Even weirder... I've come to think code coverage may be related more to security / safety than correctness. People often think they're buying safety when they go from 80% coverage to 100% coverage, and I don't think that's always true. And while it's a laudable goal, there are probably better ways to confirm safety than assuring you have 100% code coverage.
We've noticed that you can still have all sorts of problems in code that has 100% coverage. We started using a TLA+ like "environment" to automagically prove assertions about our code (mostly things like "there's only one path that gets you to this counter := counter + 1 line and there's a mutex around it.") It's a pain in the rear, but probably less of a pain in the rear than trying to prove that sort of thing with traditional tests and an exhortation that you have to have 100% coverage.
We've noticed that you can still have all sorts of problems in code that has 100% coverage. We started using a TLA+ like "environment" to automagically prove assertions about our code (mostly things like "there's only one path that gets you to this counter := counter + 1 line and there's a mutex around it.") It's a pain in the rear, but probably less of a pain in the rear than trying to prove that sort of thing with traditional tests and an exhortation that you have to have 100% coverage.