Fix sequence-ref to avoid unnecessary lookahead consumption

[NoahStoryM]

Checklist

• Bugfix
• Feature
• tests included
• documentation

Description of change

Fix sequence-ref to stop consuming extra elements from stateful sequences.

Problem

The current implementation of sequence-ref uses #:unless (j . < . i) which causes the loop to continue after reaching the target index, leading to unnecessary consumption of an additional element from the sequence. See also #5338.

Example

> (define v 0)
> (define s
    (make-do-sequence
     (λ ()
       (define (pos->element _) v)
       (define (continue-with-pos? _) (set! v (add1 v)) (< v 100))
       (values pos->element void (void) continue-with-pos? #f #f))))
> (define-values (more? get) (sequence-generate s))
> (displayln (get))
1
> (displayln (sequence-ref s 0))
2
> (displayln (sequence-ref s 0))
4
> (displayln (sequence-ref s 1))
7
> (displayln (sequence-ref s 1))
10
> (displayln (get))
12
> (displayln (sequence-length s))
87

[sorawee]

I could be missing something, but does this accomplish what you want?

(define (ref s i)
  (let ([v (for/first ([v (in-values*-sequence s)]
                       [j (in-range (add1 i))]
                       #:unless (j . < . i))
             (or v '(#f)))])
    (cond
      [(not v)
       (raise-arguments-error
        'sequence-ref
        "sequence ended before index"
        "index" i
        "sequence" s)]
      [(list? v) (apply values v)]
      [else v])))

[NoahStoryM]

I think this version looks correct, and it should behave as intended.
I’m just not sure whether this approach is better compared to the sequence-generate version.

Since this one matches the behavior of the original implementation, I’ll go ahead and use it for now.

[sorawee]

An issue with the (for ([_ (in-range i)] ...) + sequence-generate revisions above is that it's O(i) rather than O(min(i, |s|)). Maybe there's a way to make them work, but I haven't tried.

[sorawee]

Is this good to merge for the upcoming release? @mflatt