I've been working through SICP1 again. Its really a great book – and I need to say more about it on here sometime.

I am currently on Exercise 1.11, which centers around writing two separate functions that both compute a mathematical function. The first function should compute in a recursive way and the second should compute in a iterative way.

The point of writing the function in these different ways is to highlight the fact that iterative formulations are fast and take up significantly less memory; in fact, they can be computed in place/in constant space.

While I was wrote these, it hit me: It's interesting to create these functions, but it would be even more interesting to show that the computations take up the amount of space we have predicted. So, I set out to see if I could figure out how to get current size of the stack, and then to watch the stack change during the course of both executions.

As it turns out, it doesn't seem like Racket has the direct concept of a stack. This doesn't surprise me, because I understand that the concept of a stack is a little bit less clear-cut in languages that support continuations & tail call optimization.

However, I was able to find a method vector-set-performance-stats! that will provide "the number of bytes currently in use for the thread's continuation". Since a continuation represents the remaining work that a thread needs to perform as part of its computation, it seems like this would map fairly directly to the idea of the stack size showing how each function grows in space during the computation. If the recursive continuation size continually grows, and the iterative continuation size stays constant, that would be strong evidence that this would be answering my question.

Measurements of the stack size performed as predicted. During the course of the execution, the recursive continuation size kept growing. Surprisingly, the iterative version kept precisely the same. I wasn't certain at the time that this would be the case, but to me, this clearly is a success.

Ok, so how we measure the size of the continuation? The function get-continuation-size below finds the information for us.

(define (get-continuation-size)
  (let ((results (make-vector 20)))
    (vector-set-performance-stats! results (current-thread))
    ;; item 3 contains continuation size
    (vector-ref results 3)))

I'm going to keep using this code to show exactly the same thing – that functions are executing within the space we expect them to.