reframe-middleware: the ‘action first’ approach to redux

Reframe-middleware is an alternative way of handling actions in redux.dart, inspired by Clojurescript’s Re-frame.

How to use

Step #1: Add reframeReducer and reframeMiddleware to your redux.dart Store:

import 'package:reframe_middleware';

final store = Store<AppState>(
  initialState: AppState(), 
	middleware: [reframeMiddleware(), thirdPartyMiddleware, ...]);

Step #2: Define an action that extends Action and implements handle :

Synchronous, pure action:

import 'package:reframe_middleware';

class IncrementAction extends Action {
  ReframeResponse<AppState> handle(AppState state) =>
        state.copy(count: state.count + 1));

Step #3: Dispatch the action like normal:


Step #4: … There is no step #4! You’re done!

What happens next:

reframeMiddleware will accept your dispatched action and call action.handle(state):

// Type signature required by redux-dart
typedef ReduxMiddleware = void Function(Store<AppState>, dynamic, NextDispatcher);

Middleware reframeMiddleware() => (store, action, next) {
	if (action is Action)
			// StateUpdate: special action, ferries new state to reframeReducer
			..nextState.ifPresent((newState) => store.dispatch(StateUpdate(newState)))
			..effect().then((actions) => actions.forEach(store.dispatch));

	// propagate action to any 3rd party middleware
	// and, eventually, to reframeReducer

Calling action.handle(state) returns a ReframeResponse, which contains a state update and/or a side-effect:

// A side-effect asynchronously resolves to a list of additional Actions.
typedef SideEffect = Future<List<Action>> Function();
Future<List<Action>> noEffect() async => [];

class ReframeResponse<S> {
  final Optional<S> nextState;
  final SideEffect effect;

  const ReframeResponse({
    this.nextState = const Optional.absent(),
    this.effect = noEffect,

reframeMiddleware will dispatch a StateUpdate action that ferries the new state to the reducer:

..nextState.ifPresent((newState) => store.dispatch(StateUpdate(newState)))

reframeReducer does the actual state swap:

// Type signature required by flutter-redux
AppState reframeReducer(AppState oldState, dynamic action) =>
	action is StateUpdate ? action.newState : oldState

reframeMiddleware will run the side-effect and dispatch the resulting additional actions:

	..effect().then((actions) => actions.forEach(store.dispatch));


How do I handle async logic? Do I need to add thunkMiddleware?

Reframe-middleware already comes capable of handling async or impure logic — that’s what ReframeResponse.effect is for.

In reframe-middleware, asynchronous logic is ‘first class’ (built-in), not ‘second class’ (added as a dependency like thunkMiddleware).

Here’s an example of a side-effectful action:

class AsyncIncrementAction extends Action {
  ReframeResponse<int> handle(AppState state) =>
      ReframeResponse.sideEffect(() =>
          Future.delayed(milliseconds: 1000)
              .then((_) => [IncrementEvent()]));

Does this replace redux.dart or flutter-redux?

No. Reframe-middleware is supposed to be used with redux.dart (in the same way e.g. Flutter redux_persist is).

Reframe-middleware, like redux.dart, can be used with or without Brian Egan’s excellent flutter-redux.

What about missing feature from Clojurescript re-frame?

This library is only about re-frame-style middleware, adapted in a manner suitable for typed functional programming (as much as this can be done in Dart); it does not include all features of re-frame.

Derived calculations via subscriptions, etc. are thus not included. :’-(

Reframe-middleware also takes a slightly different approach to side-effects and does not use side-effect handlers as re-frame does — though it would be possible to add them, modeling an effect as an action, including e.g. ‘effect-handlers’, etc.

Isn’t this coupling reducers and actions, which is something Dan Abramov, the creator of redux.js, has warned against?

Short answer: Yes.

Long answer:

Dan has made several arguments (e.g. here and here).

The most often-cited concern appears to be about scale (“big teams can work on overlapping features without constant merge conflicts”).

This suspiciously appeals to outside contingencies to justify what is presented as a fundamental feature of redux (“The whole point of Flux/Redux is to decouple actions and reducers”).

In contrast, Clojurescript’s re-frame has broader way of thinking about why your app looks the way it does, how it can change and — most importantly — why it changes.

That is, re-frame couples an action and its state-changes and side-effects because this coupling represents a fundamental feature of how we think about how event-driven systems change.

(This part of the re-frame Readme is worth reading, even if you never use Clojurescript or re-frame.)

See ‘Motivation’ below for a longer explanation and for code comparisons of re-frame vs. traditional redux.

