02-stores.md

State in FRP?

In a pure functional language, there is no state. So how the heck can we get anything done?

There are at least two strategies in FRP:

1. use an "accumulator" variable to pass store around in function arguments
2. use a global component, say ApplicationState.

Strategy 1: Use the Scan Luke

This approach turns out to be the most similar to Flux.

Basically, in FP, you get around not being able to store state by passing around state (aka accumulating state) in function arguments. For example, the code below shows the factorial function, FP style. The function argument acc accumulates the result of successive function calls.

factorial(n, acc=1) {
  if (n >= 1) return acc
  return factorial(n-1, n*acc)
}

We are already saw this technique in reduce and scan where we "stored state" inside scan's accumulator variable. In ReactJS terminology, a store can be thought of as a model + controller (well, sort of). An FRP store is just a scan function that stores state in its accumulator variable.

For example:

//myObservable = 0, 1, 2, 3, ...
const store = myObservable.scan((values, cur) => values.concat([cur]), [])

store.onValue(x => console.log(x))
// [0]
// [0, 1]
// [0, 1, 2]
// [0, 1, 2, 3]
// etc

We just need a way to pass messages to myObservable. In FRP, we can use a bus. A bus is a special stream that can receive messages.

In BaconJS, creating a bus is a simple as:

const myObservable = new Bacon.Bus()

Messages can be anything. In the example above:

myObservable.push(0)
myObservable.push(1)
//...

Note: push values to the bus, not the stream. In the example above, store.push(0) throws an error because it is a stream, not a bus.

Variation 1: The Straightforward Approach

A straightforward implementation of a "todo" store is:

// TodoStore.js
import AppDispatcher from '...' //the bus

export function updateTodo(todos, todo) {

  todos[todo.id] = todo //replace existing todo with latest
  return todos
}

export function todoStore(todos, action) {

  switch(action.type) {
    case 'update':
      return updateTodo(todos, action.payload)
    default:
      return todos;
  }
}

const todoStoreStream = AppDispatcher.scan(scanner, {})

export default todoStoreStream

An update action looks like this:

AppDispatcher.push({
  type: 'update',
  payload: {id: 1, content: 'hello world'}
})

Note: the architecture shown above is almost a version of flux called Redux. The difference between this strategy and Redux is that the latter uses a global store, whereas this solution uses many stores. The point is that Flux and Redux both follow from FRP principles.

In Redux terminology, the store-state functions (in this case, updateTodo) are called reducers.

Variation 2: Action Streams

We define action streams to be streams that fire whenever an action occurs. Action streams are useful in a real-world app because they can be used for things like showing Toast messages in response to app events.

Another variation of the TodoStore is to define it in terms of action streams. For example:

// TodoStore.js
import AppDispatcher from '...'
import Kefir from 'kefir'

// note that we return a higher-order function
export function updateTodo(todo) {
  return todos => {
    todos[todo.id] = todo
    return todos
  }
}

export function createTodo(todo) {
  return todos => {
    todo.id = //create an id
    todos[todo.id] = todo
    return todos
  }
}

// This stream fires whenever an update action occurs.
export const updateActionStream = AppDispatcher
  // filter works just like Array.filter.
  // It fires only when the message is of type === 'update'
  .filter(x => x.type === 'update')
  .map(x => updateTodo(x.payload))

// This stream fires whenever a create action occurs.
export const createActionStream = AppDispatcher
  .filter(x => x.type === 'create')
  .map(x => createTodo(x.payload))

// In FRP libraries, "merge" fires anytime either stream fires
// So when an update action happens, the merge fires,
// returning the update function.
// When a create action happens, the merge fires,
// returning the create function
const todoStoreStream = Kefir
  .merge([createActionStream, updateActionStream])
  .scan((todos, actionFunction) => actionFunction(todos), {})

export default todoStoreStream

Actions work as before:

AppDispatcher.push({
  type: 'create',
  payload: {content: 'hello world'}
})

//But we can now listen to create actions...or at least that's the idea.
//(See below for details.)
createActionStream.onValue(() => console.log('todo created'))

The Shootout: The Straightforward Approach vs Action Streams

The Straightforward Approach is simple to write and simple to understand. Not surprisingly, it's easily unit tested (since everything is a function). The downside is boilerplate---you'll need to create something that looks like Action Streams anyway.

The main problem with Action Streams is that the action streams aren't actually that useful. Yes, they fire whenever an action happens but getting detailed information is hard. For example, look at createTodo--- how do you get the id of the newly created todo? Call the function it returns? If you look closely, that'll return a new id each time it's called. We would need to make the function idempotent (i.e., return the same value each time it's called), and this can be error prone.

//idempotent createTodo - this can be used to get the id of the newly created todo
export function createTodo(todo) {
    const id = //create id
    return todos => {
      todo.id = id
      todos[id] = todo
      return todos
  }
}

Winner: The Straightforward Approach

A Note About Observers (aka Buses are an Anti-Pattern)

The createActionStream listener defined in the example won't work. The reason is because the create event happens before the listener definition, so there's nothing to listen to. (Or, if you use a Property, the listener will get the old value). This is often cited as a reason why buses are bad.

We will examine this issue in the next article.

Strategy 2: Global State aka Dude, Where are My Stores?

Suppose we store global state in a single place called AppState. It would be really dumb if we put all (unrelated) logic in one place. Therefore, it still makes sense for stores to exist. However, because stores aren’t allowed to store/hold/accumulate state, stores become nothing but pure functions (containing domain-specific logic). Essentially, we just stumbled on the full Redux architecture.

// TodoStore.js aka TodoReducer (in Redux parlance)
import AppDispatcher from '...'

export function createTodo(todos, todo) {
  todo.id = //create an id
  todos[todo.id] = todo
  return todos
}

export function todos(todos, action) {
  switch(action.type) {
    case 'create':
      return createTodo(todos, action.payload)
    default:
      return todos;
  }
}

// AppState.js
import AppDispatcher from '...'
import TodoStore from './TodoStore'
// assuming UserStore is similar to TodoStore...
import UserStore from './UserStore'

function scanner(state, action) {
  return {
    todos: TodoStore.todos(state.todos, action),
    users: UserStore.users(state.users, action)
  }
}

const appStateStream = AppDispatcher.scan(scanner, {})
const todosStream = appState.map(appState => appState.todos)
const usersStream = appState.map(appState => appState.users)

export default {
  appStateStream,
  todosStream,
  usersStream
}

Shootout: The Straightforward Approach vs Global State Strategy

The Global State Strategy is the Straightforward Approach with all the scan functions centralized in one location. Yes, it's not too hard to write a helper function to register the scan functions for us and create AppState. However, it is still code we need to write. In contrast, the Straightforward Approach works out of the box using the FRP library.

The Global State Strategy shines when it comes time for time travel. The reason is because AppState stores the entire app state. To get the same functionality in the Straightforward Approach, we would need to write a helper function to register all stores and combine them into a single stream. The jury is still out on which helper function is simpler.

Winner: Tie for now. (Although I suspect that combining the store streams in the Straightforward Approach is relatively easy to do)