/ReactorKit

A framework for a reactive and unidirectional Swift application architecture

Primary LanguageSwiftMIT LicenseMIT

ReactorKit

Swift CocoaPods Platform Build Status Codecov CocoaDocs

ReactorKit is a framework for a reactive and unidirectional Swift application architecture. This repository introduces the basic concept of ReactorKit and describes how to build an application using ReactorKit.

You may want to see the Examples section first if you'd like to see the actual code. Visit the API Reference for code-level documentation.

Table of Contents

Basic Concept

ReactorKit is a combination of Flux and Reactive Programming. The user actions and the view states are delivered to each layer via observable streams. These streams are unidirectional: the view can only emit actions and the reactor can only emit states.

flow

Design Goal

  • Testability: The first purpose of ReactorKit is to separate the business logic from a view. This can make the code testable. A reactor doesn't have any dependency to a view. Just test reactors and test view bindings. See Testing section for details.
  • Start Small: ReactorKit doesn't require the whole application to follow a single architecture. ReactorKit can be adopted partially, for one or more specific views. You don't need to rewrite everything to use ReactorKit on your existing project.
  • Less Typing: ReactorKit focuses on avoiding complicated code for a simple thing. ReactorKit requires less code compared to other architectures. Start simple and scale up.

View

A View displays data. A view controller and a cell are treated as a view. The view binds user inputs to the action stream and binds the view states to each UI component. There's no business logic in a view layer. A view just defines how to map the action stream and the state stream.

To define a view, just have an existing class conform a protocol named View. Then your class will have a property named reactor automatically. This property is typically set outside of the view.

class ProfileViewController: UIViewController, View {
  var disposeBag = DisposeBag()
}

profileViewController.reactor = UserViewReactor() // inject reactor

When the reactor property has changed, bind(reactor:) gets called. Implement this method to define the bindings of an action stream and a state stream.

func bind(reactor: ProfileViewReactor) {
  // action (View -> Reactor)
  refreshButton.rx.tap.map { Reactor.Action.refresh }
    .bind(to: reactor.action)
    .disposed(by: self.disposeBag)

  // state (Reactor -> View)
  reactor.state.map { $0.isFollowing }
    .bind(to: followButton.rx.isSelected)
    .disposed(by: self.disposeBag)
}

Storyboard Support

Use StoryboardView protocol if you're using a storyboard to initialize view controllers. Everything is same but the only difference is that the StoryboardView performs a binding after the view is loaded.

let viewController = MyViewController()
viewController.reactor = MyViewReactor() // will not executes `bind(reactor:)` immediately

class MyViewController: UIViewController, StoryboardView {
  func bind(reactor: MyViewReactor) {
    // this is called after the view is loaded (viewDidLoad)
  }
}

Reactor

A Reactor is an UI-independent layer which manages the state of a view. The foremost role of a reactor is to separate control flow from a view. Every view has its corresponding reactor and delegates all logic to its reactor. A reactor has no dependency to a view, so it can be easily tested.

Conform to the Reactor protocol to define a reactor. This protocol requires three types to be defined: Action, Mutation and State. It also requires a property named initialState.

class ProfileViewReactor: Reactor {
  // represent user actions
  enum Action {
    case refreshFollowingStatus(Int)
    case follow(Int)
  }

  // represent state changes
  enum Mutation {
    case setFollowing(Bool)
  }

  // represents the current view state
  struct State {
    var isFollowing: Bool = false
  }

  let initialState: State = State()
}

An Action represents a user interaction and State represents a view state. Mutation is a bridge between Action and State. A reactor converts the action stream to the state stream in two steps: mutate() and reduce().

flow-reactor

mutate()

mutate() receives an Action and generates an Observable<Mutation>.

func mutate(action: Action) -> Observable<Mutation>

Every side effect, such as an async operation or API call, is performed in this method.

