choo

🚂🚋🚋🚋🚋🚋 - The little framework that could.

A framework for creating sturdy web applications. Built on years of industry experience it distills the essence of functional architectures into a productive package.

• Features
• Demos
• Usage
• Concepts
  • Models
  • Actions
  • Effects
    • HTTP
  • Subscriptions
    • server sent events
    • keyboard
    • websockets
  • Router
  • Views
    • forms
    • links
    • styles
• Rendering in Node
  • Rehydration
• API
• FAQ
• Installation
• See Also
• License

Features

• minimal size: weighing 7kb, choo is a tiny little framework
• single state: immutable single state helps reason about changes
• small api: with only 6 methods, there's not a lot to learn
• minimal tooling: built for the cutting edge browserify compiler
• transparent side effects: using effects and subscriptions brings clarity to IO
• omakase: composed out of a balanced selection of open source packages
• idempotent: renders seemlessly in both Node and browsers
• very cute: choo choo!

Demos

• Input example (@examples directory)
• HTTP effects example (@examples directory)
• Mailbox routing example (@examples directory)
• TodoMVC (github)

Usage

const choo = require('choo')

const app = choo()
app.model({
  namespace: 'input',
  state: {
    title: 'my demo app'
  },
  reducers: {
    update: (action, state) => ({ title: action.payload })
  },
  effects: {
    update: (action, state, send) => (document.title = action.payload)
  }
})

const mainView = (params, state, send) => {
  return choo.view`
    <main class="app">
      <h1>${state.input.title}</h1>
      <label>Set the title</label>
      <input
        type="text"
        placeholder=${state.input.title}
        oninput=${(e) => send('input:update', { payload: e.target.value })}>
    </main>
  `
}

app.router((route) => [
  route('/', mainView)
])

const tree = app.start()
document.body.appendChild(tree)

Concepts

choo cleanly structures internal data flow, so that all pieces of logic can be combined into a nice, cohesive machine. Internally all logic lives within models that contain several properties. subscriptions are read-only streams of data. effects react to changes, perform an action and can then post the results. reducers take data, modify it, and update the internal state.

Communication of data is done using objects called actions. Each action has any number of properties for data, and a unique type that can trigger properties on the models.

When a reducer modifies state, the router is called, which in turn calls views. views take state and return DOM nodes which are then efficiently rendered on the screen.

In turn when the views are rendered, the user can interact with elements by clicking on them, triggering actions which then flow back into the application logic. This is the unidirectional architecture of choo.

 ┌───────────────────────────┐     ┌────────┐
 │    ┌─────────────────┐    │     │  User  │
 ├────│  Subscriptions  │    │     └────────┘
 │    ├─────────────────┤    │          │
 └────│     Effects     │◀───┤          ▼
      ├─────────────────┤  Actions ┌────────┐
      │    Reducers     │◀───┴─────│  DOM   │
    Models──────────────┘          └────────┘
               │                        ▲
             State                   DOM│tree
               ▼                        │
          ┌────────┐               ┌────────┐
          │ Router │─────State ───▶│ Views  │
          └────────┘               └────────┘

• user: 🙆
• DOM: the Document Object Model is what is currently displayed in your browser
• actions: a named event with optional properties attached. Used to call effects and reducers that have been registered in models
• model: optionally namespaced object containing subscriptions, effects, reducers and initial state
• subscriptions: read-only data sources that emit actions
• effects: asynchronous functions that emit an action when done
• reducers: synchronous functions that modify state
• state: a single object that contains all the values used in your application
• router: determines which view to render
• views: take state and returns a new DOM tree that is rendered in the browser

Models

models are objects that contain initial state, subscriptions, effects and reducers. They're generally grouped around a theme (or domain, if you like). To provide some sturdiness to your models, they can either be namespaced or not. Namespacing means that only state within the model can be accessed. Models can still trigger actions on other models, though it's recommeded to keep that to a minimum.

So say we have a todos namespace, an add reducer and a todos model. Outside the model they're called by send('todos:add') and state.todos.todos. Inside the namespaced model they're called by send('todos:add') and state.todos. An example namespaced model:

const app = choo()
app.model({
  namespace: 'todos',
  model: { todos: [] },
  reducers: {
    add: (state, action) => ({ todos: state.todos.concat(action.payload) })
  }
})

In most cases using namespaces is beneficial, as having clear boundries makes it easier to follow logic. But sometimes you need to call actions that operate over multiple domains (such as a "logout" action), or have a subscription that might trigger multiple reducers (such as a websocket that calls a different action based on the incoming data).

