---------- Forwarded message ---------
From: Felix Thibault <xxxxxx@gmail.com>
Date: Sun, Aug 23, 2020 at 2:57 PM
Subject: Re: feature/underscorep branch with provisional changes to handle _
To: Adam Nelson <xxxxxx@nels.onl>


I tried using the guard procedures, today but I don't know how to troubleshoot the syntax-case macros. This is an the sort of setup I have now:

(cond-expand
  (r7rs
    (define-syntax match-prob
      (syntax-rules (_ ..1 ..* ..= @)
        ((match-prob v _ g+s (sk ...) fk i)
          (sk ... i))
        <the other patterns that can't be literals in match-two>
        ((match-prob . x)
         (match-two . x)))
     <same with match-extract-vars>)
  (r6rs
    ;guard functions
    (define (underscore? x )
      (and (identifier? x) (free-identifier=? x #'_)))
    <guards for other literals>
 (r6rs
   (syntax-case stx ()
   ((match-prob v p g+s (sk ...) fk i)
    (underscore? #'p)                        ;<-----This is where Loko chokes
    #'(sk ... i))                            ;       "Identifier out of context" 
   <other patterns that can't be literals>
  ((match-prob . x)
   (match-two . x))
  <same for match-extract-vars>))

and Chibi passes all tests, since it does the r7rs block, which is just the parts of match-two and match-extract-vars that r6rs can't handle, in the same form as before, but Loko won't load, as noted above. 

On Sat, Aug 22, 2020 at 2:56 PM Adam Nelson <xxxxxx@nels.onl> wrote:

I had to solve this same problem recently in SRFI 197:

https://github.com/ar-nelson/srfi-197/blob/draft-3/srfi-197-syntax-case.scm

I ended up with the same underscore? definition as you, but it doesn't necessarily need to be a macro; it works better as a procedure that can be called in a syntax-case fender. The definition of ellipsis? might be useful, too: in r6rs, you can get a reference to the ellipsis symbol with #'(... ...). I don't know if this is documented anywhere; I only discovered it in a post Marc made on a Guile mailing list: https://www.mail-archive.com/xxxxxx@gnu.org/msg14722.html

On 8/22/20 1:57 PM, Felix Thibault wrote:
It occurs to me that these changes convert:

invoke-macro-with-problematic-literal -> macro-with-problematic-literal

to

invoke-catch-problematic-literal -> catch-problematic-literal -> macro-without-problematic-literal

and if that is a valid approach, I could re-write it as

invoke-catch-underscore->catch-underscore->catch-dot-dot-one ... -> catch-at-sign -> macro-with-no-problematic-literals

On Fri, Aug 21, 2020 at 10:34 PM Felix Thibault <xxxxxx@gmail.com> wrote:
I added these macros:
(cond-expand
  (r7rs
    (define-syntax underscore?
      (syntax-rules (_)
        ((_ _ kt kf) kt)
        ((_ x kt kf) kf))))
  (r6rs
    (define-syntax underscore?
      (lambda (stx)
        (syntax-case stx ()
          ((_ x kt kf)
           (if (and (identifier? #'x)
               (free-identifier=? #'x #'_))
               #'kt
               #'kf)))))))


match-two and match-extract-vars are the two functions that have underscore literals so I replaced match-two's invocation with

(define-syntax match-underscore
  (syntax-rules ()
    ((match-underscore v p g+s (sk ...) fk i)
     (underscore? p
                  (sk ... i)
                  (match-two v p g+s (sk ...) fk i)))))

and all of match-extract-vars invocations with:

(define-syntax match-extract-underscore
  (syntax-rules ()
    ((match-extract-underscore p (k ...) i v)
     (underscore? p
                 (k ... v)
                 (match-extract-vars p (k ...) i v)))))

and removed the lines that matched _ from match-two and match-extract-vars.

Chibi and Loko are both passing.