How I use mithril

Posted on April 21, 2016

Please be aware this post was written when Mithril 0.2x was state of the art. Mithril 1.0 has built in support for streams and closure components; and enables writing in a declarative fashion. Many of the points I make here are still valid, but need to be read in the context of the mithril 0.2x API. The specific code samples and techniques used do not apply if you are using Mithril 1.0

I’ll leave this article here for posterity, and when I’ve got more experience with 1.0 I’ll post a follow up. In the interim please feel free to ask questions in the mithril chatroom.

Happy Coding - James

Mithril is a tiny library for writing interactive web applications. It’s so small it may seem difficult to know where to start.

Mithril is used by large in production apps, but its footprint is ~8kb. Yes it’s website could use a bit of spit and polish but it truly is the best MVC framework we have in Javascript right now.

I’ve previously written about other aspects of mithril, like m.prop and how cleanly it solves a lot of common problems I’ve experienced in other frameworks like Backbone.

In this post, I’d like to cover how I structure components and share data throughout the app.

Mithril gives us a lot of flexibility when defining components.
But I don’t write components using the traditional Mithril API. I am going to gradually introduce my approach so we can stay focused on each layer.

If you haven’t tried mithril, or you are still finding your feet, please, do not take this article as gospel. The approach I use, works for me, and my usecases. I recommend trying mithril out and get a feel for your own style.

Never use `this`

This isn’t just for javascript hipster reasons. Using this introduces bugs, data hiding issues, verbosity, and encourages the use of methods instead of functions.

Simply by using the revealing module pattern, we can see the majority of our app as data processing and reuse functions across components.

Some people may see the revealing module pattern as a hack because they are used to thinking of constructors and classes - but all a constructor is meant to do is return a state object, so we are actually making it more explicit.

//traditional mithril revealing module approach
var Example = {
  controller: function(){
    var count = m.prop(0)
    var checked = m.prop(false)
    
    return { count, checked }
  }
  
  view: function(state){
    return m('div', [
      m('input[type=number]', { 
        onchange: m.withAttr('value', state.count) 
      })
      ,m('p', 'The number is: '+ state.count() )
      ,m('input[type=checkbox]', { 
        checked: state.checked() 
        , onchange: m.withAttr('checked', state.checked) 
      })
    ])
  }
}

When I started with mithril I had separate namespaces for the model and the view-model. I don’t do that anymore. Dividing our data into separate namespaces for its origin and intent is noise. And often the line between view state and model state can blur over time. Though, I do think the concept of a view model is valuable.

Closure components

The traditional mithril component API is a struct with a controller constructor function, and a view function that accepts the controller as its first argument.

I think this is a great low level API, it is flexible and it allows for a variety of component patterns to work and still feel native.

However, I write my component as a function that returns a view function. All the model state lives within the closure.

This isn’t just because closures are cool, and this is lame. It’s because decoupling the view from the controller is only useful if you reuse base views across components. But in practice, that doesn’t happen often.

Using a closure also reduces library specific machinery. I don’t need to rely on the somewhat magical behaviour that mithril passes the controller instance to the view function. I can just reach the state directly using a standard javascript lanaguage behaviour.

You’ll notice our view also no longer needs to prefix each prop with state. So the view itself is a lot less ceremonial.

function Example(){

  var count = m.prop(0)
  var checked = m.prop(false)

  return function view(){
    return m('div', [
      m('input[type=number]', { 
        onchange: m.withAttr('value', count) 
      })
      ,m('p', 'The number is: '+ count() )
      ,m('input[type=checkbox]', { 
        checked: checked() 
        , onchange: m.withAttr('checked', checked) 
      })
    ])
  }
}

Of course, if this Component doesn’t obey the traditional mithril API, how do we mount it, or render it?

A simple utility.

function component(f){
   return { controller: f, view: function(view){ return view() } }
}

We can then plugin to the normal mithril machinery.
Here I’m injecting our component into a parent component in 3 different ways.


function ParentComponent(){
  
  var example = Example()
  
  return function(){
    return m('div', 
      
      // { controller: Example, view: function(view){ return view() } }
      component(Example)
      
      // view()
      ,example()
      
      // Allow passing in parameterized data
      ,m.component(component(Example), { data: [1,2,3] })
    )
  }
}

The above just shows what’s possible. I tend to export the traditional API as a module. And then other devs do not need to know that I used a closure component behind the scenes.