In these cases you probably want to have a model that doesn't use namespaces, and has access to the full application state. Try and keep the logic in these models to a minimum, and declare as few reducers as possible. That way the bulk of your logic will safely shielded, with only a few points touching every part of your application.

Effects

Side effects are done through effects declared in app.model(). Unlike reducers they cannot modify the state by returning objects, but get a callback passed which is used to emit actions to handle results. Use effects every time you don't need to modify the state object directly, but wish to respond to an action.

A typical effect flow looks like:

1. An action is received
2. An effect is triggered
3. The effect performs an async call
4. When the async call is done, either a success or error action is emitted
5. A reducer catches the action and updates the state

HTTP

choo ships with a built-in http module that weighs only 2.4kb:

const http = require('choo/http')
const choo = require('choo')
const app = choo()

app.model({
  effects: {
    'app:error': (state, event_ => console.error(`error: ${event.payload}`)),
    'app:print': (state, event) => console.log(`http: ${event.payload}`),
    'http:get_json': getJson,
    'http:post_json': postJson,
    'http:delete': httpDelete
  }
})

function getJson (state, action, send) {
  http.get('/my-endpoint', { json: true }, function (err, res, body) {
    if (err) return send('app:error', { payload: err.message })
    if (res.statusCode !== 200 || !body) {
      return send('app:error', { payload:'something went wrong' })
    }
    send('app:print', { payload: body })
  })
}

function postJson (state, action, send) {
  const body = { foo: 'bar' }
  http.post('/my-endpoint', { json: body }, function (err, res, body) {
    if (err) return send('app:error', { payload: err.message })
    if (res.statusCode !== 200 || !body) {
      return send('app:error', { payload:'something went wrong' })
    }
    send('app:print', { payload: body })
  })
}

function httpDelete (state, action, send) {
  const body = { foo: 'bar' }
  http.post('/my-endpoint', { json: body }, function (err, res, body) {
    if (err) return send('app:error', { payload: err.message })
    if (res.statusCode !== 200) {
      return send('app:error', { payload:'something went wrong' })
    }
  })
}

Note that http only runs in the browser to prevent accidental requests when rendering in Node. For more details view the raynos/xhr documentation.

Subscriptions

Subscriptions are a way of receiving data from a source. For example when listening for events from a server using SSE or Websockets for a chat app, or when catching keyboard input for a videogame.

An example subscription that logs "dog?" every second:

const app = choo()
choo.model({
  subscriptions: [
    (send) => setTimeout(() => send('app:print', { payload: 'dog?' }), 1000)
  ],
  effects: {
    'app:print': (state, action) => console.log(action.payload)
  }
})

Server Sent Events (SSE)

Server Sent Events (SSE) allow servers to push data to the browser. They're the unidirectional cousin of websockets and compliment HTTP brilliantly. To enable SSE, create a new EventSource, point it at a local uri (generally /sse) and setup a subscription:

const stream = new document.EventSource('/sse')

app.model({
  subscriptions: [
    function (send) {
      stream.onerror = (e) => send('app:error', { payload: JSON.stringify(e) })
      stream.onmessage = (e) => send('app:print', { payload: e.data })
    }
  ],
  effects: {
    'sse:close': () => stream.close()
    'app:error': (state, event_ => console.error(`error: ${event.payload}`)),
    'app:print': (state, event) => console.log(`sse: ${event.payload}`)
  }
})

This code does not handle reconnects, server timeouts, exponential backoff and queueing data. You might want to use a package from npm or write your own if you're building something for production.

Keyboard

Most browsers have basic support for keyboard events. To capture keyboard events, setup a subscription:

app.model({
  subscriptions: [
    function (send) {
      keyboard.onkeypress = (e) => send('app:print', { payload: e.keyCode })
    }
  ],
  effects: {
    'app:print': (state, event) => console.log(`pressed key: ${event.payload}`)
  }
})

WebSockets

WebSockets allow for bidirectional communication between servers and browsers:

const socket = new document.WebSocket('ws://localhost:8081')

app.model({
  subscriptions: [
    function (send) {
      socket.onerror = (e) => send('app:error', { payload: JSON.stringify(e) })
      socket.onmessage = (e) => send('app:print', { payload: e.data })
    }
  ],
  effects: {
    'ws:close': () => socket.close(),
    'ws:send': (state, event) => socket.send(JSON.stringify(event.payload)),
    'app:error': (state, event_ => console.error(`error: ${event.payload}`)),
    'app:print': (state, event) => console.log(`ws: ${event.payload}`)
  }
})

This code does not handle reconnects, server timeouts, exponential backoff and queueing data. You might want to use a package from npm or write your own if you're building something for production.

