[necovek]

I like to differentiate between coverage by lines and semantic coverage: sometimes you need to exercise a single line multiple times to get full semantic coverage, and better semantic coverage usually beats larger line coverage for detecting problems and preventing regressions.

[eithed]

I think mutation testing helps in defining semantic coverage, if I understand what you're saying

[necovek]

Yes, mutation testing and similar techniques like fuzzing can help with it, but sometimes you want to be more deterministic: there are usually a lot of hidden side effects that are not obvious, and probably a source of majority of software bugs today.

Eg. something as simple as

  function foo(url) {
     data = fetch(url);
     return data;
  }

has a bunch of exceptions that can happen in fetch() not covered with your tests, yet you can get 100% line coverage with a single deterministic happy path test.

Basically, any non-functional side-effect behavior is a source of semantic "hiding" when measuring test coverage by lines being executed, and there is usually a lot of it (logging is another common source of side-effects not tested well). Some languages can handle this better with their typing approaches, but ultimately, there will be things that can behave differently depending on external circumstances (even bit flips, OOMs, full disk...) that were not planned for and do not flow from the code itself.