//file: components/example.js

// es6
export default component(Example)

// commonjs

module.exports = component(Example)

Flyd Streams everywhere

Flyd is a tiny library. It lets you pass values to subscribers and set a new value at any time. It goes with mithril like Peanut Butter and Jam - its API is almost identical to m.prop.

A quick introduction to flyd

In flyd we get and set and values the same way that we use props.

var a = stream(0)
a() //=> 0
a(1) 
a() //=> 1

But we can also subscribe to changes. And create dependent streams using map

var a = stream(0)
var b = a.map(multiply(2))

a(2)
b() //=> 4

It comes in handy in so many ways. Let’s take a look!

Replace prop’s with streams.

Everything still works

var f = require('flyd')

function Example(){

  var count = f.stream(0)
  var checked = f.stream(false)

  return function view(){
    return m('div', [
      m('input[type=number]', { 
        onchange: m.withAttr('value', count) 
      })
      ,m('p', 'The number is: '+ count() )
      ,m('input[type=checkbox]', { 
        checked: checked() 
        , onchange: m.withAttr('checked', checked) 
      })
    ])
  }
}

Not very exciting, but worth stating.

Subviews

Below we subscribe to changes to our state streams (count and checked).
We then declare view streams that will be regenerated whenever the state changes.

We are creating streams based on the value of our prop’s. They will always be up to date. And when we execute our sub streams as functions, we are not rerunning the view itself, it is cached.

It’s a shame mithril doesn’t know about this caching, so it will still do a diff against the DOM. But for complex view calculations, you can ensure, you are only generating that vdom when you need to.

It would also be quite simple to return { subtree: retain} on subsequent calls to a stream. Might be a nice experiment to try.


var f = require('flyd')

function Example(){

  var count = f.stream(0)
  var checked = f.stream(false)

  var p = count.map(function(n){
    return m('p', 'The number is: '+ count)
  })
  
  var numberInput = count.map(function(n){
    return m('input[type=number]', { 
      onchange: m.withAttr('value', count)
      value: n
    })
  })
  
  var checkbox = checked.map(function(value){
    m('input[type=checkbox]', { 
      checked: value
      , onchange: m.withAttr('checked', checked) 
    })
  })

  var view = f.combine(function(){
    return m('div', [
      numberInput()
      ,p()
      ,checkbox()
    ])
  }, [ checkbox, numberInput, p])

  return view
}

In a real world scenario, I probably wouldn’t define those particular subviews as streams. There is no real benefit. But it is nice that our component is a declarative depedency tree. It’s simple to see what state triggers changes to a particular view.

But I do use this for complex situations like occlusion culling, where you need to calculate how many items are on screen and their respective positions. It’s nice to know that calculcation is only happening when a dependency triggers it, and not on every redraw.

Cross component communication

The observer pattern suffers from “Call from where?”, and is too heavy. Simply sharing props, leads to managing when to redraw, and spaghetti code.

But streams are undirectional and clearly state their dependencies and flow. They are the perfect tool for sending data through multiple levels of components in your application.

Below is an example of a Sidebar. It animates on the x axis, and its position will affect the content pane to its right. We’ve got a SidebarManager that sets up the streams and passes it to the sub components Sidebar and ContentPane.

The thing I really like about streams, is that we are not listening to events.
We are declaring relationships. It’s not a matter of “when this changes, do that” but “This is that transformed by a function”

function SidebarManager(){

  var hidden = f.stream(false)
  var offset = hidden.map(function(hidden){
    return hidden ? 40 : 400
  })
  
  var sidebar = Sidebar({ hidden, offset })
  var content = ContentPane({ offset })
  
  return function(){
    return m('div', [
      sidebar()
      content()
    ])
  }
}

function Sidebar(world){
  var style = world.offset.map(function(x){
    return {
      left: -440+'px'
      transform: 'translateX('+x+'px)'
    }
  })
  
  var view = style.map(function(style){
    return m('div', { style },  ...)
  })
  
  return view
}

function ContentPane(world){
  var style = world.offset.map(function(x){
    return {
      transform: 'translateX('+x+'px)'
    }
  })
  
  var view = style.map(function(style){
    return m('div', { style }, ... )
  })
  
  return view
}

Easier to debug

flyd streams are easier to debug for 2 reasons.

