Post

100% Code Coverage Doesn't Mean It Works

Tests are great. I like testing, tests, and talking about testing and tests.

I think code coverage is useful. It can identify gaps in testing, providing a “hey, maybe you should test over here” notion. But, 100% coverage doesn’t mean everything works perfectly. 100% test coverage doesn’t mean anything works properly. It means all the code was executed in some fashion. All statements were visited. Basically, the only thing it shows is the code didn’t unexpectedly throw an unhandled error.

No assertions, 100% coverage

Here’s an example of 100% code coverage with no asserts:

1
2
3
function isNumberSeven(number) {
  return number === 7;
}
1
2
3
4
5
6
// Jasmine
describe('isNumberSeven tests', () => {
  it('runs', () => {
    isNumberSeven();
  });
});

Todd Gardner refers to this as “Assertion-Free Testing” in his Software Testing for Failed Projects talk at NDC Oslo 2016.

Assertions! And, more tests!

Using assertions (and, more tests) dramatically improves the quality of the test suite.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
// Jasmine
describe('isNumberSeven tests', () => {
  it('returns true for 7', () => {
    expect(isNumberSeven(7)).toBe(true);
  });

  it('returns false for 6', () => {
    expect(isNumberSeven(6)).toBe(false);
  });

  it('returns false for 8', () => {
    expect(isNumberSeven(8)).toBe(false);
  });
});

Use code coverage, not because you “have to”

Code coverage can be a useful tool to show trends and identify potential gaps in testing. But, using it as a blind metric could be harmful. If you’re mandated to abide by a certain percentage of coverage, you can technically achieve the goal without getting the value you’d expect out of good, well-authored tests.

This post is licensed under CC BY 4.0 by the author.