I don't like lumping "add test & pass" together. It's very easy to accidentally write a test that doesn't test what you think it tests, and actually passes eve

I don't like lumping "add test & pass" together. It's very easy to accidentally write a test that doesn't test what you think it tests, and actually passes even without the "fix" applied. I do think it's very important to verify that the test fails without the associated change applied. This breaks `git bisect`, so maybe they shouldn't be separate commits in the public commit log, but they should be separate commits while you're working on it.