func mutate(action: Action) -> Observable<Mutation> {
  switch action {
  case let .refreshFollowingStatus(userID): // receive an action
    return UserAPI.isFollowing(userID) // create an API stream
      .map { (isFollowing: Bool) -> Mutation in
        return Mutation.setFollowing(isFollowing) // convert to Mutation stream
      }

  case let .follow(userID):
    return UserAPI.follow()
      .map { _ -> Mutation in
        return Mutation.setFollowing(true)
      }
  }
}

reduce()

reduce() generates a new State from a previous State and a Mutation.

func reduce(state: State, mutation: Mutation) -> State

This method is a pure function. It should just return a new State synchronously. Don't perform any side effects in this function.

func reduce(state: State, mutation: Mutation) -> State {
  var state = state // create a copy of the old state
  switch mutation {
  case let .setFollowing(isFollowing):
    state.isFollowing = isFollowing // manipulate the state, creating a new state
    return state // return the new state
  }
}

transform()

transform() transforms each stream. There are three transform() functions:

func transform(action: Observable<Action>) -> Observable<Action>
func transform(mutation: Observable<Mutation>) -> Observable<Mutation>
func transform(state: Observable<State>) -> Observable<State>

Implement these methods to transform and combine with other observable streams. For example, transform(mutation:) is the best place for combining a global event stream to a mutation stream. See the Global States section for details.

These methods can be also used for debugging purposes:

func transform(action: Observable<Action>) -> Observable<Action> {
  return action.debug("action") // Use RxSwift's debug() operator
}

Advanced

Service

ReactorKit has a special layer named Service. A service layer does the actual business logic. A reactor is a middle layer between a view and a service which manages event streams. When a reactor receives an user action from a view, the reactor calls the service logic. The service makes a network request and sends the response back to the reactor. Then the reactor create a mutation stream with the service response.

Here is an example of service:

protocol UserServiceType {
  func user(id: Int) -> Observable<User>
  func follow(id: Int) -> Observable<Void>
}

final class UserService: Service, UserServiceType {
  func user(id: Int) -> Observable<User> {
    return foo()
  }
  
  func follow(id: Int) -> Observable<Void> {
    return bar()
  }
}

Global States

Unlike Redux, ReactorKit doesn't define a global app state. It means that you can use anything to manage a global state. You can use a Variable, a PublishSubject or even a reactor. ReactorKit doesn't force to have a global state so you can use ReactorKit in a specific feature in your application.

There is no global state in the Action → Mutation → State flow. You should use transform(mutation:) to transform the global state to a mutation. Let's assume that we have a global Variable which stores the current authenticated user. If you'd like to emit a Mutation.setUser(User?) when the currentUser is changed, you can do as following:

var currentUser: Variable<User> // global state

func transform(mutation: Observable<Mutation>) -> Observable<Mutation> {
  return Observable.merge(mutation, currentUser.map(Mutation.setUser))
}

Then the mutation will be emitted each time the view sends an action to a reactor and the currentUser is changed.

View Communication

You must be familiar with callback closures or delegate patterns for communicating between multiple views. ReactorKit recommends you to use reactive extensions for it. The most common example of ControlEvent is UIButton.rx.tap. The key concept is to treat your custom views as UIButton or UILabel.

view-view

Let's assume that we have a ChatViewController which displays messages. The ChatViewController owns a MessageInputView. When an user taps the send button on the MessageInputView, the text will be sent to the ChatViewController and ChatViewController will bind in to the reactor's action. This is an example MessageInputView's reactive extension:

extension Reactive where Base: MessageInputView {
  var sendButtonTap: ControlEvent<String> {
    let source = base.sendButton.rx.tap.withLatestFrom(...)
    return ControlEvent(events: source)
  }
}

You can use that extension in the ChatViewController. For example:

messageInputView.rx.sendButtonTap
  .map(Reactor.Action.send)
  .bind(to: reactor.action)

Testing

ReactorKit has a built-in functionality for a testing. You'll be able to easily test both a view and a reactor with a following instruction.

What to test

First of all, you have to decide what to test. There are two things to test: a view and a reactor.

  • View
    • Action: is a proper action sent to a reactor with a given user interaction?
    • State: is a view property set properly with a following state?
  • Reactor
    • State: is a state changed properly with an action?

