Thomas Burkhart explained to me why he didn't like Redux. Reducers as methods of action classes were shown to me by Scott Stoll and Simon Lightfoot.

The Flutter packages I've authored:


Marcelo Glasberg:
https://github.com/marcglasberg
https://twitter.com/glasbergmarcelo
https://stackoverflow.com/users/3411681/marcg
https://medium.com/@marcglasberg

[2.0.6] - 2019/10/07

  • Added sync_async_test.dart
  • Doc warning about async reducer returning completed future (missing await).

[2.0.5] - 2019/10/05

  • Better typing of StoreProvider.dispatch and StoreProvider.dispatchFuture.

[2.0.4] - 2019/10/01

  • StoreTester: waitUntilError and waitUntilErrorGetLast.

[2.0.3] - 2019/09/21

  • NavigateAction tests.
  • Navigation arguments.

[2.0.2] - 2019/09/21

  • UserExceptionAction.

[2.0.1] - 2019/09/19

  • Fix: UserException.dialogContent accepts String as cause.

[2.0.0] - 2019/09/17

  • Breaking change: ErrorObserver API.
  • StoreTester parameter: shouldThrowUserExceptions (see issue).

[1.4.3] - 2019/09/15

  • Alternative: Use AsyncRedux with Provider (package provider_for_redux).

[1.4.1] - 2019/09/06

  • Flutter Awesome badge, and Pub badge.

[1.4.0] - 2019/09/02

  • Fix: dispatchFuture getter in ReduxAction.

[1.3.9] - 2019/08/31

  • NavigateAction.navigatorKey getter.

[1.3.8] - 2019/08/30

  • Alternatives to the Connector (StoreProvider static methods).
  • Waiting until an Action is finished (dispatchFuture).

[1.3.7] - 2019/08/28

  • ModelObserver and DefaultModelObserver.

[1.3.5] - 2019/08/27

  • README improvement.

[1.3.3] - 2019/08/26

  • StoreConnector's converter and model parameters.

[1.2.3] - 2019/08/23

  • StoreTester timeout message.

[1.2.0] - 2019/08/22

  • Doc improvement. StoreTester improvements.

[1.1.3] - 2019/08/21

  • StoreTester: waitCondition and waitConditionGetLast.

[1.1.2] - 2019/08/13

  • README improvement.

[1.1.1] - 2019/08/10

  • Ignore actions in the StoreTester.

[1.1.0] - 2019/08/07

  • Correct stacktrace for unwrapped action errors.

[1.0.9] - 2019/08/07

  • Error message improvement.

[1.0.4] - 2019/08/05

  • Store tester.

[1.0.0] - 2019/08/05

  • Initial commit.

example/README.md

Examples #

  1. main

    This example shows a counter and a button. When the button is tapped, the counter will increment synchronously.

    In this simple example, the app state is simply a number (the counter), and thus the store is defined as Store<int>. The initial state is 0.

  2. main_increment_async

    This example shows a counter, a text description, and a button. When the button is tapped, the counter will increment synchronously, while an async process downloads some text description that relates to the counter number.

  3. main_before_and_after

    This example shows a counter, a text description, and a button. When the button is tapped, the counter will increment synchronously, while an async process downloads some text description that relates to the counter number.

    While the async process is running, a redish modal barrier will prevent the user from tapping the button. The model barrier is removed even if the async process ends with an error, which can be simulated by turning off the internet connection (putting the phone in airplane mode).

  4. main_static_view_model

    This example shows how to use the same ViewModel architecture of flutter_redux. This is specially useful if you are migrating from flutter_redux.
    Here, you use the StoreConnector's converter parameter, instead of the model parameter. And ViewModel doesn't extend BaseModel, but has a static factory: converter: (store) => ViewModel.fromStore(store).

  5. main_before_and_after_STATE_test

    This example displays the testing capabilities of AsyncRedux: How to test the store, actions, sync and async reducers, by using the StoreTester. Important: To run the tests, put this file in a test directory.

  6. main_show_error_dialog

    This example lets you enter a name and click save. If the name has less than 4 chars, an error dialog will be shown.

  7. main_navigate

    This example shows a route in the screen, all red. When you tap the screen it will push a new route, all blue. When you tap the screen again it will pop the blue route.

  8. main_event_redux

    This example shows a text-field, and two buttons. When the first button is tapped, an async process downloads some text from the internet and puts it in the text-field. When the second button is tapped, the text-field is cleared.

    This is meant to demonstrate the use of events to change a controller state.

    It also demonstrates the use of an abstract class to override the action's before() and after() methods.

  9. main_dispatch_future

    This example demonstrates how to get a Future that completes when an action is done. It shows a list of number descriptions. If you pull to refresh the page (scroll above the top of the page) a RefreshIndicator will appear until the list is updated with different data.

Use this package as a library

1. Depend on it

Add this to your package's pubspec.yaml file:


dependencies:
  async_redux: ^2.0.6

2. Install it

You can install packages from the command line:

with Flutter:


$ flutter pub get

Alternatively, your editor might support flutter pub get. Check the docs for your editor to learn more.

3. Import it

Now in your Dart code, you can use:


import 'package:async_redux/async_redux.dart';
  
Popularity:
Describes how popular the package is relative to other packages. [more]
83
Health:
Code health derived from static analysis. [more]
99
Maintenance:
Reflects how tidy and up-to-date the package is. [more]
100
Overall:
Weighted score of the above. [more]
91
Learn more about scoring.

We analyzed this package on Oct 11, 2019, and provided a score, details, and suggestions below. Analysis was completed with status completed using:

  • Dart: 2.5.1
  • pana: 0.12.21
  • Flutter: 1.9.1+hotfix.4

Platforms

Detected platforms: Flutter

References Flutter, and has no conflicting libraries.

Health suggestions

Fix lib/src/store.dart. (-0.50 points)

Analysis of lib/src/store.dart reported 1 hint:

line 923 col 50: 'ignoreChange' is deprecated and shouldn't be used.

Fix lib/src/store_tester.dart. (-0.50 points)

Analysis of lib/src/store_tester.dart reported 1 hint:

line 530 col 8: The declaration '_addAll' isn't referenced.

Format lib/src/navigate_action.dart.

Run flutter format to format lib/src/navigate_action.dart.

Format lib/src/user_exception.dart.

Run flutter format to format lib/src/user_exception.dart.

Dependencies

Package Constraint Resolved Available
Direct dependencies
Dart SDK >=2.2.2 <3.0.0
flutter 0.0.0
logging ^0.11.3+2 0.11.3+2
Transitive dependencies
collection 1.14.11 1.14.12
meta 1.1.7
sky_engine 0.0.99
typed_data 1.1.6
vector_math 2.0.8
Dev dependencies
flutter_test
test ^1.6.3