4.30 Exercise 4.30

Consider the for-each procedure:

(define (for-each proc items)

(if (null? items)

'done

(begin (proc (car items))

(for-each proc (cdr items)))))

Ben is calling is like so, and claims that its side effects are performed:

(for-each (lambda (x) (newline) (display x))

(list 57 321 88))

He very well may be right, because these functions are not being passed side-effectful operation as arguments.

However, I would claim that this only works correctly given certain assumptions about how the newline and display functions are implemented in our evaluator. Remember that the only difference between our original evaluator and this one is that compound procedures are non-strict in their arguments. This leaves open the case where side-effectful operations could be passed as arguments to functions, in which case they well may not be forced. Consider the two following functions:

(define (p1 x)

(set! x (cons x '(2)))

x)

(define (p2 x)

(define (p e)

e

x)

(p (set! x (cons x '(2)))))

With our current evaluator, (p1 1) creates the cons cell (1 . 2) and (p2 1) does not. With Cy’s proposed change, they both do. This is because p1 evaluates its side effect as one of a sequence of expressions in the function definition, while with p2 the side-effecting computation binds to a new value e that is itself evaluated, but not forced, by the inner procedure p, leaving x unchanged.

Cy is right that, if the given for-each example works in the original evaluator, it will continue to work in his – if eval is enough to force a side effect, then actual-value will not change this. However, it would not be right to say that all for-each calls will be handled identically. Consider the following:

(define count 0)

(define (id x) (set! count (+ count 1)) x)

(define (perform x) x 'ok)

(for-each (lambda (x) (perform (id x))) '(1 2 3))

count

Under the original evaluator, count will still be 0, but with Cy’s change, it will be 3.

Laziness and side effects being mixed makes me very uncomfortable, but at the moment, I think Cy’s change might be more predictable for thinking about when (or whether) side effects occur.