Also, any reason not to add the tests upfront? I think it'd make the changes more obvious if the tests were added upfront or in a lockstep, and the implementation incrementally fixing the todo_wine.
Well, the whole test logic WRT actual samples will be just broken until the implementation is there and can probably only be skipped. So I considered that doing that upfront and updating the test on the way won't make most interesting parts clearer and will only add some back and force test tweaking along the patchset. Still if you think that is more convenient I can do that, should I?