Router

[docs wip]

Views

[docs wip]

forms

[docs wip]

links

[docs wip]

styles

[docs wip]

Rendering in Node

Sometimes it's necessary to render code inside of Node; for serving hyper fast first requests, testing or other purposes. Applications that are capable of being rendered in both Node and the browser are called isomorphic.

Rendering in Node is slightly different than in the browser. First off, to maintain performance all calls to subscriptions, effects, and reducers are disabled. That means you need to know what the state of your application is going to be before you render it - no cheating!

Secondly, the send() method inside router and view has been disabled. If you call it your program will crash. Disabling all these things means that your program will render O(n), which is super neat. Off to 10.000 QPS we go!

To render in Node call the .toString() method instead of .start(). The first argument is the path that should be rendered, the second is the state:

const http = require('http')
const client = require('./client')  // path to client entry point
http.createServer(function (req, res) {
  const html = client.toString('/', { message: 'hello server!' })
  res.setHeader('Content-Type', 'text/html; charset=utf-8')
  res.end(html)
})

In order to make our choo app call app.start() in the browser and be require()-able in Node, we check if module.parent exists:

const choo = require('choo')
const app = choo()

app.router((route) => [
  route('/', (params, state, send) => choo.view`
    <h1>${state.message}</h1>
  `)
])

if (module.parent) module.exports = app
else document.body.appendChild(app.start())

Rehydration

Now that your application is succesfully rendering in Node, the next step would be to make it load a JavaScript bundle once has loaded the HTML. To do this we will use a technique called rehydration.

Rehydration is when you take the static, server-rendered version of your application (static HTML, dehydrated because it has no logic) and rehydrate it by booting up the JS and attaching event handlers on the DOM to make it dynamic again. It's like restoring flavor to cup noodles by adding hot water.

Because we're using something called morphdom under the hood, all we need is point at an id at the root of the application. The syntax for this is slightly different from what we've seen so far, because we're updating a dehydrated DOM nodes to make them dynamic, rather than a new DOM tree and attaching it to the DOM.

const choo = require('choo')
const app = choo()

app.router((route) => [
  route('/', (params, state, send) => choo.view`
    <h1 id="app-root">${state.message}</h1>
  `)
])

if (module.parent) module.exports = app
else app.start('#app-root'))

When the JS is booted on top of the dehydrated application, it will look for the #app-root id and load on top of it. You can choose any name you like for the id, but make sure it's the same on every possible top level DOM node, or else things might break. Furthermore to ensure things go smoothly, try and keep the initial state identical on both the server and the client.

And that's it! If you want to go down the route of mad performance, consider make all first request static and caching them using something like bl, nginx, varnish or a global CDN.

API

app = choo()

Create a new choo app

app.model(obj)

Create a new model. Models modify data and perform IO. Takes the following arguments:

• namespace: optional namespace that prefixes the keys in state, reducers and effects. Also limits actions called by send() to in-namespace only.
• state: object. Key value store of initial values
• reducers: object. Syncronous functions that modify state. Each function has a signature of (action, state)
• effects: object. Asyncronous functions that perform IO. Each function has a signature of (action, state, send) where send is a reference to app.send()

choo.view`html`

Tagged template string HTML builder. See yo-yo for full documentation. Views should be passed to app.router()

app.router(params, state, send)

Creates a new router. See sheet-router for full documentation. Registered views have a signature of (params, state, send), where params is URI partials.

html = app.toString(route, state)

Render the application to a string of HTML. Useful for rendering on the server. First argument is a path that's passed to the router. Second argument is the state object. When calling .toString() instead of .start(), all calls to send() are disabled, and subscriptions, effects and reducers aren't loaded. See rendering in Node for an in-depth guide.

tree = app.start(rootId?, opts)

Start the application. Returns a tree of DOM nodes that can be mounted using document.body.appendChild(). If a valid id selector is passed in as the first argument, the tree will diff against the selected node rather than be returned. This is useful for rehydration. Opts can contain the following values:

• opts.history: default: true. Enable a subscription to the browser history API. e.g. updates the internal state.location state whenever the browser "forward" and "backward" buttons are pressed.
• opts.href: default: true. Handle all relative <a href="<location>"></a> clicks and update internal state.location accordingly.

