James fuller asks:

any thoughts about which layer of the application we should be using a DI container like Aura.Di? Highest layer possible?

Twitter is too constrained and ephemeral for a good response, so I'll answer that question here.

First, we need to remember that a Dependency Injection container and a Service Locator are indistinguishable from an implementation perspective. (This is a view I accepted only recently.) The difference is in their use: if a container is ever placed into an object so the object can retrieve its own dependencies, that container is being used as a Service Locator. Conversely, if the container stays entirely outside an object so that dependencies are pushed into the object, the container is being used for dependency injection.

If we accept that description, the proper layer for a DI container is outside every other layer in the application (with the possible exception of Factory classes). That means the DI container is part of the bootstrapping code that sets up the application, and not part of anything else.

As an example, we can imagine a web application bootstrap script that looks like this:

// set up the autoloader
include '/path/to/autoload.php';

// create and set up a Container
$container = new Container;

// retrieve a front controller service and run it
$front_controller = $container->get('front_controller');

The purpose of the bootstrap is to create and configure the container with all its services and settings, then pull one object out of the container and invoke it. None of the objects in the system has to know that the container even exists. All of the objects are being created via the container, so in a way the container "contains" the entire object graph.

The part I had the hardest time getting about real dependency injection was this: If you have a class 2 or 3 layers down in the application, and that class needs dependencies, how do they get injected? Don't you have to pass the dependencies through the intervening layers, and then doesn't that break a lot of rules, along with violating plain common sense? And what if that class needs access to several different but related objects, but doesn't necessarily know what they are in advance?

The key for me was realizing that all object creation work happens in the container and not in the classes themselves. This is an alien way of looking at things for most PHP developers, who are more used to Singleton, Registry, and Service Locator, not to mention static calls to ActiveRecord objects. All of these patterns of usage mess with separation of dependencies, but are very common in PHP land, and form the basis of most PHP developers' understanding of how to work with dependencies.

It took me a year to really "get" the concept of dependency injection, under tutelage from guys like Jeff Moore and Marcus Baker, along with the writings of Misko Hevery. Once I understood it, I wrote the Aura.Di readme file as an introductory document to step readers through the various stages of dependency allocation. Perhaps that document will be helpful to others, even if they don't choose Aura.Di as a container.


If you work with a legacy application, you already know how hard it is to track dependencies throughout the code. Sometimes the dependencies are global, sometimes they are created inside that class that needs them, and sometimes there is a service locator implementation floating around inside the classes. This makes it frustrating to track down bugs and make wide-ranging changes, and terribly difficult to write tests because everything depends on everything else.

But it doesn't have to stay that way. My new book, Modernizing Legacy Applications in PHP, gives detailed instructions that will lead you step-by-step from an include-oriented mess of page scripts to a modern architecture using dependency injection techniques.

You can buy the book in early-access mode today, or sign up on the mailing list below to get more information as it becomes available!

Are you stuck with a legacy PHP application? You should buy my book because it gives you a step-by-step guide to improving you codebase, all while keeping it running the whole time.