Re: how to write macroexpand

[Per Bothner <per at bothner dot com>]

On 10/17/2010 02:52 PM, Helmut Eller wrote:
> * Per Bothner [2010-08-12 21:37] writes:
>
> > So my question is: What do you need macro-expand for?  The main use I
> > can think
> > of is when debugging macros.  Note that the --debug-print-expr command-line
> > option prints out "expanded" syntax, though the syntax isn't actually valid
> > Scheme - just a pretty-printed form of the Expression tree.
>
> Below is the code for a simple macroexpand that uses the first stage of
> the compiler and turns the AST back to a s-expressions which are
> hopefully more readable than the debug dumps.

Nice - I'd like to include this in gnu/kawa/slib.  A couple of issues:

(1) I need some kind of copyright statement (as part of the source file):
preferably either that you're releasing the code to the public domain,
or you're assigning copyright to me on behalf of the Kawa project.

(2) I was trying to figure out if there is a standard name for this kind
of function.  SRFI-96 http://srfi.schemers.org/srfi-96/srfi-96.html does
have macroexpand, but only in the context of supporting defmacro. I see that
Guile has a macroexpand, but I haven't found the documentation, and it
doesn't seem to do as much.  Racket has expand-syntax and related forms,
but these yield syntax objects, not sexprs.  Common Lisp has macroexpand,
but it doesn't seem like it does a full expansion - just at the toplevel.

The name macroexpand is fine - I just wonder if someone knows if other 
Schemes
use a different name.

(3) We should have at least a handful of test-cases to make sure I
don't inadvertently break macroexpand, given that it reaches into
implementation internals.

> Unfortunately, some
> fields in the IfExps class are private so we can't do that for ifs.

I just checked in accessors for that.

> Another issue are resolved functions in the AST that can't be easily
> mapped back to the source names.

I guess there are cases where we prematurely create a QuoteExp rather
than a ReferenceExp.  But the few tests I tried seemed to work OK.

> ...

> The pretty printer doesn't turn quote back to ' which would improve
> readability and it's even worse for backquote.

Fixing that would be tricky, since the right thing is content-dependent.
A macro-expand->string function (that returns a string rather than an
sexpr) might be able to do a better job.

> Note also that nothing is printed for '#!void.

I changed DisplayFormat.java so that readable output of Values.empty
returns "#!void".  That does not handle the general case of multiple values,
which are still printed non-readably.

Thanks for this contribution!
--
	--Per Bothner
per@bothner.com   http://per.bothner.com/