You can subscribe to changes and log them
Every stream implements the toString interface so the chrome dev tools will show the current value.

Notice currentTime, pointer.scale and other fields are all streams. But we get to see the current value rendered inline!

So I hope I’ve shown that flyd+mithril is golden. Please try it out!

Opt out of Mithril when appropriate

Mithril makes it astonishingly convenient to jump in and out of virtual dom and direct dom. So if you need to just get in there and play with the dom directly, it’s straightforward.

There is no shame in this! If you want to getComputedStyle or access parentNode then all power to you.

In mithril it’s as simple as using the config function.

m('div', { config: 
  function(el, firstTime, context){
    if(firstTime){
      el.parentNode.style.backgroundColor = 'red'
    } else {
      //subsequent calls
    }
  } 
})

When I first saw that API, it scared me. Often I just want access to an element.

But, In plain old mithril, we could just use a prop as the config function, and because the el is the first argument, we’ll store the element in the prop.

//controller
var container = m.prop()

//view
m('div', { config: container })

And with flyd, we can actually do some work when that container arrives.

var container = f.stream()

var computedStyle = container.map(getComputedStyle)

// log the computed style every redraw
computedStyle.map(
  console.log.bind(console, 'computedStyle: ')
)

//view
m('div', config: container)

We’ve got this great way of interacting with the DOM now. But there are some things virtual doms are not good at, so it’s great its so simple to transfer back and forth between mithril and the “real world”.

Avoid the Model Layer

Models are closely coupled to views. The sooner we admit this, the more trouble we avoid. So let’s define them in the same file, and in most cases, lets just define them within in the controller.

In some cases it makes sense to put a model in a separate file, but do this after you’ve found you need shared state across routes.

You also, do not need object relationship management (ORM).

The whole get/set/save mentality is more complicated than just using fetch({ method: 'POST', body: json })

Here is the process:

Fetch it (fetch(...).then(data))
Render it data().map( i => m('div', i.data ))
Save it fetch({ method: PUT, body: JSON.stringify(data) })

This should be a util function, not an application layer and should all happen in close proximity to the view.

Avoid m.request

m.request is the XHR utility built into mithril. I personally don’t like utilities that do multiple things in multiple ways. m.request returns both a Promise and a m.prop, it also can control the redraw behaviour of your app and handle casting the data into a custom Model type.

The Promise implementation is custom, and not A+ compliant. m.request is like a swiss army knife, and if that appeals to you, use it by all means. But I prefer to compose separate tools that do 1 thing well. Its easier to debug, and reason about later.

Here’s an example of using m.request

You can initialize your props immediately and mithril will automatically redraw for you when the request completes.


var users = m.request({ method: 'GET' , url: '/users' })

users() //=> undefined
users.then(function(serverResponse){
  serverResponse == users()
})

I would just use fetch.

var users = m.prop([]) // A default value so we can draw immediately.
fetch('/users') //GET by default
  .then( r => r.json() )
  .then(users) // save to prop
  .then(m.redraw)

I personally like that I am specifically setting the data to the prop and manually triggering a redraw. It encourages us to draw immediately with an empty dataset, instead of loading. It’s also simple to add some logs, or other transformations within the promise chain.

Conclusion

In this post I’ve demonstrated how I like to use mithril. But the great thing about mithril is that you can choose your own way to approach these problems. Mithril has smart design assumptions built in, but it also is flexible enough to let you opt out at any time.

The upcoming version of mithril will make this even more true. Mithril’s router, rendering, etc are all separate modules that you can opt out of at will.

If you found this article interesting, I recommend hanging out in the mithril chat.

Thank you for reading. If you’d like to reach out you can get in touch on Mastodon, BlueSky.

How I use mithril

Posted on April 21, 2016

Never use `this`

Closure components

Flyd Streams everywhere

A quick introduction to flyd

Replace prop’s with streams.

Subviews

Cross component communication

Easier to debug

Opt out of Mithril when appropriate

Avoid the Model Layer

Avoid m.request

Conclusion

Recent articles

Previous articles

Feeds

Never use this

Closure components

Flyd Streams everywhere

A quick introduction to flyd

Replace prop’s with streams.

Subviews

Cross component communication

Easier to debug

Opt out of Mithril when appropriate

Avoid the Model Layer

Avoid m.request

Conclusion

Recent articles

Previous articles

Feeds

Never use `this`