Skip to content

WHY_DI.md

Latest commit

 

History

History
27 lines (18 loc) · 3.9 KB

WHY_DI.md

File metadata and controls

27 lines (18 loc) · 3.9 KB

Why use dependency injection?

This document describes the basics of the dependency injection pattern; what it is; and why it is a good pattern to apply to your app development? The texts below uses the term DI as short for dependency injection.

A somewhat real example to illustrate

Instead of describing the pattern in abstract terms, let's use a simple view controller based example to understand it. If you are interested in an abstract description, Wikipedia has a great article.

Let's say we are developing a photo browsing app, where we have a view controller that displays a set of photos retrieved from the server. In this extremely simple app, we have a PhotosViewController that displays the photos, and a PhotosService that encapsulates the logic of requesting photos from our server. The PhotosViewController implements the view logic while the PhotosService contains the HTTP request sending and response parsing logic. Without using DI, our PhotosViewController would instantiate a new instance of the PhotosService in its init or viewDidLoad method. Then it can use the service object to request photos when it sees fit.

Now let's step back and analyze our code. In its current state, the PhotosViewController and PhotosService are tightly coupled. This leaves us with a few issues:

  1. We cannot change PhotosService without having to also change PhotosViewController. This may seem fine with just two classes, but in a real-world scenario with hundreds of classes this would significantly slowdown our app iteration.
  2. We cannot switch out the PhotosService class without having to change PhotosViewController. Let's imagine we have a better PhotosServiceV2 class we now want our view controller to use, we'll have to dig into the implementation of PhotosViewController to make changes.
  3. We cannot unit test PhotosViewController without also invoking the PhotosService implementation.
  4. We cannot develop PhotosViewController independently and concurrently with the PhotosService. This may not seem like a big deal with our extremely simple app, in a real-world team, our engineers would be blocked constantly.

Let's apply the DI pattern to our app. With DI, we will have a third class, in Needle's terms a Component class, that instantiates the PhotosService and pass it into the PhotosViewController as a protocol. Let's call this protocol PhotosServicing. Now our PhotosViewController no longer knows anything about the concrete implementation of PhotosService. It simply uses the passed in PhotosServicing protocol to perform its logic.

With DI applied, let's revisit the issues we had before:

  1. We can freely change the implementation of PhotosService without affecting our PhotosViewController.
  2. We can simply update our DI Component class to instantiate and pass PhotosServiceV2 into PhotosViewController, as long as the implementation still conforms to the PhotosServicing protocol. This allows us to freely switch implementations of the photos service without having to change anything in the view controller.
  3. We can properly unit test PhotosViewController by injecting, aka passing in, a mock PhotosServicing object.
  4. As soon as the PhotosServicing protocol is defined, we can independently and concurrently develop PhotosService and PhotosViewController implementations.

Dependency injection terminologies

Before moving on, let's define some terminology that is frequently used with the DI pattern. In our example app above, the PhotosService is typically referred to as a "dependency". Our PhotosViewController class is sometimes referred to as the "dependent" or "consumer". The act of passing an instance of PhotosServicing into the PhotosViewController is called "inject". In summary, our simple DI setup injects the PhotosServicing dependency into the consumer PhotosViewController.