Hi! I’m setting up kover in a legacy project befor...

# test

[jean]

Hi! I’m setting up kover in a legacy project before I start breaking apart the spaghetti code. I some modules have 100% class coverage but the other metrics, method, branch, line are at 0% and instruction is at 1,2%. I didn’t find an explanation on what is measured to give those numbers. Are private functions took in account for the total of testable code for example? If yes, should I make them intern so I can test them?

[CLOVIS]

class = ok if at least one thing from the class has been called (including constructors, static initializers, etc) → mostly useless
method = ok if the method has been called
line = ok if the execution went through this line
branch = ok if the execution went through this line

For example:

if (foo) {
    println("a")
    println("b")
}

Here, there are 3 lines and 1 branch (the entire if)

foo?.bar ?: return "foo"

Here, there is 1 line and 3 branches (everything not null, foo is null, bar is null)

[CLOVIS]

Are private functions took in account for the total of testable code for example?

Yes

If yes, should I make them intern so I can test them?

No. If a test calls a public function, and that public function calls the private function, then the private function is marked as covered. If the private function is marked as uncovered, that means not a single test went through it, even through public functions. That either means you don't test the public functions, or the private function is dead code.

[jean]

Ok, thanks for the explanation! But it’s quite weird I get so low results then, I do test a good amount of code.

[CLOVIS]

Do you use mocks?

[jean]

Yes, we have a lot of them actually. The legacy codebase I work on has a very strong coupling between its components making it difficult to test without mocks

[CLOVIS]

I don't know your particular codebase, but a very common mistake people do with mocks is to mock everything.

Let's say you have the flow:
FooResource → FooService → FooRepository

People write a test for FooResource and mock FooService. That test only tests your serialization library, not any of your code! It's a very common thing.

[CLOVIS]

I've also seen people write (pseudo-code, I don't know which mocking lib you use)

test {
    mock fooService.create() → Foo("foo")

    assert fooService.create() == Foo("foo")
}

This tests literally none of your codebase, at all, but it's surprisingly common to find in mocked codebases

[jean]

We use mockk, so it’s similar to your example. I guess I need to focus on a particular test class first and see how I can improve it to get much better coverage. Then we can expand to the other classes. Thanks for the help!

[jean]

I see a lot of generated code included in the report (I’m focusing on generated classes for Room first). I’m trying to exclude them with this

reports.filters.excludes {

        annotatedBy("androidx.room.*")
        annotatedBy("*Composable")
        annotatedBy("*Preview")

        annotatedBy("*Generated")
        annotatedBy("org.junit.jupiter.api.Test")
    }

But it doesn’t work. All the class annotated with @Generated are still picked up. Do you know how I can prevent Kover of checking them?

[CLOVIS]

I thought this would have worked… Maybe ask in #CDLD3845Q or directly in the Kover GitHub?