This site runs best with JavaScript enabled.

Making your UI tests resilient to change

Software Engineer, React Training, Testing JavaScript Training

Photo by Warren Wong


User interface tests are famously finicky and prone to breakage. Let's talk about how to improve this.

You're a developer and you want to avoid shipping a broken login experience, so you're writing some tests to make sure you don't. Let's get a quick look at an example of such a form:

Login form from the Bookshelf App

1const form = (
2 <form onSubmit={handleSubmit}>
3 <div>
4 <label htmlFor="username">Username</label>
5 <input id="username" className="username-field" />
6 </div>
7 <div>
8 <label htmlFor="password">Password</label>
9 <input id="password" type="password" className="password-field" />
10 </div>
11 <div>
12 <button type="submit" className="btn">
13 Login
14 </button>
15 </div>
16 </form>
17)

Now, if we were to test this form, we'd want to fill in the username, password, and submit the form. To do that properly, we'd need to render the form and query the document to find and operate on those nodes. Here's what you might try to do to make that happen:

1const usernameField = rootNode.querySelector('.username-field')
2const passwordField = rootNode.querySelector('.password-field')
3const submitButton = rootNode.querySelector('.btn')

And here's where the problem comes in. What happens when we add another button? What if we added a "Sign up" button before the "Login" button?

1const form = (
2 <form onSubmit={handleSubmit}>
3 <div>
4 <label htmlFor="username">Username</label>
5 <input id="username" className="username-field" />
6 </div>
7 <div>
8 <label htmlFor="password">Password</label>
9 <input id="password" type="password" className="password-field" />
10 </div>
11 <div>
12 <button type="submit" className="btn">
13 Sign up
14 </button>
15 <button type="submit" className="btn">
16 Login
17 </button>
18 </div>
19 </form>
20)

Whelp, that's going to break our tests. But that'd be pretty easy to fix right?

1// change this:
2const submitButton = rootNode.querySelector('.btn')
3// to this:
4const submitButton = rootNode.querySelectorAll('.btn')[1]

And we're good to go! Well, if we start using CSS-in-JS to style our form and no longer need the username-field and password-field class names, should we remove those? Or do we keep them because our tests use them? Hmmmmmmm..... 🤔

So how do we write resilient selectors?

Given that "the more your tests resemble the way your software is used, the more confidence they can give you", it would be wise of us to consider the fact that our users don't care what our class names are.

So, let's imagine that you have a manual tester on your team and you're writing instructions for them to test the page for you. What would those instructions say?

  1. get the element with the class name username-field
  2. ...

"Wait," they say. "How am I going to find the element with the class name username-field?"

"Oh, just open your devtools and..."

"But our users wont do that. Why don't I just find the field that has a label that says username?"

"Oh, yeah, good idea."

This is why Testing Library has the queries that it does. The queries help you to find elements in the same way that users will find them. These queries allow you to find elements by their label, placeholder, text contents, alt text, title, display value, role, test ID.

That's actually in the order of recommendation. There certainly are trade-offs with these approaches, but if you wrote out instructions for a manual tester using these queries, it would look something like this:

  1. Type a fake username in the input labeled username
  2. Type a fake password in the input labeled password
  3. Click on the button that has text sign in

And that would help to ensure that you are testing your software as closely to how it's used as possible. Giving you more value from your test.

What's with the data-testid query?

Sometimes you can't reliably select an element by any of the other queries. For those, it's recommended to use data-testid (though you'll want to make sure that you're not forgetting to use a proper role attribute or something first).

Many people who hit this situation, wonder why we don't include a getByClassName query. What I don't like about using class names for my selectors is that normally we think of class names as a way to style things. So when we start adding a bunch of class names that are not for that purpose it makes it even harder to know what those class names are for and when we can remove class names.

And if we simply try to reuse class names that we're already just using for styling then we run into issues like the button up above. And any time you have to change your tests when you refactor or add a feature, that's an indication of a brittle test. The core issue is that the relationship between the test and the source code is too implicit. We can overcome this issue if we make that relationship more explicit.

If we could add some metadata to the element we're trying to select that would solve the problem. Well guess what! There's actually an existing API for this! It's data- attributes! For example:

1function UsernameDisplay({user}) {
2 return <strong data-testid="username">{user.username}</strong>
3}

And then our test can say:

1const usernameEl = getByTestId('username')

This is great for end to end tests as well. So I suggest that you use it for that too! However, some folks have expressed to me concern about shipping these attributes to production. If that's you, please really consider whether it's actually a problem for you (because honestly it's probably not as big a deal as you think it is). If you really want to, you can compile those attributes away with babel-plugin-react-remove-properties.

Conclusion

You'll find that testing your applications in a way that's similar to how your software is used makes your tests not only more resilient to changes, but also provide more value to you. If you want to learn more about this, then I suggest you read more in my blog post Testing Implementation Details.

I hope this is helpful to you. Good luck!

Discuss on TwitterEdit post on GitHub

Share article
loading relevant upcoming workshops...

TestingJavaScript.com

Your essential guide to flawless testing.

Jump on this self-paced workshop and learn the smart, efficient way to test any JavaScript application. 🏆

Start Now
Kent C. Dodds

Kent C. Dodds is a JavaScript software engineer and teacher. He's taught hundreds of thousands of people how to make the world a better place with quality software development tools and practices. He lives with his wife and four kids in Utah.