Unit tests are the developers number one safety net. Let that sink in. This is the number one reason for writing unit tests.
Unit tests are written by developers for developers to ensure that the code works as expected and handles happy and sad paths correctly. With enough unit test coverage the tests enable a safe environment for refactoring and rewriting code.
Unit test scope
Unit test should test a single thing, a method or function call and it should test only one use case within. In other words a unit test should test a function with a single input. This is a important guideline to understand. When a unit test tests a function with single input it makes the test isolated, repeatable and predictable.
Example of good tests:
@Test
fun findsAddress() {
val address = findAddress("Stevens street 35", "Southport", "Australia")
assertThat(address).isNotNull()
}
@Test
fun doesNotFindAddress() {
val address = findAddress("Stevens street 697", "Southport", "Australia")
assertThat(address).isNull()
}
If a test contains multiple inputs for a single function it's not anymore isolated. Say a test has seven different inputs and it calls the testable function seven times but with the third input it fails. Because it fails with the third input the remaining four inputs are not tested. Now you have in your hands a test that is proven to work for the first two inputs and fail for the third but you have no idea whether it works with the remaining four inputs or not. When you start to fix the implementation your safety net can fail for seven different inputs and if any of them fails you can never be sure if the remaining inputs work or not.
Example of a bad test:
@Test
fun findAddress() {
val validAddress = findAddress("Stevens street 35", "Southport", "Australia")
assertThat(validAddress).isNotNull()
val invalidAddress = findAddress("Stevens street 697", "Southport", "Australia")
assertThat(invalidAddress).isNull()
}
Writing a separate test case for each input surely adds the number of lines of code but it's not a bad thing when it also gives a better safety net when the code needs to be changed.
Exception to the rule
Of course there's a exception to the rule and in this case there are at least two, first one is called parameterized tests and second one is called property based testing.
Example of a parameterized test:
@ParameterizedTest
@ValueSource(ints = {1, 35, 50})
fun findsAddresses(streetNumber: Int) {
val address = findAddress("Stevens street $streetNumber", "Southport", "Australia")
assertThat(address).isNotNull()
}
Test logic and logic in tests
Test should be testing the logic of the implementation, they should not introduce any logic itself.
I've seen multiple times a test suite that has a bunch of logic in it, evaluations and conditionals that either manipulate the inputs, choose what functions to call or choose what assertions should be evaluated.
The problem with this is that the logic in tests itself introduces logic with various inputs and outputs and edge cases that themselves can introduce unexpected behaviour and the logic of the tests is not itself tested and verified in any way.
Avoid introducing logic just for tests. It adds complexity and possibility of invalid test functionality. Keep the unit tests as simple as possible.
Writing unit tests
I do try to write tests with TDD approach but I don't find it natural in all situations and in those situations I mix it up and write some of tests after I've written the initial implementation. I personally don't have a strong opinion on when and how you should write your tests as long as the tests are written and they cover the expected use cases and exception cases before the code ends up in trunk/main. Experiment and find a way to write tests that suits you.
Next part
In the next part I'll be writing about another testing subject, integration testing.