In the realm of programming, “unit testing” is an automated way of testing small, individual “units” of your code to ensure that when given a specific set of inputs they return a specific set of outputs in the way you expected them to.
If you’ve never delved into unit testing, you probably won’t realize all the ways it can improve your code and make you a more efficient programmer in the process.
Here are seven benefits the other UGRC developers have taught me or that I have discovered on my own since they introduced me to unit testing a year and a half ago.
1: You Know Your Logic Is Correct
This is the obvious benefit of unit testing. If you write tests that cover the right cases, you know that each tested part of your code works the way it should. This gives you peace of mind (and job security), especially when whatever your code spits out is being presented to the Mayor.
2: You Understand Your Code Better
Writing tests is all about verifying your assumptions of what your code does at a very detailed level. Often, when we’re working with new languages or new libraries, we tend to “program by accident.” We take example code from the docs (or Stack Overflow) and make it fit our problem. We don’t necessarily know why it works, just that it does (maybe).
Creating unit tests forces you to think about every step, especially if you have to create stand-in results from other functions that are input to your current function. Instead of copying a few lines and trial-and-erroring your way through, you have to really learn what they do and what kind of data they process.
In addition, if you write your tests first and then write code that makes them pass (test-driven development), you will have already written code that consumes your function. You’ll know how your function fits into the flow of your main program.
3: You Catch Your Coding Errors Faster
Because you understand your code better, you catch errors faster. Sometimes, this can happen even before you run your tests. As you build a test, thinking about what the input and output data should look like, you realize that your function doesn’t do exactly what you thought it did, or outputs data in a different structure than you thought.
Obviously, a test that fails will also reveal errors. Sometimes they’re errors with how your function processes data. Other times its the grand revelation that your function isn’t doing what you thought, like that time I realized that my function actually returned a list instead a single string.
This can be especially helpful when manipulating large pandas data frames or other complex data structures.
4: You’re (mostly) Protected From Breaking Your Code When You Update It
Have you ever gone back to your code a year later to make a small change and suddenly it doesn’t work anymore? Unit testing (and it’s cousin, integration testing) can help you ensure that your code is still giving the right outputs after you make changes.
This can be especially useful when you’re asked to update code someone else wrote (or that you wrote three years ago!). Once you’ve got some tests written against the original code, you can now make changes without worrying that you’re missing some deviously subtle edge case.
5: You Can Easily Test A Lot of Inputs and Edge Cases
Before I used unit testing, I would have some test datasets that I’d feed into my completed program to see if I got the right data out on the other end. “Testing” this way requires finding or crafting a new dataset for each set of conditions you want to test. It’s slow, inefficient, and makes it easy to miss a test case or skip it out of laziness.
Because unit testing involves programmatically setting up the inputs to your functions, you can quickly and easily set up tests for a whole range of conditions and cases. This is especially handy for code that requires you to select data through either a GUI, a command line, or a settings file. More tested conditions gives you greater confidence that your code works (see #1).
6: You Write More Self-Contained, “Atomic” Code
One of the guiding principles for writing functions and classes is that they do one thing and one thing only. In scholarly circles, this is known as the single responsibility principle (Sandi Metz has some rules on this that will really push you).
In practice, its very easy for functions to grow. I often start programming with a broad view of how to solve a problem, maybe three or four general steps. Each step becomes its own function. Then, each function grows as I figure out how to solve the smaller problems within that step.
The most useful (and easiest to write) unit tests only test one thing, one solution to a problem, at a time. Does that complex logical expression work the way I think it does? Does my translation of math to code work the way it should?
Because each test requires calling a function, thinking about making a solution easy to test also means wrapping that solution in its own, small, self-contained function. Your functions become simpler and more “atomic” (as in being broken down to the most basic level possible, not as in big kabooms).
7: You Write More Flexible, De-Coupled Code
This next benefit follows on from writing smaller, more self-contained functions/methods.
“Coupling” in programming refers to code that explicitly or implicitly relies on variables and objects from other parts of your program. Highly coupled code is fragile code: what seems to be a simple change to one part of the program could have hidden side effects down the line that will have you tearing your hair out trying to debug.
As you get used to writing unit tests, you’re constantly thinking “what data will I need to perform this one operation?”, “what other functions and libraries does this use?”, and “what other bits of my program does this need to run?” You’re always identifying what data structures you’ll have to create and which objects or methods you’ll need to create stand-ins for (referred to as mock objects).
Thus, as you’re writing your code, you’re aware of just how dependent the current function is on other functions and classes in your code and how painful it could be to make a change. You can then refactor your design now instead of later, when that “one small change that will only take 15 minutes” has introduced bugs that you’ve spent the last two hours debugging.