[r6rs-discuss] [Formal] Internal define-syntax lexical scoping violation

[AndrevanTonder]

---
This message is a formal comment which was submitted to formal-comment at r6rs.org, following the requirements described at: http://www.r6rs.org/process.html
---
Name        : Andre van Tonder
Email       : andre at het.brown.edu
Type        : defect
Priority    : major
Component   : Expansion process
Version     : 5.91
Pages       : 27-28
Dependencies: Sections 9.3 (Syntax definitions)

Summary:
--------

Lexical scoping violation for internal define-syntax.

Description:
------------

In the following example, the expansion algorithm, as described in r6rs, 
will consider (bar x) an expression and throw a syntax violation stating that 
expressions may not precede definitions.

    (let ((bar 1))
      (bar x)
      (define-syntax bar
        (syntax-rules ()
          ((bar x) (define x 1))))
      x)

This syntax error violates lexical scoping, specifically the sentence "Keyword 
bindings established by define-syntax are visible throughout the body in which 
they appear, ..." on p. 29.  By lexical scoping, the result should be 1.  I 
think it can be made 1 if the expansion algorithm is made multi-pass (see 
below).

I am unsure if this case is covered by the constraint on p.28, which reads:
"It is a syntax violation if the keyword that identifies one of the body forms 
as a definition (derived or core) is redefined by the same definition or a later 
definition in the same body."  When (bar x) is expanded, it is not yet known 
that bar will be a keyword.  In any case, bar is never "re"-defined.

Even if it is intended to be covered by the constraint, it will not be detected 
by the algorithm in the first paragraph on page 28.

Suggestion:
-----------

Perhaps lexical scoping with internal define-syntax can be better handled by a 
multipass expansion algorithm instead of single-pass.  On each pass, the 
expansion process would restart, using syntax definitions discovered on previous 
passes, until no more syntax definitions are discovered.

It seems that such an algorithm would give the result 1 above.  The 
implied reprocessing of forms, however, may cause problems with macros that 
update expand-time state (e.g. record definitions).