FAQ

Why did you build this?

choo is nothing but a formalization of how I've been building my applications for the past year. I originally used virtual-dom with virtual-app and wayfarer where now it's yo-yo with send-action and sheet-router. The main benefit of using choo over these technologies separately is that it becomes easier for teams to pick up and gather around. The code base for choo itself is super petite (~200 LOC) and mostly acts to enforce structure around some excellent npm packages. This is my take on modular frameworks; I hope you'll find it pleasant.

Why is it called choo?

Because I thought it sounded cute. All these programs talk about being "performant", "rigid", "robust" - I like programming to be light, fun and non-scary. choo embraces that.

Also imagine telling some business people you chose to rewrite something critical to the company using the choo framework. 🚂🚋🚋🚋

Why is it a framework, and not a library?

I love small libraries that do one thing well, but when working in a team, having an undocumented combination of packages often isn't great. choo() is a small set of packages that work well together, wrapped in an an architectural pattern. This means you get all the benefits of small packages, but get to be productive right from the start.

How does choo compare to X?

Ah, so this is where I get to rant. choo (chugga-chugga-chugga-choo-choo!) was built because other options didn't quite cut it for me, so instead of presenting some faux-objective chart with skewed benchmarks and checklists I'll give you my opinions directly instead. Ready? Here goes:

• react: react is kind of big (155kb was it?), has a lot of new, odd words and does weird things with versioning. They also like classes a lot, and enforce a lot of abstractions. It also encourages the use of JSX and babel which break JavaScript, The Language™. And all that without even making clear how code should flow, which is bad in a team setting. I don't like complicated things and in my view react is one of them. react is not for me.
• mithril: never used it, never will. I didn't like the API, but if you like it maybe it's worth a shot - the API seems small enough. I wouldn't know how pleasant it is past face value.
• preact: a pretty cool idea; seems to fix most of what is wrong with react - except what is broken by design (the API). It also doesn't fix the large dependencies react seems to use (e.g. react-router and friends). If react is your jam, and you will not budge, sitting at 3kb this is probably a welcome gift.
• angular: definitely not for me. I like small things with a clear mental model; angular doesn't tick any box in my book of nice things.
• angular2: I'm not sure what's exactly changed, but I know the addition of TypeScript and RxJS definitely hasn't made things simpler. Last I checked it was ~200kb in size before including some monstrous extra deps. I guess angular and I will just never get along.
• mercury: ah, mercury is an interesting one. It seemed like a brilliant idea until I started using it - the abstractions felt heavy, and it took team members a long time to pick up. In the end I think using mercury helped greatly in getting choo where it is now.
• deku: deku is fun. I even contributed a bit in the early days. It could probably best be described as "a functional version of react". The dependence on JSX isn't great, but give it a shot if you think it looks neat.

Which packages was choo built on?

• views: yo-yo
• models: send-action, xtend
• routes: sheet-router
• http: xhr

Does choo use a virtual-dom?

choo uses morphdom, which diffs real DOM nodes instead of virtual nodes. It turns out that browsers are actually ridiculously good at dealing with DOM nodes, and it has the added benefit of working with any library that produces valid DOM nodes. So to put a long answer short: we're using something even better.

What packages do you recommend to pair with choo?

• tachyons - functional CSS for humans
• sheetify - modular CSS bundler for browserify
• pull-stream - minimal streams

How can I optimize choo?

choo really shines when coupled with browserify transforms. They can do things like reduce file size, prune dependencies and clean up boilerplate code. Consider running some of the following:

• unassertify - remove assert() statements which reduces file size. Use as a --global transform
• varify - replace const with var statements. Use as a --global transform
• uglifyify - minify your code using UglifyJS2. Use as a --global transform
• bulkify - transform inline bulk-require calls into statically resolvable require maps

Hey, doesn't this look a lot like Elm?

Yup, it's greatly inspired by the elm architecture. But contrary to elm, choo doesn't introduce a completely new language to build web applications.

Is it production ready?

Sure.

Installation

$ npm install choo

See Also

• budo - quick prototyping tool for browserify
• stack.gl - open software ecosystem for WebGL

License

MIT