I don't think historical precedent shows that the preferred form for modifying GPLed works is the full set of data contained within a revision control system - during the bitkeeper era you were unable to obtain most of the kernel metadata without agreeing to onerous license conditions, but nobody at the time seemed to argue that this was a GPL violation and I note that Debian didn't stop distributing the kernel during that period. Is there an argument to be made that the preferred form for modification includes individual patches if that's the way the maintainers work? I honestly don't know. Have we held anyone to that standard in the past? Absolutely not, and so I don't think it's terribly useful to start talking about "evading the GPL" when the result is equivalent to things that we've accepted as GPL-compliant in the past.