Motivation: Actions, not ‘pure state updates’ (reducers), are the core of redux

The most important question in re-frame and redux is ‘What does an action mean?’ i.e. ‘Which state changes and side-effects does this action cause?’

Re-frame is the ‘action first’ way of reasoning about your app.

Libraries like Elm, re-frame, and redux.js are all fundamentally based on two principles:

  1. UI is explained by state (i.e. state at some time t), and
  2. state is explained by actions (i.e. state at t0 + action => state at t1)

The change produced by an action is described by some function f: f(state at t0, action) => state at t1

The most important thing, then, is to understand what an action means — i.e. which state changes and side-effects does this action cause?

Re-frame makes this easy:

  1. each action has its own action-handler which describes both the state updates and the side-effects caused by the action, and
  2. an action and its handler are co-defined, i.e. we cannot define an action without also defining its handler, and every dispatched action merely calls its own handler (action.handle(state, …))

In contrast, traditional redux makes this hard:

  1. an action can map to multiple reducers and/or middlewares, and so the resultant state changes and side-effects must be coordinated across (in the worst case) every reducer and middleware, and
  2. an action is not bound to any given reducer or middleware; we have to manually connect them (e.g. switch, if-else, TypedReducer etc.) and write a test, and
  3. side-effects are treated as second-class; we must add e.g. thunkMiddleware as a dependency.

(See below for an explanation of each point, along with code examples.)

Motivation in depth, with code samples:

Re-frame: An action’s event-handler is the single place to understand an action’s state updates and side-effects

class SetCountAction extends Action {
  final int number;

  const SetCountAction(this.number);

  ReframeResponse<AppState> handle(AppState state) =>
        state.copy(counter: state.counter.copy(count: number)));

Traditional redux: There is no single place to understand an action’s state-updates and side-effects

// spread throughout our codebase, found via e.g. use search:
// my_dogs_module/x.dart:
reducerX(state, actionA)

// my_friends_module/k.dart:
reducerK(state, actionA)

// my_favorite_module/favorite.dart
reducerF(state, actionA)

(And those are just the reducers in traditional redux — don’t forget the middlewares too!)

Conflicts are also hard to spot when the state change and side effect logic of an action are not centralized. Let’s take a look inside reducerX and reducerF:

reducerX(state, actionA) => state.counter + 1
reducerF(state, actionA) => state.counter - 1 // oops!

Re-frame: an action’s handler is co-defined and guaranteed to be called (no room for error)

Every action extends from this class and must implement handle:

abstract class Action {
  const Action();

  ReframeResponse<AppState> handle(AppState state);

As we saw above, re-frame uses a single middleware whose only responsibility is to call action.handle and run the state changes and side-effects described in the handler response.

Traditional redux: actions must be manually connected to reducer(s)/middleware(s) in a verbose, error-prone manner

An action is not bound to any given reducer or middleware; we have to manually connect them (e.g. switch, if-else, TypedReducer etc.) and write a test to ensure an action is being consumed.

Consider a common way of matching a redux action to its reducer and/or middleware(s):

if action is ActionA: 
	return reducerA(action);
else if action is ActionB:
	return reducerB(action);
else if ...
else ...

Example from redux-dart docs:

// from redux-dart's
  if (action is AddItemAction) {
    return new AppState(
      new List.from(state.items)..add(action.item), 
  } else if (action is RemoveItemAction) {
    return new AppState(
      new List.from(state.items)..remove(action.item), 
  } else if (action is PerformSearchAction) {
    return new AppState(state.items, action.query);
  } else {
    return state;

A typed version of a reducer is sometimes offered as an alternative:

TypedReducer<A, Z>(reducerZ),
TypedReducer<B, Y>(reducerY),
TypedReducer<C, X> ...

Example from redux-dart docs:

// from redux-dart's

// Compose these smaller functions into the full `itemsReducer`.
Reducer<List<String>> itemsReducer = combineReducers<List<String>>([
  // Each `TypedReducer` will glue Actions of a certain type to the given 
  // reducer! This means you don't need to write a bunch of `if` checks 
  // manually, and can quickly scan the list of `TypedReducer`s to see what 
  // reducer handles what action.
  new TypedReducer<List<String>, AddItemAction>(addItemReducer),
  new TypedReducer<List<String>, RemoveItemAction>(removeItemReducer),

(Note the justification in the comment in the above code — if the goal is to quickly identify ‘which reducer handles which action’, isn’t it better to just have action-handlers?)