Introduction to dcl

By  on  

I'm incredibly honored to have Eugene Lazutkin author for David Walsh Blog. Eugene has written much of the vector graphic code for the Dojo Toolkit's dojox/gfx (and subsequent charting and drawing resources) library, a library I consider to be mind-blowingly awesome. Eugene chose to write about dcl, an ultra-flexible, tiny OOP JS library.

dcl is a minimalistic yet complete JavaScript package for node.js and modern browsers. It implements OOP with mixins + AOP at both "class" and object level, and works in strict and non-strict modes.

The simplest way to learn something is to dive right in. Let's implement a simple widget based on reactive templating: when we change parameters of a widget, they are immeditely reflected in a web page.

Assuming that we run our code using AMD format in browser, our "code shell" will look like that:

  ["dcl", "dcl/bases/Mixer", "dcl/mixins/Cleanup", "dcl/advices/memoize"],
  function(dcl, Mixer, Cleanup, memoize){
    // our code goes here

As the first step let's code our data model:

var Data = dcl(Mixer, {
  declaredClass: "Data",
  updateData: function(data){
    dcl.mix(this, data);

We derived our class using single inheritance from Mixer, which comes with dcl. Mixer is a very simple base. All it does is it copies properties of the first constructor argument to an instance.

Obviously in this simple example we could just call updateData() from our constructor, but let's assume that a constructor and an updater can do (slightly) different things and we want to keep them separately.

declaredClass is completely optional, yet recommended to be specified (any unique human-readable name is fine), because it is used by debugging helpers included with `dcl`.

Now let's code our nano-sized template engine, which substitutes strings like this: ${abc} with properties taken directly from an instance ( in this case). Something like that:

var Template = dcl(null, {
  declaredClass: "Template",
  render: function(templateName){
    var self = this;
    return this[templateName].replace(/\$\{([^\}]+)\}/g, function(_, prop){
      return self[prop];

We specify what template to use by name, which is a property name on an object instance, and it fills out a template string using properties specified on an object.

This is another demonstration of single inheritance: our Template is based on a plain vanilla Object, like any JavaScript's object, which is indicated by using null as a base.

What else do we need? We need a way to manage our DOM node:

var Node = dcl([Mixer, Cleanup], {
  show: function(text){
      this.node.innerHTML = text;
  destroy: function(){
      this.node.innerHTML = "";

The code above provides a way to show some HTML, and clears out its presentation when we destroy() a widget.

It uses two bases: already mentioned Mixer is used to get a property in during initialization (node in this case), and Cleanup, which again comes with dcl. The latter chains all destroy() methods together and provides a simple foundation for clean up management, so all resources can be properly disposed of.

What we did up to this point is we came up with very small manageable orthogonal components, which reflect different sides of our widget, and can be combined together in different configurations. Let's put them all together now:

var NameWidget0 = dcl([Data, Template, Node], {
  declaredClass: "NameWidget0",
  template: "Hello, ${firstName} ${lastName}!"

var x = new NameWidget0({
  node:      document.getElementById("name"),
  firstName: "Bob",
  lastName:  "Smith"
});"template")); // Hello, Bob Smith!
x.updateData({firstName: "Jill"});"template")); // Hello, Jill Smith!

It works, but it is not very coherent, and way too verbose. Don't worry, we will fix it soon.

Some readers probably noticed that we have three bases now: Data, Template, and Node, and two of them (Data, and Node) are based on Mixer. How does it work? It works fine, because underneath dcl uses C3 superclass linearization algorithm (the same one used by Python), which removes duplicates, and sorts bases to ensure that their requested order is correct. In this case a single copy of Mixin should go before both Data and Node. Read more on that topic in dcl() documentation.

Now let's address deficiencies of our implementation #0:

  • As soon as a widget is constructed, we should show text.
  • As soon as data is updated, we should show text.

Both requirements are simple and seem to call for good old-fashioned supercalls:

var NameWidget1 = dcl([Data, Template, Node], {
  declaredClass: "NameWidget1",
  template: "Hello, ${firstName} ${lastName}!",
  constructor: function(){
  updateData: dcl.superCall(function(sup){
    return function(){
      sup.apply(this, arguments);
  showData: function(){
    var text = this.render("template");;

var x = new NameWidget1({
  node:      document.getElementById("name"),
  firstName: "Bob",
  lastName:  "Smith"
// Hello, Bob Smith!

x.updateData({firstName: "Jill"}); // Hello, Jill Smith!

Much better!

Let's take a look at two new things: constructor and a supercall. Both are supposed to be supercalls, yet look differently. For example, constructor doesn't call its super method. Why? Because dcl chains constructors automatically.

updateData() is straightforward: it calls a super first, then a method to update a visual. But it is declared using a double function pattern. Why? For two reasons: run-time efficience, and ease of debugging. Read all about it in dcl.superCall() documentation, and Supercalls in JS.

While this implementation looks fine, it is far from "fine". Let's be smart and look forward: in real life our implementation will be modified and augmented by generations of developers. Some will try to build on top of it.

  • Our call to showData() in construct is not going to be the last code executed, as we expected. Constructors of derived classes will be called after it.
  • updateData() will be overwritten, and some programmers may forget to call a super. Again, they may update data in their code after our code called showData() resulting in stale data shown.

