I think automated testing, particularly automated unit testing, is a misunderstood creature. Every time I run into a “Do we write tests or do we not?” discussion in the wild, the focus is around the confidence tests give one about the functionality of the code. People in favor of tests generally argue about how you gain higher confidence about how well the code functions, and people counter that with their existing approaches like QA or the fact that their management accepts bugs as part of the process.
For me, that confidence boost isn’t the point; it’s a nice result, a by-product of the real advantage. It’s like scallions on my baked potato, or the sauteed onions on my steak, or the whipped cream on my sundae. Yes, I relish those things and they give me the extra hit of dopamine that makes me feel I have to have them…but I don’t skip the potato because I only have sour cream, and I definitely don’t refuse ice cream because there’s no toppings. The garnish is just the obvious benefit to the dish, the same way the “Confidence it will work” is the crouton on the testing salad.
Valuable tests are more than logic validators. They enforce sane engineering practices, expose complexity, put a fan behind code smells, and provide documentation that will be up-to-date as long as people are running the test suite. In short, automated tests are a way to avoid poor Quality code.
Tests Help Avoid Poor Quality Code
Quick qualifier here -- working code is good code, but we all know Quality code when we're in it. Yes, I'm
stealingapplying concepts from Pirsig again.
Poor Quality code is difficult to test. Sometimes so difficult it’s not worth it, like an abandoned copper mine that still has plenty of ore… but it’d take an investment many times over market price to extract it safely. For me, I can use the ease of writing a test to tell me a lot about the architecture, patterns, and configuration of the codebase. It’s the quickest way to identify challenges I’m going to face.
- “Oh. This method is new-ing up a dependency in-line. That’s…that’s going to be hard to mock.”
- “So part of the constructor on this service class is… calling out to a 3rd party API for part of it’s configuration?”
- “Wait. The controller is making a call directly to DatabaseA, so it can use the return to make a call to a service class that talks to DatabaseB?”
All things I’ve run into trying to write what I thought would be quick tests around legacy code, moving roughly from easiest to fix to most difficult. If you never write tests, and the app worked correctly right off the bat, and you never need to change the functionality, none of these things are problems, per se. But when was the last time you had an application go into prod without a problem? How many stakeholders have you met whose requirements are written in stone?
When I worked at a home improvement retailer, one of my jobs was to cut wood to size for customers. I was taught by a retired carpenter-turned-retailer to measure on the saw where the piece of wood would be once it was cut, then clamp scrap wood there. This gave me a guide to know when the 2×4 was cut correctly without too much conscience effort, and also prevented me from cutting too much.
For me, writing a test is a lot like clamping that scrap into place. The test is a fixed goal to hit, and you know without question and without thinking about it while you’re coding if you’ve hit it or not. If tests are failing (and that includes refusing to compile or build), the code isn’t right. Could I cut 2x4s by measuring each 18 inch length out individually? Yes, but it’s harder than just cutting till I can’t reach the scrap anymore. Could I write that class to cover all the use cases without a test for each one? Probably, but it’s definitely going to be harder.
I write tests because they force me to think about the problem, to break it down to testable pieces, and figure out how to keep it testable. I’m not trying to combine the “what” and “how” in the same thought. Tests also force me to implement that code in a way that is testable –and testable code is (typically) easy to change, easy to diagnose, easy to plug into different use cases.
The beauty of arrange/act/assert, especially in an xUnit style test with a minimum of shared setup, is each test is explicit about how the system under test behaves under different conditions. If you’re confirming that a specific result happens based on configuration, you have to put that value in the test. If the data context has to return a specific value, you have to specify it in the test where it’s visible to anyone.
Which means that a year from now, when you need to update a switch statement — you already know what all the values correspond to, without looking up the requirements doc from two projects ago. You won’t need to spend as much time explaining the code to someone — the tests lay it all out, in all the variability. The dependencies are documented, the expected behavior of dependencies is laid out.
Recently (and this has happened more than a few times) I went to the tests as the first stop in a bug squash. I quickly realized that none of the tests covered the scenario where the main dependency throws a null reference exception; as a result, the code was just logging the exception and returning an inappropriate value. I was able to write a test that replicated the situation, and then put a bug fix in without ever actually debugging the app.
Having up-to-date tests is like having up-to-date documentation. You don’t need to debug the application to figure out what it’s doing under the hood, you already know by following the story told by the tests.
And When Coupled With Test Driving…
So all of that above is primarily based on experiences I’ve had trying to wrap legacy code, and it boils down to “Tests help me understand the code so I can improve it safely.”
But…what if you were able to avoid the whole “this needs redesigned before we can add the feature” part? What if I told you there was a way to build that same quality into your code, right from the start?
This is the obligatory TDD plug. I don’t want to harp on it — I love TDD, and even I can’t stand most of the TDD missionaries out there — but again, I view the tests the same way I view configuring the IDE, using git aliases, customizing my PowerShell profile, using Resharper. It’s a tool that allows me to work with the code in a way that drives Quality. Writing a test for an empty service class is going to keep the problem I’m trying to solve very small. And if I’m trying to solve a small problem by “using the the least amount of code to make the test pass,” I’m far less likely to over-engineer a situation. This keeps my code lightweight, flexible, and simple.
As the problems become more complicated, in the “Only update the database if these 3 conditions are true and also it’s Tuesday” vein, so does my code…but incrementally, and in a way that doesn’t break previous passing tests. I’m already avoiding regressions and we’ve never deployed this code. My code is only as complex as it needs to be (if I stay disciplined), and the fewer moving parts the fewer things that can spawn bugs.
Tests, whether before or after writing your prod code, are going to drive Quality. I just prefer to be efficient and find out I’m making a mess before I commit any changes.
Wrapping It All Up
I’ve worked in shops with no automated tests. I’ve worked in TDD shops. I’ve worked in shops that half-assed testing. I’ve learned you do not need tests to write and change working code, but that tests make the job infinitely easier. And when delivering software isn’t an absolute struggle, I write far better code.
Code that can’t be tested without a lot of work is smelly. Writing tests in that case is like opening the refridgerator door — without opening that door you never smell the fact that last week’s leftovers are ready for the trash. Tests, if nothing else, tell the story of how your code is supposed to function — far better than writing a README or walking someone through the entire application.
These two items are the things I’ve come to appreciate about automated tests far more than the pat “I know the code works because tests.”