There are many situations where the extra constraints on pure functions are very useful. Pure functions can't segfault. They can't overwrite memory used by other parts of your program due to stray pointer references. They can't phone home to Google. Verification and optimization is often much simpler when you know that some functions are pure. Reactive programming frameworks that execute callbacks in response to events can cause really weird bugs if the callback mutates state somewhere. I could go on and on. Nobody is arguing for pure functions all the way down. A program's top-level functions will usually be side-effecting, and then you can call out to pure code any time. In practice, a surprising amount of application logic can end up being pure functions.