Modern SwiftUI: Dependencies

It doesn’t matter how much time you spend writing “clean” code with precisely modeled domains if you don’t also control your dependencies. Uncontrolled dependencies make it difficult to run your application in Xcode previews, simulators, and devices; make it difficult to write tests; and make your code base just harder to deal with.

What is a dependency?

Dependencies are the types and functions in your application that need to interact with outside systems that you do not control. Classic examples of this are API clients that make network requests to servers, but also seemingly innocuous things such as the UUID and Date initializers, file access, user defaults, and even clocks and timers, can all be thought of as dependencies.

You can get really far in application development without ever thinking about dependency management (or, as some like to call it, “dependency injection”), but eventually, uncontrolled dependencies can cause many problems in your code base and development cycle:

• Uncontrolled dependencies make it difficult to write fast, deterministic tests because you are susceptible to the vagaries of the outside world, such as file systems, network connectivity, internet speed, server uptime, and more.
• Many dependencies do not work well in SwiftUI previews, such as location managers and speech recognizers, and some do not work even in simulators, such as motion managers, and more. This prevents you from being able to easily iterate on the design of features if you make use of those frameworks.
• Dependencies that interact with 3rd party, non-Apple libraries (such as Firebase, web socket libraries, network libraries, video streaming libraries, etc.) tend to be heavyweight and take a long time to compile. This can slow down your development cycle.

Controlling dependencies

For the reasons above, and a lot more, it is highly encouraged that you to take control of your dependencies rather than let them control you.

In fact, in our series on “Modern SwiftUI,” where we rebuild Apple’s “Scrumdinger” application from scratch, we came face-to-face with this lesson as soon as we introduced code that called out to Apple’s Speech framework. We found that directly accessing Speech APIs from our feature completely broke the preview, making it difficult to iterate on the UI and functionality. We were forced to run the full app in the simulator, which destroyed the fast iteration cycle that previews give us.

So, we decided to take control of our dependence on Speech (and a lot of other dependencies too!) by putting an interface that we own in front of the framework:

struct SpeechClient {
  var authorizationStatus:
    @Sendable () -> SFSpeechRecognizerAuthorizationStatus
  var requestAuthorization:
    @Sendable () async -> SFSpeechRecognizerAuthorizationStatus
  var startTask:
    @Sendable (SFSpeechAudioBufferRecognitionRequest) async -> AsyncThrowingStream<
      SpeechRecognitionResult, Error
    >
}

Then we made use of our new Dependencies library to inject the dependency into the feature that needs to interact with the Speech framework:

class RecordMeetingModel: ObservableObject {
  @Dependency(\.speechClient) var speechClient
  …
}

So we no longer reach for Speech APIs directly, and instead we only go through the speechClient interface. For example, when asking for speech recognition authorization:

let authorization = await self.speechClient.authorizationStatus() == .notDetermined
  ? self.speechClient.requestAuthorization()
  : self.speechClient.authorizationStatus()

With that little bit of upfront work we were able to restore functionality in our previews because we can now mock the speech client to act as if the user had previously authorized speech recognition:

struct RecordMeeting_Previews: PreviewProvider {
  static var previews: some View {
    NavigationStack {
      RecordMeetingView(
        model: withDependencies {
          $0.speechClient.authorizationStatus = { .authorized }
        } operation: {
          RecordMeetingModel(standup: .mock)
        }
      )
    }
  }
}

The Speech framework isn’t the only dependency we controlled in our Standups application. We also controlled our dependence on the Date and UUID initializers, our dependence on clocks for time-based asynchrony, our dependence on the file system for persisting application data, and even our dependence on an AVAudioEngine for playing sound effects in the app.

Until next time…

That’s it for now. We hope you have learned a bit about why dependencies can be so pernicious and how you might take back control over them. Be sure to check out our Dependencies library to start wrangling in your dependencies today!

Check back in tomorrow for the 5th, and final, part of our “Modern SwiftUI” blog series, where we show how to write deep, nuanced tests now that our application is built with a precisely modeled domain and all dependencies controlled.