How I write unit tests for existing code
I have a confession to make: even though I recognize the value of automated testing, I never got a hang of TDD, and all of my projects got their tests written after the fact. As a result, I had to come up with my own techniques for designing test suites. Here’s one: start with a test plan.
First, look at the signatures of your methods or functions1, and also at their docs. Think about actions they perform and relations between their inputs and outputs. Do not look at their implementations, because that’ll spoil your high-level outlook. At this stage, all you care about is descriptions. You should end up with something like this:
("strprintf() fills a string with provided parameters", "[strprintf]")
TEST_CASE{
("Empty format results in empty string")
SECTION{
}
("%i format is replaced by an integer")
SECTION{
}
("%s format is replaced by a string")
SECTION{
}
}
Now jump into low-level mode, and implement the actual checks. Still, do not look at method implementations. Try to get as many tests as possible to pass; comment out the rest.
After that, you’re finally allowed to look at methods’ code. Figure out why failing tests fail. A coding error? Wrong assumptions? Make all the tests pass.
Finally, look at the test coverage, and try to increase it as much as you can. Make an extra effort to ensure that your descriptions could be understood by someone who doesn’t know the implementation details.
There are a couple benefits to writing tests this way:
- Descriptions are (mostly) tied to methods’ purpose, not the way they’re coded.
- Separate “high-level” stage helps you notice interactions between methods (e.g. connection between serialization and deserialization).
- By the time you get to “low-level” stage, you already have a good understanding of what properties you’re testing and where, so it’s easier to write minimal tests that only check the essential. This promotes DRY code.
- I find it easier to do “high-level” and “low-level” thinking in separate passes, rather than constantly switching between the two.
Signatures contain name, input arguments (with types if your language supports that), and output type (if your language support that).↩︎
Your thoughts are welcome by email
(here’s why my blog doesn’t have a comments form)