I'm recently hired as part of a team at my work whose focus is on writing acceptance test suites for our company's 3D modeling software. We use an in-house C# framework for writing and running them which essentially amounts to subclassing the TestBase class and overriding the Test() method, where generally all the setup, testing, and teardown is done.
While writing my tests, I've noticed that a lot of my code ends up being boilerplate code I rewrite often. I've been interested in trying to extract that code to be reusable and make my code DRYer, but I've struggled to find a way to do so without overabstracting my tests when they should be largely self-contained. I've tried a number of approaches, but they've run into issues:
- Using inheritance: the most naive solution, this works well at first and lets me write tests quickly but has run into the usual trappings, i.e., test classes becoming too rigid and being unable to share my code across cousin subclasses, and logic being obfuscated within the inheritance tree. For instance, I have abstract RotateVolumeTest and TranslateVolumeTest classes that both inherit from ModifyVolumeTest, but I know relatively soon I'm going to want to rotate and translate a volume, so this is going to be a problem.
- Using composition through interfaces: this solves a lot of the problems with the previous approach, letting me reuse code flexibly for my tests, but it leads to a lot of abstraction and seeming 'class bloat' -- now I have IVolumeModifier, ISetsUp, etc., all of which make the code less clear in what it's actually doing in the test.
- Helper methods in a static utility class: this has been very helpful, especially for using a Factory pattern to generate the complex objects I need for tests quickly. However, it's felt 'icky' to put some methods in there that I know aren't actually very general, instead being used for a small subset of tests that need to share very specific code.
- Using a testing framework like xUnit.net or similar to share code through [SetUp] and [TearDown] methods in a test suite, generally all in the same class: I've strongly preferred this approach, as it offers the reusability I've wanted without abstracting away from the test code, but my team isn't interested in it; I've tried to show the potential benefits of adopting a framework like that for our tests, but the consensus seems to be that it's not worth the refactoring effort in rewriting our existing test classes. It's a valid point and I think it's unlikely I'll be able to convince them further, especially as a relatively new hire, so unless I want to make my test classes vastly different from the rest of the code base, this one's off the table.
- Copy and paste code where it's needed: this is the current approach we use, along with #3 and adding methods to TestBase. I know opinions differ on whether copy/paste coding is acceptable for test code where it of course isn't for production, but I feel that using this approach is going to make my tests much harder to maintain or change in the long run, as I now have N places I need to fix logic if a bug shows up (which plenty have already and only needed to be fixed in one).
At this point I'm really not sure what other options I have but to opt for #5, as much as it slows down my ability to write tests quickly or robustly, in order to stay consistent with the current code base. Any thoughts or input are very much appreciated.
I personally believe the most important thing for a successful testing framework is abstractions. Make it as easy as possible to write the test. The key points for me are that you will end up with more tests and the writer focuses more on what they are testing than how to write the test. Every testing framework I have seen that doesn't focus on abstraction has failed in more ways than one and ended up being maintainability nightmares.
If the logic is not used anywhere else but a single test class then leave in that test class but refactor later if it is needed in more than one place
I would opt in for all except #5.