This project was created related to this article. Please have a read to see what it is all about.
!NOTE This project is not officially released, as it would require more testing and a maintainer.
With the success of Immer there is no doubt that developers has no problems writing pure code in an impure API. The mutation API of JavaScript is straight forward and expressive, but the default result is impure, going against the immutable model favoured by React. With immer-store we allow Immer to take even more control and basically gets rid of reducers, dispatching and action creators.
Instead of having to write this:
// action creator
const receiveProducts = (products) => ({
type: RECEIVE_PRODUCTS,
products
})
// thunk
const getProducts = () => async (dispatch) => {
dispatch(receiveProducts(await api.getProducts()))
}
// reducer
const productsById = produce((draft, action) => {
switch (action.type) {
case RECEIVE_PRODUCTS:
action.products.forEach((product) => {
draft[product.id] = product
})
return
}
})
You can just write this:
const getProducts = async ({ state }, payload) => {
const products = await api.getProducts()
products.forEach((product) => {
state.products[product.id] = product
})
}
Everything is still immutable.
immer-store takes inspiration from the "api"-less API of overmindjs. The codebase is rather small and commented, so you can take a dive into that. A quick summary though:
- With a combination of chosen API and Proxies immer-store exposes a state object to your actions that produces Immer drafts under the hood. The concept of a draft is completely hidden from you. You only think actions and state
- It supports changing state asynchronously in your actions
- Because immer-store exposes an action API it also has access to the execution of the action and access to state. That means it is able to batch up mutations and notify components at optimal times to render
- Instead of using selectors to expose state to components the useState hook tracks automatically what state you access and subscribes to it. That means there is no value comparison in every single hook on every mutation, but immer-store tells specifically what components needs to update related to matching batched set of mutations
- immer-store has a concept of effects which is a simple injection mechanism allowing you to separate generic code from your application logic. It also simplifies testing
- The library is written in Typescript and has excellent support for typing with minimal effort
import React from 'react'
import { render } from 'react-dom'
import { createStore, Provider, useState, useActions } from 'immer-store'
const store = createStore({
state: {
title: ''
},
actions: {
changeTitle: ({ state }, title) => {
state.title = title
}
}
})
function App() {
const state = useState()
const actions = useActions()
return (
<div>
<h1>{state.title}</h1>
<button onClick={() => actions.changeTitle('New Title')}>
Change title
</button>
</div>
)
}
render(
<Provider store={store}>
<App />
</Provider>,
document.querySelector('#app')
)
As your application grows you want to separate things a bit more.
// store/index.js
import * as home from '../pages/home'
import * as issues from '../pages/issues'
export const config = {
state: {
home: home.state,
issues: issues.state
},
actions: {
home: home.actions,
issues: issues.actions
}
}
// index.js
import { createStore } from 'immer-store'
import { config } from './store'
const store = createStore(config)
This structure ensures that you can split up your state and actions into different domains. It also separates the definition of your application from the instantiation of it. That means you can easily reuse the definition multiple times for testing purposes or server side rendering.
Instead of importing 3rd party libraries and writing code with browser side effects etc. immer-store allows you to separate it from your application logic in the actions. This creates a cleaner codebase and you get several other benefits:
- All the code in your actions will be domain specific, no low level generic APIs
- Your actions will have less code and you avoid leaking out things like URLs, types etc.
- You decouple the underlying tool from its usage, meaning that you can replace it at any time without changing your application logic
- You can more easily expand the functionality of an effect. For example you want to introduce caching or a base URL to an HTTP effect
- You can lazy-load the effect, reducing the initial payload of the app
So you decide what you application needs and then an effect will provide it:
export const config = {
state: {...},
actions: {...},
effects: {
storage: {
get: (key) => {
const value = localStorage.getItem(key)
return typeof value === 'string' ? JSON.parse(value) : value
},
set: (key, value) => {
localStorage.setItem(key, JSON.stringify(value))
}
},
http: {
get: async (url) => {
const response = fetch(url)
return response.json()
}
}
}
}
Note! When using Typescript you have to define your effects as functions (as shown above), not methods. This is common convention, though pointing out it being necessary for typing to work.
You got typing straight out of the box with immer-store.
const store = createStore({
state: {
title: ''
},
actions: {
changeTitle: ({ state }, title: string) => {
state.title = title
}
}
})
store.state.foo // string
store.actions.changeTitle // (title: string) => void
As you scale up you want to do this:
import { IAction } from 'immer-store'
import * as home from '../pages/home'
import * as issues from '../pages/issues'
const state = {
home: home.state,
issues: issues.state
}
const actions = {
home: home.actions,
issues: issues.actions
}
const effects = {}
export const config = {
state,
actions,
effects
}
// This type can be used within the pages to define actions
export interface Action<Payload = void>
extends IAction<Payload, typeof state, typeof effects> {}
And then in for example pages/home/
/*
./index.ts
*/
export { state } from './state'
export * as actions from './actions'
export const Home: React.FC = () => {}
/*
./state.ts
*/
type State {
title: string
}
export const state: State = {
title: ''
}
/*
./actions.ts
*/
import { Action } from '../store'
export const changeTitle: Action<string> = ({ state }, title) => {
state.home.title = title
}
Since immer-store is tracking what components actually use it has a pretty nice optimization especially useful in lists. You can target a piece of state and then ensure that the component only renders again if that specific piece of state changes.
function Todo({ id }: { id: string }) {
const todo = useState((state) => state.posts[id])
const actions = useActions()
return (
<li>
<h4>{todo.title}</h4>
<input
checked={todo.completed}
onChange={() => actions.toggleTodo(id)}
/> {todo.description}
</li>
)
}
Since we target the todo itself this component will only reconcile again if any of its accessed properties change, for example the completed state. If you were to rather pass the todo as a prop also the component passing the todo would need to reconcile on any change of the todo. This is an optimization that only makes sense in big lists where individual state items in the list change. Normally when you want to target state you can just:
const SomeComponent: React.FC_ = () => {
const { home, issues } = useState()
return (
...
)
}
immer-store includes the reselect library for cached computed state. immer-store exposes a createComputed and useCOmputed hook that allows you to compute state:
const getTitle = (state) => state.title
const titleUpperCase = createComputed(
[getTitle],
(title) => title.toUpperCase()
)
const SomeComponent: React.FC_ = () => {
const title = useComputed(titleUpperCase)
return (
...
)
}
The computed concept is just a tiny wrapper around reselect to manage updates.
immer-store knows a lot about your application:
- It knows what actions are changing what state
- It knows what state all components are looking at
- It knows which component renders related to what state change
It is possible to increase the insight even more to get a similar development tool experience as overmindjs.
- Immer provides the concept of snapshots and patching. This is available, just not implemented. The question is what kind of API you want. An example would be to expose a default effect which allows you to start tracking changes to a specific state path with a limit of number of changes. Then you could takes these snapshots and patch them back in