# Fun With Symbolic Derivatives in Lisp.

## Table of Contents

## Code as Data

I don't often get to use lisp at work. Indeed, best chance I have to actually code lisp is if I decide to automate something in emacs. Though, even then I'm likely to just use an org-mode buffer as what people are calling notebooks nowadays.

That said, I do want to take a stab at writing something explaining what the lisp community means when they say Code as Data.

## The basic claim

The basic claim is simply that you can treat code as data. This doesn't sound as impressive to folks that are used to "eval" in languages like JavaScript. In large, the basic examples using eval don't help. Lets look at a function that simply performs some math and how we can construct it using eval, as well.

First, the straight forward way of defining the function.

(defun playing/function (x) (+ (* x 2) 12))

playing/function

Nothing fancy. You can call this in the standard way:

(playing/function 12)

36

Now, lets see how we could split the body from the function header.

(defvar playing/body '(+ (* x 2) 12))

playing/body

At face value, we just have a new variable "playing/body". What does it look like?

playing/body

(+ (* x 2) 12)

What can we do with this, then? Well, we could just append it to the require syntax and evaluate it to get our function. Let us try that.

(eval (append '(defun playing/eval-function (x)) (list playing/body)))

playing/eval-function

Just to confirm, we can call this in the same way as we had previously.

(playing/eval-function 12)

36

## But isn't this the same as any language with "eval"?

An obvious and straight forward objection. I presume this typically means doing like the following in JavaScript.

function example(x) { return x * 2 + 12; } console.log(example) console.log(example(12)) var body = "return x * 2 + 12;" console.log(body) eval("function eval_example(x) {" + body + "}"); console.log(eval_example) console.log(eval_example(12))

[Function: example] 36 return x * 2 + 12; [Function: eval_example] 36

As we can see, it is possible to have a variable that stores the "body" of a function separately from the actual definition of it. Yes, there was some funky string concatenation overhead, but the spirit of the argument is that this is not fundamentally different.

## So, what makes the lisp example different?

Rather than just talk about the differences, lets have some fun.
Recently, I linked someone the section on symbolic derivatives in lisp
from the original SICP lectures.^{1}

For brevity, I'm just going to quickly run through all of the functions the SICP defined. I keep the "playing/" prefix to keep from littering my current emacs session, and I obviously port the functions to elisp, from scheme.

First, the primitives used.

(defun playing/variable? (x) (symbolp x)) (defun playing/same-variable? (v1 v2) (and (playing/variable? v1) (playing/variable? v2) (eq v1 v2))) (defun playing/make-sum (a1 a2) (list '+ a1 a2)) (defun playing/sum? (x) (and (listp x) (eq (car x) '+))) (defun playing/addend (s) (cadr s)) (defun playing/augend (s) (caddr s)) (defun playing/make-product (m1 m2) (list '* m1 m2)) (defun playing/product? (x) (and (listp x) (eq (car x) '*))) (defun playing/multiplier (p) (cadr p)) (defun playing/multiplicand (p) (caddr p))

Then, the derivative function.

(defun playing/deriv (exp var) (cond ((numberp exp) 0) ((playing/variable? exp) (if (playing/same-variable? exp var) 1 0)) ((playing/sum? exp) (playing/make-sum (playing/deriv (playing/addend exp) var) (playing/deriv (playing/augend exp) var))) ((playing/product? exp) (playing/make-sum (playing/make-product (playing/multiplier exp) (playing/deriv (playing/multiplicand exp) var)) (playing/make-product (playing/deriv (playing/multiplier exp) var) (playing/multiplicand exp)))) (t (error "unknown expression type -- DERIV" exp))))

Now, lets see if it works.

(playing/deriv playing/body 'x)

(+ (+ (* x 0) (* 1 2)) 0)

Not at all reduced form. And, for fun, the SICP section goes over how to fix this at a first order. So, lets quickly see that here.

(defun playing/make-sum (a1 a2) (cond ((playing/=number? a1 0) a2) ((playing/=number? a2 0) a1) ((and (numberp a1) (numberp a2)) (+ a1 a2)) (t (list '+ a1 a2)))) (defun playing/make-product (m1 m2) (cond ((or (playing/=number? m1 0) (playing/=number? m2 0)) 0) ((playing/=number? m1 1) m2) ((playing/=number? m2 1) m1) ((and (numberp m1) (numberp m2)) (* m1 m2)) (t (list '* m1 m2)))) (defun playing/=number? (exp num) (and (numberp exp) (= exp num)))

With that, lets try the deriv function one more time on the original. Showing the definition of "playing/body" to remind us what it was.

(playing/deriv playing/body 'x)

2

In case you forgot what the original was, here it is again.

playing/body

(+ (* x 2) 12)

## But who cares, I'm sure you could write a deriv function for javascript.

Probably, but consider the above a bit longer. Before, we showed that you could use eval to just stitch together the body straight to the function. But, since we didn't do anything to the body, it was natural to think this is akin to string concatenation into eval. But, we could also easily use eval to create a symbolic derivative.

For example:

(eval (append '(defun playing/eval-function-derivative (x)) (list (playing/deriv playing/body 'x))))

playing/eval-function-derivative

Calling this will give us the results we expect.

(mapcar #'playing/eval-function-derivative '(1 2 3 4 5 6 7 8 9 10))

2 | 2 | 2 | 2 | 2 | 2 | 2 | 2 | 2 | 2 |

Which… is just a constant, and kind of boring. Lets see if it can deriv something a little more interesting.

(eval (append '(defun playing/eval-function-derivative-2 (x)) (list (playing/deriv '(* x x) 'x))))

playing/eval-function-derivative-2

Looking at our values now, we see:

(mapcar #'playing/eval-function-derivative-2 '(1 2 3 4 5 6 7 8 9 10))

2 | 4 | 6 | 8 | 10 | 12 | 14 | 16 | 18 | 20 |

Still not exactly an interesting function, but quite clear that this is no longer a constant.

## Note all that was missing.

At no point did we have to write a parser. At no point did we have to really worry about stitching together syntax into the code we created. At no point did we have to use a special macro syntax, even.

Instead, we could simply treat the body of our function as any other data element in our code, and we were able to write a symbolic deriv function that was capable of not just telling us the derivative, but doing so in a way that we could turn into an executable function fairly easily.

Note also, that if there was something "not function like" in our function body, it wouldn't just happily execute the malicious code, but would error out.

(condition-case err (eval (append '(defun playing/eval-function-derivative-unsafe (x)) (list (playing/deriv '(eval "malicious") 'x)))) (error ;; we have to trap the error to show it in the output (concat "We got an error: " (error-message-string err))))

"We got an error: Symbolâ€™s function definition is void: playing/deriv"

Does this mean eval is safe? No. Please don't take it as that. Just realize that the eval of lisp is a lot more powerful because of how much more you can do with the data that you put into an eval. It is not just some opaque string that gets to enjoy all of the benefits of your language. It is a first class list of elements that you can inspect and have fun with.

## Further reading.

Please don't let the hasty treatment of the SICP lectures I did above
prevent you from reading that book. I also got a great deal of fun
out of watching the videos.^{2}

Similarly, don't get scared away from lisp just because I chose to use elisp. I'll confess I just picked elisp because I didn't want to install anything on the machine I'm currently on. Tried to hammer this page out without getting sucked into a rabbit hole of caring how my machine was setup.