I've never understood the purpose of significantly modifying a program in memory. Won't you just have to redo all your changes, this time to the source program,

I've never understood the purpose of significantly modifying a program in memory. Won't you just have to redo all your changes, this time to the source program, once you've settled on a design? [0] As long as you have fast, incremental compilation, doing it in memory first seems like a waste of time. It mostly seems useful for small, disposable changes, to explore while debugging or whatnot. (I suppose hacking on a process as it runs in production, or some other scenario where "halt, compile, restart" is impossible, is also a good reason, but personally I'd prefer saving process state across restarts, like XMonad does.) [0] Of course, that doesn't apply if you're building the program entirely in memory, such as Smalltalk with its image-based programming environment.