Obviously we can write lengthy comments documenting our "implementation decisions", and suggesting future programmers ways to do it right, but who reads docs and comments especially when writing "industrial" code in a crunch time?

It would be nice to solve those problems in a clean elegant way. Is it even possible? Of course. That's why we have AOP.

Let's rewrite our attempt #1:

var NameWidget2 = dcl([Data, Template, Node], {
  declaredClass: "NameWidget2",
  template: "Hello, ${firstName} ${lastName}!",
  constructor: dcl.after(function(){
  updateData: dcl.after(function(){
  showData: function(){
    var text = this.render("template");;

var x = new NameWidget2({
  node:      document.getElementById("name"),
  firstName: "Bob",
  lastName:  "Smith"
// Hello, Bob Smith!

x.updateData({firstName: "Jill"}); // Hello, Jill Smith!

Not only we got a (slightly) smaller code, now we are guaranteed, that showData() is called after all possible constructors, and after every invokation of updateData(), which can be completely replaced with code that may use supercalls. We don't really care --- we just specified code, which will be executed *after* whatever was put there by other programmers.

Now imagine that our user wants to click on a name, and get a pop-up with more detailed information, e.g., an HR record of that person. It would make sense to keep the information in one place, yet render it differently. And we already have a provision for that: we can add another template property, and call render() with its name:

var PersonWidget1 = dcl(NameWidget2, {
  declaredClass: "PersonWidget1",
  detailedTemplate: "..."

var x = new PersonWidget1({
  node:      document.getElementById("name"),
  firstName: "Bob",
  lastName:  "Smith",
  position:  "Programmer",
  hired:     new Date(2012, 0, 1) // 1/1/2012
// Hello, Bob Smith!

var detailed = x.render("detailedTemplate");

In the example above I skipped the definition of a detailed template for brevity. But you can see that we can add more information about person, and we can define different templates when a need arises.

Imagine that we profiled our new implementation and it turned out that we call render() method directly and indirectly very frequently, and it introduces some measurable delays. We can pre-render a template eagerly on every data update, yet it sounds like a lot of work for several complex templates, and some of them are not even going to be used. Better solution is to implement some kind of lazy caching: we will invalidate cache on every update, yet we will build a string only when requested.

Obviously such changes involve both Data and Template. Or it can be done downstream in NameWidget or PersonWidget. Now look above and please refrain from doing those changes: so far we tried to keep our "classes" orthogonal, and caching is clearly an orthogonal business.

dcl already provides a simple solution: memoize advice. Let's use it in our example:

var PersonWidget2 = dcl(NameWidget2, {
  declaredClass: "PersonWidget2",
  detailedTemplate: "...",
  // memoization section:
  render:     dcl.advise(memoize.advice("render")),
  updateData: dcl.advise(memoize.guard ("render"))

With these two lines added our render() result is cached for every first parameter value ("template" or "detailedTemplate" in our case), and the cache will be invalidated every time we call updateData().

In this article we presented dcl package. If you plan to use it in your Node.js project install it like this:

npm install dcl

For your browser-based projects I suggest to use volo.js:

volo install uhop/dcl

The code is an open source on under New BSD and AFL v2 licenses.

This article didn't cover a lot of other things provided by dcl:

  • Avoid the double function pattern in your legacy projects using inherited() supercalls.
  • Use AOP on object-level --- add and remove advices dynamically in any order.
  • Specify "before" and "after" automatic chaining for any method.
  • Use debug helpers that come with dcl.
  • Leverage a small library of canned advices and mixins provided by dcl.

If you want to learn more about it, or just curious, you can find a lot of information in the documentation.

Happy DRY coding!

Eugene Lazutkin

About Eugene Lazutkin

Eugene's interests include programming fundamentals, complex systems, performance, and graphics. In his professional life he wrangles with Dojo and Django, in his time off he contributes to open source projects. Eugene prefers to deal with dynamic languages, especially JavaScript and Python, and likes constructing challenging web applications.

Track.js Error Reporting

Recent Features

  • Serving Fonts from CDN

    For maximum performance, we all know we must put our assets on CDN (another domain).  Along with those assets are custom web fonts.  Unfortunately custom web fonts via CDN (or any cross-domain font request) don't work in Firefox or Internet Explorer (correctly so, by spec) though...

  • CSS Gradients

    With CSS border-radius, I showed you how CSS can bridge the gap between design and development by adding rounded corners to elements.  CSS gradients are another step in that direction.  Now that CSS gradients are supported in Internet Explorer 8+, Firefox, Safari, and Chrome...

Incredible Demos

  • MooTools Zoomer Plugin

    I love to look around the MooTools Forge. As someone that creates lots of plugins, I get a lot of joy out of seeing what other developers are creating and possibly even how I could improve them. One great plugin I've found is...

  • Introducing MooTools ScrollSpy

    I've been excited to release this plugin for a long time. MooTools ScrollSpy is a unique but simple MooTools plugin that listens to page scrolling and fires events based on where the user has scrolled to in the page. Now you can fire specific...


  1. Wow great info :)

    • Jaataveda

      very nice

Wrap your code in <pre class="{language}"></pre> tags, link to a GitHub gist, JSFiddle fiddle, or CodePen pen to embed!