To gain a better understanding of the decisions knockoutjs made in terms of implementation, I decided to try and build my own version. I named it sea as a homonym for 'see' which is somehow related to observable properties, I swear. I started with a simple HTML page that basically contained a series of "things that should work". The first was a simple div that followed the mouse cursor using two observables to track position.

After a few more examples similar to that, I realized this library had two testable aspects: the frp parts, which can exist in any JS enviroment, and the data binding parts, which require a DOM. I decided to formalize and write some actual tests using mocha, because I'd heard you could use it in a browser and node.

And that's when things got difficult. Sure, mocha supports browser-based tests, but it leaves the part about getting your tests into the HTML harness up to you. You could just load a single JS file, but I bet your library is a bit more complex than just one file, and you're probably using a module system to help keep things... modular (such as requireing your assert library). So if you're using requirejs or browserify, it's up to you to write a build step to make it easy to consume the tests.

I also looked into using phantomjs or one of myriad modules that purported to easily get mocha working in phantomjs. The problem is that they all assume a build system (like grunt), assume you're writing non-requireable code, or assume you're using requirejs. Then there's the problem of getting the test output... out... of phantomjs. So for now, that will wait.

There was one more problem that greatly exacerbated my troubles: I wanted to use the exports interface that mocha provides. Every example I've seen of using mocha in a browser uses either the tdd interface or the bdd interface. These are especially suited for the browser because it's relatively easy to expose the interfaces globally in the browser environment, while the exports interface requires a more complex shim. I find both of them a little verbose, expecially after writing so many tests for vash using vows.

Here is the typical bdd interface:

describe('sea.observable', function(){

  it('defaults to undefined', function(){
      var o = sea.observable();
      assert.equal(o(), undefined);

  it('updates value', function(){
      var o = sea.observable(null);
      assert.equal(o(), null);
      assert.equal(o(2), 2);


That's fine, but I find the repeated describe and it distracting. Here's the same using the exports interface:

exports.observable = {

  'defaults to undefined': function(){
    var o = sea.observable();
    assert.equal(o(), undefined);

  ,'updates value': function(){
    var o = sea.observable(null);
    assert.equal(o(), null);
    assert.equal(o(2), 2);


Honestly, it's basically the same, I know. For more complex suites I've seen it get very difficult to read, but at this point I'm mostly arguing a personal preference.

Final goals for this testing environment:

  • Write "mocha" tests using the exports interface
  • Be able to require the in-progress library, along with anything else needed, like assert or chaijs or sinonjs
  • Write the tests without caring if they'll be running in a browser or node

Step 1: Project Structure

This is a simple project, and contains the following files:

/index.js        # frp-parts, node-only
/dom.js          # data binding, requires DOM
/package.json    # typical, but will have build/test commands
  runner.html    # the mocha test harness
  test.sea.js    # the test, run using mocha(1) or in runner.html

Using the library is simple:

var sea = require('./index')

var firstName = sea.observable('Johnny');
var fullName = sea.computed(function(){
  return firstName() + ' Tatlock';

console.log(firstName()) // 'Johnny'
console.log(fullName()) // 'Johnny Tatlock'


console.log(firstName()) // 'James'
console.log(fullName()) // 'James Tatlock'

And using the data-binding components (assumes a DOM):

<ul data-foreach="items()">
  <li data-text="typeof $data === 'function' ? $data() : $data"></li>

<script type="text/javascript">
var sea = require('./dom') // decorates sea with data-binding stuff

var items = sea.observableArray(['a', 'b', 'c']);

// tell sea to parse the entire DOM
  items: items

// in 2 seconds, add some more elements, this will cause the ul to
// render more items
  items.push('d', 'e', 'f')
}, 2000)

If you've ever used knockoutjs, then this should look very familiar. I changed the syntax a bit for data-binding to simplify my job (I didn't want to write a full parser, so instead of using a single data-bind attribute, each data- attribute is matched against a valid registered binding).

