[racket] Syntax classes as first-class values

From: Alexander D. Knauth (alexander at knauth.org)
Date: Tue Dec 2 17:35:50 EST 2014

I got to thinking about this, and I know that they can’t work as purely first-class values, because syntax-parse needs to know the attributes of a syntax-class at compile time, but would it be possible to have a system sort of like in racket/unit where units can be first-class values but you can attach the compile-time information like attributes to them with a form like define-unit-binding?  Or would that not work for some other reason?

On Oct 21, 2014, at 3:04 PM, Jack Firth <jackhfirth at gmail.com> wrote:

> With syntax/parse, would it be possible to make syntax classes first class values instead of just a macro extension on top of syntax parse? That way you could parameterize syntax classes over arbitrary values, including other syntax classes, to create composable syntax classes. For instance, I find it very common when writing complex macros that I need to represent a series of optional keyword clauses like so:
> 
> (define-syntax foo
>   (syntax-parser
>     [(_ req1 req2
>          (~optional (~or (~seq #:kw1 op1)
>                                  (~seq #:kw2 op2)
>                                  (~seq #:kw3 op3)))
>          ...)
>      ; do something ]))
> 
> But this allows for items to be repeated and can be very verbose if the clauses are complex. I can define a syntax class for each of the optional clauses:
> 
> (define-splicing-syntax-class op1
>   (pattern (~seq #:kw1 op1-pattern)))
> 
> (define-splicing-syntax-class op2
>   ...
> ...
> 
> I'd like to also be able to define a function that takes those syntax classes and returns a new syntax class representing an optional series of any of those classes where each can be given once:
> 
> (define (optional-sequence-of class-a class-b class-c)
>   (splicing-syntax-class
>     (pattern (~optional (~or (~once (~var a class-a))
>                                           (~once  (~var b class-b))
>                                           (~once (~var c class-c)))) ...)))
> 
> And then my original macro would just be:
> 
> (define-syntax foo
>   (syntax-parser
>     [(_ req1 req2 (~var optionals optional-sequence-of op1 op2 op3))
>     ; do something ]))
> 
> The reason I think this would be useful is that you could re-use the optional-sequence-of class in other macros since it's generalized. You can make some general "template" syntax classes for common macro tasks, such as a sequence of pairs of unbound unique identifiers with expressions (for let-like macros) or for a definition of a function's arguments (so Racket's definition of a function ((f a b) g) and Typed Racket's definition of a function ((f [a : Number] [b : Number]) [g : Number Number -> Number) could re-use a lot of the same macro plumbing).
> 
> I don't know how possible this is, or what issues would need to be ironed out to make it particularly useful, but I welcome any thoughts on the matter.
> ____________________
>  Racket Users list:
>  http://lists.racket-lang.org/users



Posted on the users mailing list.