mokacoding

unit and acceptance testing, automation, productivity

Better tests for delegates

Delegation is a powerful design patter. The Apple frameworks Foundation and UIKit use it in many places, such as UITableViewDelegate or URLSessionDelegate.

Testing how an object interacts with its delegate is straightforward. Define a spy test double, use it to keep track of the methods called, and parameters passed to them, assert the behavior you are expecting by looking into the state of the spy.

For example, we can test how a ResourceFetcher type calls its delegate upon a successful fetch like this.

import XCTest
import Nimble

class ResourceFetcherTests: XCTestCase {

  func testCallsDelegateOnSuccess() {
    let resourceFecther = ResourceFetcher()
    let delegateSpy = ResourceFetcherDelegateSpy()
    resourceFecther.delegate = delegateSpy

    resourceFecther.fetch()

    expect(delegateSpy.didCallResourceFetched).toEventually(beTrue())
    expect(delegateSpy.didCallResourceFetchedValue).toEventually(equal(Resource(id: 1)))
  }
}

protocol ResourceFetcherDelegate: class {

  func resourceFetcher(_ resourceFetcher: ResourceFetcher, fetched resource: Resource)
}

class ResourceFetcher {

  weak var delegate: ResourceFetcherDelegate?

  func fetch() {
    // In a real system, you would have something like a network request here.
    DispatchQueue.global(qos: .background).asyncAfter(deadline: .now() + 0.1) { [weak self] in
      guard let self = self else { return }

      self.delegate?.resourceFetcher(self, didFetch: Resource(id: 1))
    }
  }
}

lass ResourceFetcherDelegateSpy: ResourceFetcherDelegate {

  private(set) var didCallResourceFetched = false
  private(set) var didCallResourceFetchedValue: Resource?

  func resourceFetcher(_ resourceFetcher: ResourceFetcher, didFetch resource: Resource) {
    didCallResourceFetched = true
    didCallResourceFetchedValue = resource
  }
}

I talk more about this testing tactic here.

Behaviour vs. Implementation

I want my test suites to help developers as their codebase evolves. This means we should be able to refactor code without having to update its tests. In the words of Corey Haines in this podcast interview, "the test suite should serve you; you shouldn't serve it."

Test suites that focus on asserting behavior rather than implementation serve us better in the long run. Focusing on behavior means that we can change implementation details, such as method names or the order of their parameters, and perform refactors like extract method, without having to update our tests.

The use of a delegate is an implementation detail. In the example of ResourceFetcher, the behavior is the act of communicating the fetch has completed successfully, the fact that a delegate is used is an implementation detail.

On the other hand, we cannot test ResourceFetcher in isolation without taking into account the implementation detail that is its delegate. Testing components in isolation is usually a valuable thing to do; it ensures they are flexible, easy to consume, and easier to extract.

Let's say we wanted to use the word didFetch instead of fetched in our ResourceFetcherDelegate, to be consistent with the naming conventions used in the Apple libraries (tableView(_:didSelectRowAt:), urlSession(_:didReceive:completionHandler:) etc.)

- func resourceFetcher(_ resourceFetcher: ResourceFetcher, fetched resource: Resource)
+ func resourceFetcher(_ resourceFetcher: ResourceFetcher, didFetch resource: Resource)

To make our test read consistently, we would have to update them:

  // ResourceFetcherTests.swift
- expect(delegateSpy.didCallResourceFetched).toEventually(beTrue())
- expect(delegateSpy.didCallResourceFetchedValue).toEventually(equal(Resource(id: 1)))
+ expect(delegateSpy.didFetchResourceCalled).toEventually(beTrue())
+ expect(delegateSpy.didFetchResourceValue).toEventually(equal(Resource(id: 1)))

  // ResourceFetcherDelegateSpy.swift
  class ResourceFetcherDelegateSpy: ResourceFetcherDelegate {

-   private(set) var didCallResourceFetched = false
-   private(set) var didCallResourceFetchedValue: Resource?
+   private(set) var didFetchResourceCalled = false
+   private(set) var didFetchResourceValue: Resource?

    func resourceFetcher(_ resourceFetcher: ResourceFetcher, didFetch resource: Resource) {
-     didCallResourceFetched = true
-     didCallResourceFetchedValue = resource
+     didFetchResourceCalled = true
+     didFetchResourceValue = resource
    }

I know, it's a small thing. However, small things matter, and they do add up. Plus, let me repeat it one more time, I'm not interested in what the delegate method is called, but in the behavior it captures.

If a new reader of the codebase, which includes our future selves, sees a test suite made up of tests and assertions biased towards behavior they'll be more likely to adopt the same style.

If we agree on the value of testing how objects call their delegates, and on the value of focusing on behavior rather than implementation, is there a way for us to write these tests without being coupled to implementation details?

A behavior-oriented delegate test

A way to mitigate the intrinsic dependency on implementation when testing delegation is to shift the API we use in the tests from focusing on the method and property names towards the kind of behavior the delegate is notified about.

In the ResourceFetcher example above we could use something like this:

// ResourceFetcherTests.swift
func testCallsDelegateOnSuccess() {
  let resourceFecther = ResourceFetcher()
  let delegateSpy = ResourceFetcherDelegateSpy()
  resourceFecther.delegate = delegateSpy

  resourceFecther.fetch()

  expect(delegateSpy.wasCalledForSuccessfulFetch(of: Resource(id: 1)))
      .toEventually(beTrue())
}

// ResourceFetcherDelegateSpy.swift
class ResourceFetcherDelegateSpy: ResourceFetcherDelegate {

  private var fetchedResource: Resource?

  func wasCalledForSuccessfulFetch(of resource: Resource) -> Bool {
    return fetchedResource == resource
  }

  func resourceFetcher(_ resourceFetcher: ResourceFetcher, didFetch resource: Resource) {
      fetchedResource = resource
  }
}

When reading the test, we don't focus on the implementation details of the method name and its parameters, but instead on what they achieve. Those details, by the way, are one click away; we just need to drill into the implementation of the wasCalledForSuccessfulFetch(of:) method in the Spy.

Notice as well how the diff for a change in the delegate method name looks like this time:

  class ResourceFetcherDelegateSpy: ResourceFetcherDelegate {

    private var fetchedResource: Resource?

    func wasCalledForSuccessfulFetch(of resource: Resource) -> Bool {
      return fetchedResource == resource
    }

-   func resourceFetcher(_ resourceFetcher: ResourceFetcher, didFetch resource: Resource) {
+   func resourceFetcher(_ resourceFetcher: ResourceFetcher, fetched resource: Resource) {
        fetchedResource = resource
    }
  }

Because of the abstraction layer between how the delegate works and how we check if it's consumed properly, we had to update only the Spy implementation details, without having to touch the test using it.

Yes, this is a small achievement. But once again, small achievements all add up to make a huge difference in the long run.


I believe tests that focus on behavior rather than implementation are the most useful and sustainable kind of tests.

Depending on the business logic we are implementing it's not always possible to hide implementation details from the tests. In those occasions we can still try to mitigate them as much as possible. Hiding the implementation details of how a delegate is called behind a method named after the behavior being notified to the delegate is one such mitigation tactic.

What do you think of this approach? Have you used something similar? How do you minimize the number of implementation details you assert in your tests?