Step 2: The Tests

The node tests look like (abbreviated):

var assert = require('assert')
  , sea = require('sea')

exports.observableArray = {

  'with no args defaults to empty array': function(){
    var a = sea.observableArray();
    assert.equal(a().length, 0);

  ,'.push updates a computed': function(){
    var a = sea.observableArray()
      , c = sea.computed(function(){
        return a().join(',')

    assert.equal(c(), '');
    assert.equal(c(), '1');
    a.push('2', '3');
    assert.equal(c(), '1,2,3');

And the DOM tests (abbreviated):

var assert = require('assert')
  , sea = require('../dom')

// skip running if not in a browser
if(typeof window === 'undefined') return;

var scratch = document.querySelector('#scratch');
var $  = function(query, ctx){
  return [] || scratch).querySelectorAll(query));

exports['Data Binding'] = {

  'data-text': {

    before: function(){
      var html = '<p data-text="t()"></p>';
      scratch.innerHTML = html

    ,'observable updates value': function(){

      var t = sea.observable('Hello!');
      sea.applyBindings({ t: t }, scratch);
      var p = $('p')[0];

      assert.equal(p.textContent, 'Hello!');

      assert.equal(p.textContent, 'Goodbye!');

For the most part, writing a test for the browser or node is exactly the same in terms of structure. Obviously the browser test won't run in node because of lack of DOM objects, but the structure of the tests are the same.

Step 3: Bundling

Next we need to bundle the tests so that require works. It's pretty easy, but took a bit of fiddling with browserify to figure out:

node_modules/browserify/bin/cmd.js ./test/test.dom.js --standalone tests > test.bundle.js

This command:

  1. Runs browserify on test/test.dom.js, which spiders through and includes the dependencies, including index.js (sea), dom.js, assert and other things (like the process shim and such).
  2. Uses the --standalone flag with argument tests, which wraps the bundle in a UMD guard under the name of 'tests'. In the absense of a require function in the target environment, the bundle will be attached to the global window as window.tests.
  3. Dump the resulting output into test.bundle.js.

The key step is #2, otherwise there would be no way to reference the tests from the test runner, which is integral when using the exports interface. Browserify is really good at script interop, allowing you to explicitly control what is exported.

Step 4: Hacking the Harness

The last step is to do something super hacky: copy mocha's exports interface, modify it slightly, and place it into the typical test harness:

  <meta charset="utf-8">
  <title>Mocha Tests</title>
  <link rel="stylesheet" href="../node_modules/mocha/mocha.css" />
  <div id="mocha"></div>
  <div id="scratch"></div>
  <script src="../node_modules/mocha/mocha.js"></script>
  <script src="../test.bundle.js"></script>
    // This code is directly from mocha/lib/interfaces/exports.js
    // with a few modifications
    (function manualExports(exports, suite){
      var suites = [suite];


      function visit(obj) {
        var suite;
        for (var key in obj) {
          if ('function' == typeof obj[key]) {
            var fn = obj[key];
            switch (key) {
              case 'before':
              case 'after':
              case 'beforeEach':
              case 'afterEach':
                suites[0].addTest(new Mocha.Test(key, fn));
          } else {
            var suite = Mocha.Suite.create(suites[0], key);
    }(tests, mocha.suite));


The key here is that I'm manually passing in tests, defined by our test bundle, into a modified version of mocha's exports interface. The only modifications were to access mocha using the global Mocha, and to be able to pass in the exports object and suites, instead of relying on mocha's require event.

Step 5: Script it

I could go with a build system, like grunt or a Makefile, but that's too much for this tiny project. A few simple additions to the scripts field of my package.json will do fine:

"scripts": {
  "pretest": "node_modules/browserify/bin/cmd.js ./test/test.dom.js --standalone tests > test.bundle.js",
  "test": "node_modules/mocha/bin/mocha --recursive --ui exports"

If you use require and browserify, you really don't need a true build system. The worst is having to manually specify files for inclusion in said build system, and require takes care of that.


Even simple projects quickly get complex when you want to be able to test in a browser and node. With the power of browserify and mocha, a lot of the hard things are taken care of, leaving just a little bit of undefined but required glue. Hopefully this helps the next time you start a small project and want some tests!