From: William D Clinger
Date: Sat, 16 Jun 2007 21:43:39 -0400
Subject: [Formal] please make square brackets more useful

Submitter: William D Clinger
Issue type: Enhancement
Priority: Minor
Component: Lexical Syntax
Report version: 5.94
Summary: please make square brackets more useful

Full description of issue:

The addition of square brackets to Scheme's lexical
syntax has been controversial for several reasons.
Most obviously, it creates a redundant alternative
to Lisp's established lexical syntax for lists, for
the sole purpose of allowing programmers to express
their intentions through some subtly creative choice
of parenthetical styles.  Since all such choices are
equally valid under the proposed standard, we are
likely to have unproductive disputes concerning the
One True Way to choose between round and square
brackets.

These disputes have actually been going on for some
time, but the addition of square brackets to the
draft R6RS brought this dispute to the fore and
makes its resolution both urgent and necessary.

Jonathan Rees has proposed an elegant solution that
would make square brackets more useful to all known
factions in this dispute.  I endorse his suggestion
by submitting it, verbatim, as a formal comment.

Will

                                * * *

(If someone else would care to turn the below into a formal comment
and/or SRFI, to be submitted under my or another name, I would
be delighted; and in fact if no one wants to go
to that effort that speaks poorly of the idea and it should probably
drop. I am sorry but I do not read r6rs-discuss very often so will
not see followup unless you cc me.  -Jonathan)


There is a feud between a faction that wants to be able to enlist two
distinct kinds of brackets to help beginners to learn Scheme, and/or
to help anyone to write what they consider to be clearer code, and a
faction that is horrified at the loss of 1/64 of the valuable ASCII
character space for a purpose that to them isn't justified, preventing
the use of brackets for legitimate extensions to Scheme.

No one will change anyone else's mind, since this conflict has been
raging unresolved since the 1970's.

Here is a compromise that satisfies concerns on both sides.  Define
[a b c] to be a special reader syntax for the four-element list
(*square-bracketed* a b c), just as '(a b c) is special reader syntax
for the two-element list (quote (a b c)).  By "*square-bracketed*" here
I really do mean the symbol of that name (although it could be some
other symbol).

For example, it will make sense to define:

(define-syntax let
   (syntax-rules (*square-bracketed*)
     ((let ([?var ?val] ...) ?body ...)
      ((lambda (?var ...) ?body ...) ?val ...))))

if you want to require that let-bindings be written using square
brackets, or

(define-syntax let
   (syntax-rules (*square-bracketed*)
     ((let ([?var ?val] ...) ?body ...)
      ((lambda (?var ...) ?body ...) ?val ...))
     ((let ((?var ?val) ...) ?body ...)
      ((lambda (?var ...) ?body ...) ?val ...))))

if you want to allow either syntax.  If you want [a b c] to be an
abbreviation for (list a b c), you can do

(define *square-bracketed* list)

and so on.

Note that this improves on synonymizing [ ] with ( ) for teaching
purposes in that it enables error checking - in every syntactic
position, you can enable or disable either kind of parenthesis
selectively.

The same kind of thing could also be done with curly brackets, using a
different marker.  (N.b. the Unicode names for the characters are LEFT
and RIGHT SQUARE BRACKET and LEFT and RIGHT CURLY BRACKET.)

WRITE could be made to generate [...] on output.

If I remember correctly, we did something like this for T (Yale
Scheme) in about 1983 or 1984, but I don't remember it being used.
Perhaps the feature was undocumented or experimental.

You may very well object that this is unhygienic - that the fact
that the name *square-bracketed* is free in [...] is in poor taste.
But I will point out that this is already true of quote (we all
ignored Brian C Smith way back when). The proposal is a
compromise, and the question is whether the benefits of its
peacekeeping effect compensates for the embarrassment it causes.


RESPONSE:

The goal of the report with respect to brackets is to allow programmers
the freedom to use brackets in place of parentheses where they please,
according to their individual tastes.  The suggested change would leave
the decision up to the language designers and/or system or library
implementors, i.e., those who write the transformers that either allow or
disallow the use of brackets.  It would also inhibit programmers from
using brackets in list data, if they so desire, and complicate
program-transformation code that deals with source-level expressions. 
Therefore, this change has not been adopted.  Implementors wishing to
allow the use of square brackets for some other purpose are free to do so
in code that follows an implementation-dependent #!<identifier> prefix as
generally described in the lead-in to chapter 4 of the main report.

A section describing suggested bracket usage, based on common practice,
has been added to the nonnormataive appendices.  This may stave off some
of the disputes feared by the comment submitter (or it may not).