f(x) =
// ...
f(x + 1)
It's called like this: f =
x = x + 1
f()
x = x - 1
The return address from the function call references the code that undoes modification to global state (all locals and parameters have static storage when there is no recursion support), rather than relying on copies of parameters being made as in a modern language.The second is an extension of the first, using goto and labels instead of function calls, but it only works if you don't need to distinguish between different call sites. If the recursive function only ever calls itself in one place, and the mutation it needs to make in order to "recurse" is always the same (so it can be consistently undone), then the call stack can be eliminated too.
However, in modern systems (or basically anything you'll ever see today, including the cheapest microcontrollers), the stack doesn't just keep track of where to return to after each function completes. It also stores all the local variables used by the function. This is vital because it allows the variables to be restored after you return from a 'deeper' function. A convenient side effect is that when the function completes and the frame is popped off the stack, the references to the local references are destroyed and the data can be garbage-collected.
In this sense, local variables are directly related to stack frames. If your system doesn't store local variables in the stack frame, it almost certainly doesn't store them anywhere else (stack frames are the natural place for them), and you've only got globals to work with.
But his approach does in fact conserve stack space.
Also, by "return stack", he isn't referring to the mechanism for returning values, he is referring to the mechanism for knowing which address to jump (return) to when a function is finished. In x86, this is still done with a stack, but the stack modifications are done "behind the scenes" by the call/ret instructions.
When you pass function parameters you usually need to push onto the stack or risk overwriting the value of the named variable in the current execution context, since it has the same name or register in the new execution context.
Again, mildly educated answer so if somebody can correct me, please do.
I think you meant last-in-first-out.