Stealth Code

Automated Testing: TDD, Experimentation, and the Front-End


Bob Owen

Bob Owen

Automated Testing: TDD, Experimentation, and the Front-End

Posted by Bob Owen on .

Automated Testing: TDD, Experimentation, and the Front-End

Posted by Bob Owen on .

A recent conversation amongst some of my fellow Hack Reactor alumns got me thinking about one of my favorite development practices, and why the current direction of my career (working in UX and the Front End) seems to keep pulling me away from it.

I've been practicing TDD (Test Driven Development) off and on as a software engineer since early 2000s. I'm a full on convert about it. In TDD before you begin coding the functionality of your program, you first write an automated test that validates that, what you're about to create, returns the results you would expect. Then you run the test and see that it fails. Next you write the code that makes the test pass and see that it passes. Finally you might change the code you wrote confident in the knowledge that if you break it, your tests will catch it. At Hack Reactor they refer to this testing pattern as: Red, Green, Refactor.

The beauty of TDD is that it's a forcing function. It ensures that:

  • You understand the the problem you're trying to solve. (For a given input you know what the expected output is.)
  • You've thought through the API of the problem you're trying to solve. (You are more likely to define an easy to use API if you are forced to use it up front in your tests.)
  • That you know your code works before writing other code that depends upon it. (This is huge. TDD seems slower at first, but because you know you can trust you're code it helps you work faster overall.)
  • That if something else in the system breaks your code, you have an easy way to find out about it.

Still, there are two classes of problems that make this kind of development really difficult in some cases.

One, is when you aren't exactly sure what your building. If you don't yet know that your tackling the right problem or that the solution you're hoping to create is actually achievable, then how can you even define what you are going to test?

I propose that this situation calls for a different kind of Test Driven Development... in this case your test is really to answer the above issues in the form of an experiment. This idea, is one of the central tenents of the Lean build, test, learn cycle. In this case, you're writing code to create a single instance, non-automated test. You're experimenting to see if your problem can even be solved and that your approach can even work. (Usually by looking at the parts of the problem you don't understand and taking a stab solving them.)

This isn't a new idea... several decades ago Fredrick Brooks wrote in The Mythical Man Month to "build one to throw away." The experiment you create here should not ever make it's way into production. Rather it's a kind of guide to the code you will later write for product. And you can apply TDD to that.

Another sticky area for TDD is the Front-End. It's not that people haven't tried to create tools to solve automated testing for the front-end, it's just that implementing them rarely reaps many rewards.

I see two big reasons for this:

  • I'm not aware of any front-end automated testing suites that actually test the right things.
  • The front-end, more that any other part of the system, is in a constant state of flux.

When it comes to front-end automated testing tools today, there are really only two kinds of testing they do: Some ensure that certain expected strings exist within the UI and maybe are placed as expected within the hierarchy. (This can be labels, certain types of controls, or ids associated with them.) The other is basically to do pixel (i.e. screen shot) analysis and compare the way the UI looks now with how it looked at some point before.

One issue with this is that modern UI is a lot more dynamic than it used to be. How can these systems notice that the animation is working as designed or that the sounds are being produced the right way. It's concievable that this might be solved by more advanced tools in the future.

The biggest issue, however, is that none of the things the automated tools measure actually matter very much in a production user interface. What really matters are the kinds of things we measure in a usability test:

  • Was the user able to figure out how to get the system to do what they needed to do?
  • Was the systems output what the user expected to get?
  • How long did it take for the user to accomplish a task?
  • How many mistakes did they make.
  • How did the UI make them feel about accomplishing the task?

The UI is about interfacing with a human. Unfortunately, humans are very difficult and/or expensive to automate.

Even if you can solve for that problem with some kind of mechanical turk user testing solution, there's another big problem: The UI layer of an application is likely the most mauable and changing part of a software system.

Bill Scott, VP of Product Development at PayPal, calls the UI the "experimentation layer." The companies that seem to be creating the most useful and delightful user experiences are constantly tweaking or whole-sale revamping the UI layer of thier applications. So assuming you even had the right tools to automatically measure the right aspects of a UI, the whole thing is likely to soon get ripped out from under whatever tests you put in place.

How can you automatically test a constantly changing experiment?

I'd love you hear your ideas.

View Comments...