问题
I am trying to build the built-in procedure build-list in Racket.
The built-in function works like this:
(build-list 10 (lambda (x) (* x x)))
>> '(0 1 4 9 16 25 36 49 64 81)
My implementation is a recursive definition for a recursive procedure:
(define (my-build-list-recur list-len proc)
(if (= list-len 0)
'()
(cons (proc (sub1 list-len)) (my-build-list-recur (sub1 list-len) proc))))
When I call my implementation, I have:
(my-build-list-recur 10 (lambda (x) (* x x)))
>> '(81 64 49 36 25 16 9 4 1 0)
As you might have seen, I get the same result, but in a reverse order.
What can I do to have the result in the same order as the native function?
P.S.: I have done an implementation using a recursive definition for an iterative procedure which works perfectly. I am struggling now to generate the same result with the totally recursive procedure. I already know how to solve this doubt with long tail recursion.
This is my implementation with long tail recursion:
(define (my-build-list list-len proc)
(define (iter list-len accu n)
(if (= (length accu) list-len)
(reverse accu)
(iter list-len (cons (proc n) accu) (add1 n))))
;(trace iter)
(iter list-len '() 0))
回答1:
Ok so you're looking for an answer that does not use state variables and a tail call. You want for a recursive procedure that also evolves a recursive process. Not sure why you want this other than just to see how the definition would differ. You should also read about tail recursion modulo cons (here, and on wikipedia) – it's relevant to this question.
;; recursive procedure, recursive process
(define (build-list n f)
(define (aux m)
(if (equal? m n)
empty
(cons (f m) (aux (add1 m)))))
(aux 0))
(build-list 5 (λ (x) (* x x)))
;; => '(0 1 4 9 16)
Notice how the aux call is no longer in tail position – ie, cons cannot finish evaluating until it has evaluated the aux call in its arguments. The process will look something like this, evolving on the stack:
(cons (f 0) ...)
(cons (f 0) (cons (f 1) ...))
(cons (f 0) (cons (f 1) (cons (f 2) ...)))
(cons (f 0) (cons (f 1) (cons (f 2) (cons (f 3) ...))))
(cons (f 0) (cons (f 1) (cons (f 2) (cons (f 3) (cons (f 4) ...)))))
(cons (f 0) (cons (f 1) (cons (f 2) (cons (f 3) (cons (f 4) empty)))))
(cons (f 0) (cons (f 1) (cons (f 2) (cons (f 3) (cons (f 4) '())))))
(cons (f 0) (cons (f 1) (cons (f 2) (cons (f 3) '(16)))))
(cons (f 0) (cons (f 1) (cons (f 2) '(9 16))))
(cons (f 0) (cons (f 1) '(4 9 16)))
(cons (f 0) '(1 4 9 16))
'(0 1 4 9 16)
You'll see that the cons calls are left hanging open until ... is filled in. And the last ... isn't filled in with empty until m is equal to n.
If you don't like the inner aux procedure, you can use a default parameter, but this does leak some of the private API to the public API. Maybe it's useful to you and/or maybe you don't really care.
;; recursive procedure, recursive process
(define (build-list n f (m 0))
(if (equal? m n)
'()
(cons (f m) (build-list n f (add1 m)))))
;; still only apply build-list with 2 arguments
(build-list 5 (lambda (x) (* x x)))
;; => '(0 1 4 9 16)
;; if a user wanted, they could start `m` at a different initial value
;; this is what i mean by "leaked" private API
(build-list 5 (lambda (x) (* x x) 3)
;; => '(9 16)Stack-safe implementations
Why you'd specifically want a recursive process (one which grows the stack) is strange, imo, especially considering how easy it is to write a stack-safe build-list procedure which doesn't grow the stack. Here's some recursive procedures with a linear iterative processes.
The first one is extremely simple but does leak a little bit of private API using the acc parameter. You could easily fix this using an aux procedure like we did in the first solution.
;; recursive procedure, iterative process
(define (build-list n f (acc empty))
(if (equal? 0 n)
acc
(build-list (sub1 n) f (cons (f (sub1 n)) acc))))
(build-list 5 (λ (x) (* x x)))
;; => '(0 1 4 9 16)
Check out the evolved process
(cons (f 4) empty)
(cons (f 3) '(16))
(cons (f 2) '(9 16))
(cons (f 1) '(4 9 16))
(cons (f 0) '(1 4 9 16))
;; => '(0 1 4 9 16)
This is insanely better because it can constantly reuse one stack frame until the entire list is built. As an added advantage, we don't need to keep a counter that goes from 0 up to n. Instead, we build the list backwards and count from n-1 to 0.
Lastly, here's another recursive procedure that evolves a linear iterative process. It utilizes a named-let and continuation passing style. The loop helps prevent leaking the API this time.
;; recursive procedure, iterative process
(define (build-list n f)
(let loop ((m 0) (k identity))
(if (equal? n m)
(k empty)
(loop (add1 m) (λ (rest) (k (cons (f m) rest)))))))
(build-list 5 (λ (x) (* x x)))
;; => '(0 1 4 9 16)
It cleans up a little tho if you use compose and curry:
;; recursive procedure, iterative process
(define (build-list n f)
(let loop ((m 0) (k identity))
(if (equal? n m)
(k empty)
(loop (add1 m) (compose k (curry cons (f m)))))))
(build-list 5 (λ (x) (* x x)))
;; => '(0 1 4 9 16)
The process evolved from this procedure is slightly different, but you'll notice that it also doesn't grow the stack, creating a sequence of nested lambdas on the heap instead. So this would be sufficient for sufficiently large values of n:
(loop 0 identity) ; k0
(loop 1 (λ (x) (k0 (cons (f 0) x))) ; k1
(loop 2 (λ (x) (k1 (cons (f 1) x))) ; k2
(loop 3 (λ (x) (k2 (cons (f 2) x))) ; k3
(loop 4 (λ (x) (k3 (cons (f 3) x))) ; k4
(loop 5 (λ (x) (k4 (cons (f 4) x))) ; k5
(k5 empty)
(k4 (cons 16 empty))
(k3 (cons 9 '(16)))
(k2 (cons 4 '(9 16)))
(k1 (cons 1 '(4 9 16)))
(k0 (cons 0 '(1 4 9 16)))
(identity '(0 1 4 9 16))
'(0 1 4 9 16)
来源：https://stackoverflow.com/questions/40641470/building-the-built-in-procedure-build-list-in-racket