Rx State Reducer

February 23, 2019 · View on GitHub

This repository demonstrates a pattern of asynchronous state management based on ideas of reactive programming, state reducer and unidirectional data flow. This is not a library, because you don’t need yet another library to implement this.

Running

  1. Install Carthage
brew install carthage
  1. Build the dependencies
carthage bootstrap --platform iOS

Problem

Asynchronous state management is a non-trivial modelling exercise and requires a reliable approach to keep it concise and maintainable.

Most trivial applications start without any explicit state management approach. However, things quickly get out of hand when the number of states in which a system can reside starts to grow.

Finite-State Machine and State Pattern can help manage synchronous state transitions, but are not designed to handle asynchronous behaviour.

Unless you can model your entire system synchronously, a single asynchronous source breaks imperative programming.

-- Jake Wharton

Solution

Explicitly define the state of a system, then use a reducer function to compute a new state based on the previous state. Use reactive streams to handle asynchronous tasks and make data flow in one direction.

  1. An event is received and transformed to an Action.
  2. The Action is handled and transformed to Mutation. This transformation is needed when Actions require asynchronous handling. Such Actions produce multiple Mutations (e.g., (1) loading, (2) loaded)
  3. Each Mutation is passed to the Reducer function, which uses it to transform the current State to a new State.

Where:

  • Action - an interpretation of the system or user event
  • Mutation - a result of the Action handling
  • State - a model of the system or use case
  • Reducer - a pure function which takes a State and a Mutation and produces a new State

Example

  1. View receives a user event.
  2. View transforms the event to an Action.
  3. View propagates the Action to Interactor.
  4. Interactor handles the Action and transforms it to Mutations.
  5. Interactor uses each Mutation to transform the current State to a new State.
  6. Interactor propagates the new State to View
  7. View updates itself using the new State. The unidirectional feedback loop is now complete.

Given that actions is Observable<Action>, states is Observable<State>, the unidirectional feedback loop is expressed as:

states = actions.flatMap(handleAction).scan(State.initial, accumulator: reduce)

Action and Mutation are expressed as an enum:

enum Action {
    case reload
}

enum Mutation {
    case loading
    case records([Record])
    case failure(Error)
}

State is expressed as a struct (if there is a strict number of finite states, then enum):

struct State {
    var records: [Record]
    var isLoading: Bool
    var error: Error?

    static var initial = State(records: [], isLoading: false, error: nil)
}

Reducer is a pure function:

func reduce(state: State, mutation: Mutation) -> State {
    var state = state
    
    switch mutation {
    case let records(records):
        state.records = records
    case let failure(error):
        ...
    case loading:
        ...
    }
    return state
}

The function handleAction is asynchronous:

func handleAction(_ action: Action) -> Observable<Mutation> {
    switch action {
    case reload:
        return gateway.fetchRecords()
            .map(Mutation.records)
            .startWith(Mutation.loading)
            .catchError({ .just(Mutation.failure(\$0)) })
    }
}

State introspection

Check branch state-feedback

To introspect State during handleAction it can be fed back into the loop:

func start(actions: Observable<Action>) -> Observable<State> {
    let states = BehaviorRelay(value: State.initial)
    return actions
        .withLatestFrom(states, resultSelector: { (\$0, \$1) })
        .flatMap(handleAction)
        .scan(states.value, accumulator: reduce)
        .do(onNext: states.accept)
}

The function handleAction will now take two parameters:

func handleAction(_ action: Action, state: State) -> Observable<Mutation> {
    ...
}

Further reading