Duplicate identifier in expansion

[samdphillips]

This program expands to a duplicate identifier error. Removing qi/list from the requires expands successfully.

#lang racket

(require qi qi/list)

(define ((get-ith v) i)
  (vector-ref v i))

(define-qi-syntax-rule (sum)
  (~> (-< (~> get-ith (as ref))
          (~> vector-length (as nrows)))
      (range 0 nrows)
      (map ref)
      (foldl + 0)))

(define x (for/vector ([i (in-range 10)]) (random)))
(~> (x) (sum))

let: duplicate identifier
  at: nrows
  in: (let ((nrows undefined) (ref undefined) (nrows undefined) (ref undefined) (nrows undefined) (ref undefined) (nrows undefined) (ref undefined)) (qi0->racket (thread (tee (thread (esc (#%host-expression get-ith)) (thread (esc (λ (ref2) (set! ref ref...

[samdphillips]

Here's a shorter illustration of the problem

#lang racket

(require qi qi/list)

(~> (0)
    (as x)
    (range 0 x)
    (map (+ 1)))

[samdphillips]

Observations

• bound-identifiers uses find-and-map/qi to traverse qi syntax.
• find-and-map/qi will not traverse #%host-expression forms.
• deforest-pass builds a esc form. An esc form normally expands to have a single #%host-expression parameter. Because deforest-pass runs after expansion these esc forms aren't annotated with #%host-expression
• modifying the bindings pass to not traverse into esc forms eliminates the duplicate identifier errors, but now creates identifier used out of context errors.

[countvajhula]

We discussed this on Discord and looked into it in detail in last week's meeting: Two Bugs for the Price of One. The short of it is, "Qi Enterprise Edition" is our best hope! For now, the (ugly) workaround is to use Racket bindings with qi/list, e.g.,:

(let ([x (void)])
  ((☯ (~> (esc
           (λ (v)
             (set! x v)))
          ⏚
          (range 0 x)
          (map (+ 1))))
   5))