yes, the atomicity argument is the strongest one to me as git submodules are a genuine pain to work with and still don't solve the problem well. Yet a clear transaction around a change affecting multiple components gets even more important as the rate of activity and associated contention rises.