[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Choose-Your-Own-Ellipsis

This page is part of the web mail archives of SRFI 46 from before July 7th, 2015. The new archives for SRFI 46 contain all messages, not just those from before July 7th, 2015.

On Monday, Oct 13, 2003, at 10:43 US/Eastern, Allophone Petrofsky wrote:

I don't know what the best exact syntax for this extension would be.
Here are several possibilities (using the conventions foo* == any
number of foo, and foo? == zero or one foo):

  -- Ellipsis choice before the literals, as above:
    (syntax-rules <ellipsis>? <literals> <rule>*)

  -- After the literals:
    (syntax-rules <literals> <ellipsis>? <rule>*)

  -- On a per-rule basis:
    (syntax-rules literals (<ellipsis>? <pattern> <template>) ...)

  -- After the literals, with an extra "noise word" (like cond's =>)
     to distinguish this from other syntax-rules extensions:
    (syntax-rules <literals> (ellipsis <ellipsis>)? <rule>*)

I like this idea most.  This lets me do some other macrology that
would have been rather irritating, because you no longer need to pass
the 'free pattern variables' in submacros to avoid accidental ellipsis
conflict.  I'll modify the draft to define this, but we shall first
to decide which of those four options to choose: opinions, anyone?  (I
don't care either way.)

This _would_ also eliminate the issue of matching ellipsis literally
but for one problem: we have no guards or non-linear patterns, so we
can't do something like:

(define-syntax contrived-example
  (syntax-rules ()
    ((_ (syntax-rules ?ellipsis ?literals
                   ;; ^^^^^^^^^
          ((?ignored ?foo ?ellipsis) ?template)))
                       ;; ^^^^^^^^^
     (display "You've got ellipsis!"))
    ((_ ?foo)
     (display "No ellipsis yet."))))

and have

(contrived-example (syntax-rules ::: () ((_ ?bar :::) baz)))

expand to

(display "You've got ellipsis!")

Having mentioned non-linear patterns, what are some opinions on adding
them, and what their specific semantics should be?  If I do add non-
linear patterns, should SYNTAX=? defined as

(define-syntax syntax=?
  (syntax-rules ()
    ((_ ?sk ?fk ?x ?x) ?sk)
    ((_ ?sk ?fk ?x ?y) ?fk)))

compare identifiers boundly, freely, or symbolically?

What are some thoughts on non-linear patterns and guards, or, if
guards, how they might be best added to SYNTAX-RULES?