Link vs compile vs controller

When you create a directive, you can put code into the compiler, the link function or the controller.

In the docs, they explain that:

  • compile and link function are used in different phases of the angular cycle
  • controllers are shared between directives

However, for me it is not clear, which kind of code should go where.

E.g.: Can I create functions in compile and have them attached to the scope in link or only attach functions to the scope in the controller?

How are controllers shared between directives, if each directive can have its own controller? Are the controllers really shared or is it just the scope properties?


Compile :

This is the phase where Angular actually compiles your directive. This compile function is called just once for each references to the given directive. For example, say you are using the ng-repeat directive. ng-repeat will have to look up the element it is attached to, extract the html fragment that it is attached to and create a template function.

If you have used HandleBars, underscore templates or equivalent, its like compiling their templates to extract out a template function. To this template function you pass data and the return value of that function is the html with the data in the right places.

The compilation phase is that step in Angular which returns the template function. This template function in angular is called the linking function.

Linking phase :

The linking phase is where you attach the data ( $scope ) to the linking function and it should return you the linked html. Since the directive also specifies where this html goes or what it changes, it is already good to go. This is the function where you want to make changes to the linked html, i.e the html that already has the data attached to it. In angular if you write code in the linking function its generally the post-link function (by default). It is kind of a callback that gets called after the linking function has linked the data with the template.

Controller :

The controller is a place where you put in some directive specific logic. This logic can go into the linking function as well, but then you would have to put that logic on the scope to make it "shareable". The problem with that is that you would then be corrupting the scope with your directives stuff which is not really something that is expected. So what is the alternative if two Directives want to talk to each other / co-operate with each other? Ofcourse you could put all that logic into a service and then make both these directives depend on that service but that just brings in one more dependency. The alternative is to provide a Controller for this scope ( usually isolate scope ? ) and then this controller is injected into another directive when that directive "requires" the other one. See tabs and panes on the first page of angularjs.org for an example.


I wanted to add also what the O'Reily AngularJS book by the Google Team has to say:

Controller - Create a controller which publishes an API for communicating across directives. A good example is Directive to Directive Communication

Link - Programmatically modify resulting DOM element instances, add event listeners, and set up data binding.

Compile - Programmatically modify the DOM template for features across copies of a directive, as when used in ng-repeat. Your compile function can also return link functions to modify the resulting element instances.