Re: reserved-keyword in macro

[Damien Mattei <damien.mattei@gmail.com>]

oh :-O yes it is the behavior expected : assign the 2nd element of an array
T with value 1 :
prefix notation of : (<- ($bracket-apply$ T 2) 1) is equivalent in Curly
Infix syntax to : {T[2] <- 1}
as expected when $bracket-apply$ is not bound we have the good result.
Thanks for this result.
Of course in this case  {T[2] <- 1} it is not a problem to have not
$bracket-apply$
bound,
things became harder when doing : {T[2] <- T[4]} for example because then
i want to assign the result value of T[4] to T[2] and so i need to
evaluate T[4]
before which is exactly
($bracket-apply$ T 4) and at this point $bracket-apply$ is bind to:
(define-syntax $bracket-apply$
  (syntax-rules ()

    ((_ container index)
     ;(begin ;;(display "$bracket-apply$") (newline)
     (cond ((vector? container) (vector-ref container index))
  ((hash-table? container) (hash-table-ref container index))
  (else (array-ref container index))));)

    ((_ array index1 index2 ...)
     ;(begin ;;(display "$bracket-apply$") (newline)
     (if (vector? array)
(array-n-dim-ref array index1 index2 ...)
(array-ref array index1 index2 ...)))));)

and is used in the RHS (right-hand side) of:
{T[2] <- T[4]} which expand in Curly Infix SRFI-105 to:
(<- ($bracket-apply$ T 2) ($bracket-apply$ T 4))
and then is expanded in Scheme+ in two phases:
first the RHS expr is evaluate in :
((_ ($bracket-apply$ container index) expr)
     (let ((value expr)) ;; to avoid compute it twice
in : (array-ref T 4)
but the LHS (Left hand side) is not evaluated with $bracket-apply$ but with
the macro <-
in the body of let :
(cond ((vector? container) (vector-set! container index value))
which give :
(vector-set! T 2 value) where value is the value of expr, previously
expanded and evaluated.
And we get the good result.
But for this we need to have sometimes $bracket-apply$ as a bound macro (or
procedure) and sometimes not, being a reserved keyword NOT bound.
This for me obscure WHY the keyword in syntax-rules MUST not be bound to
behave correctly but this is like that in Scheme standarts and we have to
deal with.
I already faced this problem earlier and the solution is in the previously
commented code:
 ;; (if (equal? (quote $bracket-apply$) (quote funct-or-macro)) ;; test
funct-or-macro equal $bracket-apply$
which can only be understood knowing that the macro was in the past
declared like this and commented code does not match present code,here is
the previous definition of <- :
(syntax-rules ()

    ;;  special form like : (<- ($bracket-apply$ T 3) ($bracket-apply$ T 4))

    ;; one dimension array, example: {a[4] <- 7}
    ;; $bracket-apply$ of SRFI 105
    ((_ (funct-or-macro container index) expr)
           (let ((value expr)) ;; to avoid compute it twice

       ;; (if (equal? (quote $bracket-apply$) (quote funct-or-macro)) ;;
test funct-or-macro equal $bracket-apply$

       ;; normal case
       ;; {T[2] <- 4}
       ;; {T[3] <- T[2]}

       (cond ((vector? container) (vector-set! container index value))
                 ((hash-table? container) (hash-table-set! container index
value))
                 (else (array-set! container index value)))
       value))

so the solution will be to remove $bracket-apply$ as literal in:
(define-syntax <-
  (syntax-rules ($bracket-apply$)

and some check manually in the macro with:
(if (equal? (quote $bracket-apply$) (quote funct-or-macro)) ;; test
funct-or-macro equal $bracket-apply$
to branch instead of using pattern matching.
i will code this later but this was of great help,thanks again all.

Damien

On Thu, Feb 3, 2022 at 1:52 AM Vijay Marupudi <vijaymarupudi@gatech.edu>
wrote:

> Hi Damien,
>
> I tried to run the code you provided. I ran
>
> -----------------------------------------------------------------
>
> (define-syntax <-
>   (syntax-rules ($bracket-apply$)
>     ((_ ($bracket-apply$ container index) expr)
>      (let ((value expr)) ;; to avoid compute it twice
>        (cond ((vector? container) (vector-set! container index value))
>              ((hash-table? container) (hash-table-set! container index
> value))
>              (else (array-set! container index value)));)
>        value))
>     ((_ ($bracket-apply$ array index1 index2 ...) expr)
>      (let ((value expr))
>        (if (vector? array)
>            (array-n-dim-set! array value index1 index2 ...)
>            (array-set! array index1 index2 ... value));)
>        (newline)
>        value))
>     ((_ (var ...) expr)
>      (begin
>        (display expr) (newline)
>        (let ((expr-list (call-with-values (lambda () expr) list)))
>          (assign-var (var ...) expr-list)
>          expr-list)))
>     ((_ var expr)
>      (begin
>        (set! var expr)
>        var))
>     ((_ var var1 var2 ...)
>      (<- var (<- var1 var2 ...)))))
>
> (define T (make-vector 5))
> (<- ($bracket-apply$ T 2) 1)
>
> -----------------------------------------------------------------
>
> After I ran that, T was
>
> #(#<unspecified> #<unspecified> 1 #<unspecified> #<unspecified>)
>
> Is that was you are looking for?
>
> > "A literal matches an input expression if the input expression is an
> > identifier with the same name as the literal, and both are unbound13
> > <
> https://www.gnu.org/software/guile/manual/html_node/Syntax-Rules.html#FOOT13
> >.
> > " as $bracket-apply$ is already bind to a definition the pattern will
> > not be matched:
>
> It's possible, as in my case, I did not have it bound, and it seems to
> have worked the way you expected?
>
> ~ Vijay
>