There is an interesting parallel between Mr Carmack's "inlining" observations and one of the sessions I went to at Strange Loop this year. Jonathan Edwards was

There is an interesting parallel between Mr Carmack's "inlining" observations and one of the sessions I went to at Strange Loop this year. Jonathan Edwards was trying to beat back "callback hell" (i.e. unpredictable execution order leading to unpredictable side effects) by radically simplifying the control flow of his programs and keeping the execution model dirt-simple. Both would appear to argue that it's best to arrange heavily stateful code in a simple linear sequence, and use a top-down execution model, so that stateful effects are clear and easy to predict. That being said, I've seen procedures that followed this sort of approach that were thousands of lines long. Even if we could have cut down on the ridiculous number of conditionals in that code, most of the state at that scale asymptotically approaches an undifferentiated mass of global variables. The result is testable and maintainable only via heroic effort. There have got to be limits to this kind of approach. (For Mr Edwards the solution was to break the whole thing up into a sequence of composable views, or lenses, with the interface between each stage being well-defined.) I wonder to what extent Mr Carmack's pivot to pure functions is simply an acknowledgement that there were much better ways to refactor the code than the mess of one-timer procedures that probably seemed like a good idea the first time through...