ad-hockery: /ad·hok'@r·ee/, n.
Gratuitous assumptions... which lead to the appearance of semi-intelligent behavior but are in fact entirely arbitrary. Jargon File

Decoupling from the DOM with Angular

One piece of advice you’ll run into pretty soon when working with Angular is that you should never touch the DOM outside of a directive. Especially when test-driving your components this is pretty wise. The great strength of Angular is the declarative way in which the view (HTML) works with the view model (Angular controllers). It’s almost absurd how easy it is to unit test Angular controllers. Controller functions tend to act on $scope properties, trigger or respond to events and all those things are straightforward to replicate in unit tests.

However, since Angular directives can contain controllers the temptation can be to write some pretty non-idiomatic Angular code by writing view logic that really belongs in a "pure" controller in a bloated directive that happily interacts with the DOM via the $element injected into its controller in much the way a Backbone view might.

I was recently building some fairly simple CRUD functionality for my current client. There’s a list of products and each has an edit button that pops up a Bootstrap modal containing a form with the product data.

I initially built this using:

  • a controller that is triggered by the URL route and displays the product list.

  • a directive for the form with its own controller exporting scope properties representing the selected product and scope functions that handle things like clicking the save or cancel buttons.

The product list is displayed in a table and the form is declared after it:

  <tr ng-repeat="product in products">
    <!-- you get the idea -->

<div class="modal" edit-product-form>
  <!-- a form that contains things like... -->
  <input ng-model="product.title">
  <!-- and other inputs and buttons -->

Each element in the product list contains an edit button:

<button type="button" ng-click="edit(product)">Edit</button>

Clicking the button triggers an event using $scope.$broadcast:

$scope.edit = function(product) {
  $scope.$broadcast('product:edit', product);

That event is picked up by the edit-product-form directive. It places the product in its isolate scope – which automatically populates the form with the product’s properties – and pops up the modal:

angular.module('myApp').directive('edit-product-form', function() {
  return {
    restrict: 'A',
    controller: ['$scope', '$element', function($scope, $element) {
      $scope.$on('product:edit', function(event, product) {
        $scope.product = product;


The save button handler in the directive sends another event back up the scope heirarchy using $scope.$emit and hides the modal again.

$ = function() {
  $scope.product.$save(function() {
    $scope.$emit('product:updated', $scope.product);
  function() {
    // display an error message

Fine, this works but it’s interacting with the Bootstrap modal jQuery plugin in an imperative way. When I want the modal to appear I tell it to appear explicitly. When I want it to disappear I tell it to hide. This is mixing up the logical action of placing the form data in the scope, saving changes and updating the list view with the specifics of how the form is displayed – with a Bootstrap modal dialog. The logic governing view state would be the same if showing and hiding the form were handled in a different way. I’ve violated the separation of concerns.

To summarize the problems:

  • Unit testing the code requires at least a minimal DOM even for assertions purely concerned with the scope properties.

  • If I want to use a similar modal elsewhere in the app I can’t re-use anything here unless you consider copy 'n paste to be reuse.

  • If I want to use a different widget like Foundation’s Reveal to show and hide the form I need to pick through – and be careful not to break – the code that governs the view state even though that’s a separate concern.

We can do better.

Let’s pare things back to the simplest state – what am I trying to do here? I want to hide the "edit product" form until I have selected a product to edit, show the form while I’m working on that product then hide it again once I’m finished. The view state of "user has selected a product" is represented by the existence of a property called $scope.product. With no fancy presentation or slide in/out effects I could do this by just adding and removing that scope property at the appropriate time:

$scope.edit = function(product) {
  $scope.product = product;

$ = function() {
  $scope.product.$save(function() {
  }, function() {
    // display an error message

Then in the view I can declaratively react to the presence or absence of the product property:

<div ng-show="product">

The ng-show directive hides the element unless there is a non-null, non-undefined property in the scope that matches its value.

Simple. Now the view does the right thing based on state of the view model. You may or may not have noticed but I’ve done away with the directive for the edit form now – the edit and save functions can just appear as part of the controller that manages the product list. Arguably the directive is still useful for modulatity – especially if the form is complex – but crucially I don’t need it because I don’t need to imperatively interact with the DOM.

I should be able to use a Bootstrap modal to hide and show the form instead of the basic ng-show directive without changing any of the controller code. The mechanism for showing and hiding the form is an implementation detail of the view that’s nothing to do with the view model. Managing that kind of separation of concerns is the promise of Angular in a nutshell.

To do this I’ll add the Boostrap modal class to my view:

<div class="modal fade" trigger="product">

Don’t worry about the trigger attribute – I’ll get to that.

Then I’ll create a new directive that is activated simply by the presence of the modal class on the element.

angular.module('myApp').directive('modal', function() {
  return {
    restrict: 'C',
    controller: ['$scope', '$element', '$attrs', function($scope, $element, $attrs) {
      $scope.$watch($attrs.trigger, function(newValue, oldValue) {
        if (!!newValue && !oldValue) {
        if (!!oldValue && !newValue) {

The restrict: 'C' means the directive is activated by a class name that is the same as the directive name. In other words any element that has class="modal" will have this directive applied.

The directive uses the $scope.$watch function to respond to changes in a scope property defined by the trigger attribute on the directive’s element. Note that $scope.trigger still works as is if, like me, you’re fussy about validity and declare the attribute as data-trigger="product".

The $watch callback is invoked every time the named scope property changes and is passed the previous and new values of that property. In this example I only care whether the property has changed from null/undefined to an object or vice-versa. I don’t want to call $element.modal 'show' again if the modal is already visible but $ has just changed to point to a different object. (If you haven’t seen the !! operator in JavaScript before it’s simply a way of coercing any type to boolean).

I’ve now got two components – a controller that purely interacts with scope properties and a directive that responds to scope changes and calls a jQuery plugin method.

  • Both those are very simple to unit test.

  • Neither is coupled to the other except via the trigger attribute in the view.

  • I can re-use the modal directive in similar scenarios elsewhere in the system by just adding class="modal" and an appropriate trigger attribute.

  • If I want to use a different UI widget I only have to change the directive.

Learning to think about separation of concerns in this way is key to getting the most out of Angular. Remember controllers are for managing view state and directives are for managing the view implementation. If you find yourself mixing those concerns step back and think about how you can separate them. The resulting code will be easier to follow, easier to change and easier to test.

Web Statistics