Indeed. I'm not a fan of 100% code coverage tests in any language, but it's especially apparent in Haskell where, as you said, if code compiles (and you keep most of your functions pure), there are whole classes of bugs that simply cannot exist.