Oct 9, 2023
You test for expected behavior against your public interfaces. You check happy path, error paths, edge cases like empty states, and so on.
One should be able to make a change to the internal implementation and have nothing break, and have no tests fail.
I have a dependency injection library called Factory. It contains, at the moment, about 112 tests with 100% code coverage.
Yet I can make an internal change and have nothing break, in which case I can be fairly confident that should I push that code to production none of my users should see an issue.
There have also been occasions where I've made a quick "improvement'" to the code and had a test fail. That's the "sensor" portion of your argument and I agree with that part.