View testing

A view can be tested with a stub reactor. A reactor has a property stub which can log actions and force change states. If a reactor's stub is enabled, both mutate() and reduce() are not executed. A stub has these properties:

var isEnabled: Bool { get set }
var state: Variable<Reactor.State> { get }
var action: ActionSubject<Reactor.Action> { get }
var actions: [Reactor.Action] { get } // recorded actions

Here are some example test cases:

func testAction_refresh() {
  // 1. prepare a stub reactor
  let reactor = MyReactor()
  reactor.stub.isEnabled = true

  // 2. prepare a view with a stub reactor
  let view = MyView()
  view.reactor = reactor

  // 3. send an user interaction programatically
  view.refreshControl.sendActions(for: .valueChanged)

  // 4. assert actions
  XCTAssertEqual(reactor.stub.actions.last, .refresh)
}

func testState_isLoading() {
  // 1. prepare a stub reactor
  let reactor = MyReactor()
  reactor.stub.isEnabled = true

  // 2. prepare a view with a stub reactor
  let view = MyView()
  view.reactor = reactor

  // 3. set a stub state
  reactor.stub.state.value = MyReactor.State(isLoading: true)

  // 4. assert view properties
  XCTAssertEqual(view.activityIndicator.isAnimating, true)
}

Reactor testing

A reactor can be tested independently.

func testIsBookmarked() {
  let reactor = MyReactor()
  reactor.action.onNext(.toggleBookmarked)
  XCTAssertEqual(reactor.currentState.isBookmarked, true)
  reactor.action.onNext(.toggleBookmarked)
  XCTAssertEqual(reactor.currentState.isBookmarked, false)
}

Sometimes a state is changed more than one time for a single action. For example, a .refresh action sets state.isLoading to true at first and sets to false after the refreshing. In this case it's difficult to test state.isLoading with currentState so you might need to use RxTest or RxExpect. Here is an example test case using RxExpect:

func testIsLoading() {
  RxExpect("it should change isLoading") { test in
    let reactor = test.retain(MyReactor())
    test.input(reactor.action, [
      next(100, .refresh) // send .refresh at 100 scheduler time
    ])
    test.assert(reactor.state.map { $0.isLoading })
      .since(100) // values since 100 scheduler time
      .assert([
        true,  // just after .refresh
        false, // after refreshing
      ])
  }
}

Conventions

ReactorKit suggests some conventions to write clean and concise code.

  • You must create a reactor outside of the view and pass it to the view's reactor property.

    Good

    let view = MyView()
    view.reactor = MyViewReactor(provider: provider)

    Bad

    class MyView: UIView, View {
      init() {
        self.reactor = MyViewReactor()
      }
    }

Examples

  • Counter: The most simple and basic example of ReactorKit
  • GitHub Search: A simple application which provides a GitHub repository search
  • RxTodo: iOS Todo Application using ReactorKit
  • Cleverbot: iOS Messaging Application using Cleverbot and ReactorKit
  • Drrrible: Dribbble for iOS using ReactorKit (App Store)
  • Passcode: Passcode for iOS RxSwift, ReactorKit and IGListKit example

Dependencies

Requirements

  • Swift 4
  • iOS 8
  • macOS 10.11
  • tvOS 9.0
  • watchOS 2.0

Installation

ReactorKit officially supports CocoaPods only.

Podfile

pod 'ReactorKit'

Contribution

Any discussions and pull requests are welcomed 💖

  • To development:

    $ TEST=1 swift package generate-xcodeproj
  • To test:

    $ swift test

Community

Who's using ReactorKit


StyleShare Kakao Wantedly

DocTalk Constant Contact Wantedly

Are you using ReactorKit? Please let me know!

Changelog

  • 2017-04-18
    • Change the repository name to ReactorKit.
  • 2017-03-17
    • Change the architecture name from RxMVVM to The Reactive Architecture.
    • Every ViewModels are renamed to ViewReactors.

License

ReactorKit is under MIT license. See the LICENSE for more info.