podcast interview about mutation testing at Google for a variety of languages

Thank you for this, this is a strategy I'd not considered feasible against the explosion of test cases that an all-branches coverage report can cause (even if you pack together tests so that there's orthogonal coverage of two or more areas in one test case without an early failure clouding the validity of a later negative test result).

I think this conversation hides an elephant, too. There's a difference in approach between the London Style and the Detroit Style for an integrated test suite[1]. The London 'Extreme Tuesday Club' is an ongoing meetup founded in the 2000's to discuss Extreme Programming, while the wiki.c2.com community that sprung up around Kent Beck's development of Extreme Programming (and Agile) while working a project for Chrysler in Detroit.

London Style breaks the test scope into everything that has an interface -- and assumes a stable code base, possibly a practice where legacy interfaces don't change but implementation details might -- so there's scope to test all the viable code paths. (Plus you get to reject bad data when it's an unexpected use of the existing code.) There can be a reluctance to rework interfaces because you also have to keep a test suite 'always green' for no feature gain -- the stability of the code can ossify the project.

Detroit Style is a scaffold to a changing code base, or changing interfaces, so you have only relevant test code for the things you want to protect from regressions. Investing in mutation cases for a fluid code-base might not be worth the time, but the fluidity of the Linux code-base shows that change is ongoing and we need to run test cases to protect against regressions.

These approaches probably should overlap, but 'should is considered harmful' for the number of unchecked assumptions it brings to the conversation.

1: https://zone84.tech/architecture/london-and-detroit-schoo...

K3n.