Luckily there were no hidden false negatives.

Turns out the tests were correctly configured in CMake.

posix.cfg and gnu.cfg both contain the mmap() function leading to different tests results depending on the order of the libraries.

I am not sure how this could be handled properly. It is properly correct but the gnu test should not depend on posix and there should possibly some indication that data was changed as it might not be intentional.

This is another reason to only include libraries if necessary - see https://trac.cppcheck.net/ticket/11287.