With all-due respect to other philosophies than mine, that sounds like a nightmare. And I like tests. But everything has a cost, and I can't imagine ever getting anything significant done with such a straight-jacket, I'd spend all day tracking down weird compiler quirks.

I do a lot of cross platform stuff, and just having warnings as errors (a good idea) means a lot of build fixing because clang and MSVC disagree on things. If I were writing code in MSVC and I had to deal with code coverage reports from clang breaking the build, I would be paralyzed.

"Code quality" isn't a thing that can be trivially measured, and in my experience people that fixate on certain metrics tend to miss the forest for the trees.