[r6rs-discuss] Low-level macro systems (Was: On naming the next scheme(s))

[Brian Mastenbrook]

On Aug 25, 2009, at 2:14 AM, Peter Bex wrote:

>> If you're thinking of
>> unhygienic macros, I think this would be a horrible idea; lexical
>> scoping of *all* identifiers, including those in syntax definitions,
>> is central to the Scheme idea.
>
> Not necessarily.  I think all low-level systems allow for breaking of
> hygiene, but that's not the reason to prefer a low-level system.
> A low-level system is useful because it allows one to explain macros
> in terms of familiar Scheme primitives as cons, car and cdr.
> This simplicity fits well with Scheme, and I think it would also serve
> well in an educational setting - having to explain that high-level
> macro systems are their own language separate from Scheme is yet  
> another
> layer of complexity.


 From an implementer's perspective, `syntax-rules' does indeed look  
like something that's difficult to explain, but in my experience  
having arbitrary code running in the macro-expansion phase raises its  
own can of worms. It's very easy for new programmers to get confused  
as to what is running when, and what needs to be around for certain  
things to happen. `syntax-rules' provides a flexible, easy-to-use  
macro system that retains one of the core ideas of the language, that  
of static scoping. The idea of template expansion as a means of  
metaprogramming is easily explained and avoids the need to discuss  
issues of phase.

Additionally, low-level macro systems are a contentious issue, and I  
don't believe that the implementor or user has yet settled on any  
consensus on the issue. Implementations of the R5RS are still being  
developed with a variety of low-level macro systems. It's to the R5RS  
user's benefit that a consistent means of hygienic macro definition is  
provided by all of these implementations. Rather than scrap `syntax- 
rules' in the Thing One language, I'd rather see extensions to `syntax- 
rules' standardized and included. SRFI-46 is an obvious candidate for  
this as it addresses a fundamental issue in syntax-defining syntax not  
accounted for by the R5RS. I also believe that a standard facility for  
limited breaking of hygiene could be included which would address a  
substantial fraction of the demand for a low-level macro system. I've  
sketched out what this might look like here:

http://paste.lisp.org/display/86014

The `define-syntax/identifier-context' ugliness is present because  
`syntax-rules' templates do not bind the first element of the template  
as a pattern variable, but it's necessary to recover hygienic context  
information.
--
Brian Mastenbrook
brian at mastenbrook.net
http://brian.mastenbrook.net/