I found an inconsistency in an app, which had to do with the color of a certain action’s title.

Easy fix, I thought. I can fix this without too much work.

So that’s what I did. And I was right — it was an easy fix. However I wanted to go a step further, to prevent this from happening in the future, and to validate the current behavior.

The problem

The problem was a UIAlertAction having a style of .default (making the button blue), where it was expected to be .destructive (which would make it red). The action was deleting a post, after all. The code looked something like this, and was inside a UIViewController subclass:

func presentActionsAlert(for input: [String]) {
    let alertController = UIAlertController(
        title: nil,
        message: nil,
        preferredStyle: .actionSheet

    for value in input {
        switch value {
        case "share":
            let action = UIAlertAction(title: "share", style: .default)
        case "delete":
            let action = UIAlertAction(title: "delete", style: .default)

    let cancelAction = UIAlertAction(title: "cancel", style: .cancel)

    present(alertController, animated: true)

The fix was easy:

- let action = UIAlertAction(title: "delete", style: .default)
+ let action = UIAlertAction(title: "delete", style: .destructive)

So I committed the fix, but before opening a pull request I wanted to check if all tests still succeeded (as to not waste resources on our continuous integration infrastructure).

Guess what. All tests passed. 😱

So I took another look at the function, and thought… this makes sense. How would I ever test this?

Improving testability

So I refactored the function to only create the UIAlertController, without the side effect of presenting it.

- func presentActionsAlert(for input: [String]) {
+ func actionsAlert(for input: [String]) -> UIAlertController {

- present(alertController, animated: true)
+ return alertController

What is nice about this, is that we reduced our function to a function that does not do more than getting our input and providing us with an output1.

At this point, I was ready to write a unit test, that would validate some assumptions — in this case checking that if we pass in a "delete" as input, expecting a .destructive style as the output. If this would be changed in the future, the corresponding unit test would fail.

Lets take a look at this unit test.

class MyTest: XCTestCase {
    let myViewController = MyViewController()

    func test_that_aDeleteInput_has_aDestructiveStyle() {
        let input = "delete"
        let output = myViewController.actionsAlert(for: [input])

            output.actions.first { $0.title == input }?.style,
            "A `delete` input should have a `.destructive` style."

And there we go: our assumption is being validated in a test2. 🎉


It is good to take a look and see if you can improve a piece of code, especially when it involved a contained fix like this one. Check if there is something to be done to leave the code a bit cleaner than you found it3.

  1. If that makes you think about functional programming: you’re right. In fact, in its current form, this is a pure function. And if pure functions are awesome for something, it is testability! 

  2. In the real test case I wrote, I validated some other things, like there being a .cancel action present, as well as the actions’s count when we pass in some input that would end up in the default case. 

  3. Now that the function does not rely on anything within the UIViewController, we could even take it out of there. Tearing down Massive ViewControllers one function at a time.