Re: comments on new-model.txt

[Lynn Winebarger]

On Monday 09 September 2002 17:55, Marius Vollmer wrote:
> Lynn Winebarger <address@hidden> writes:
> 
> >       So, we have two SCM types, ilocs and variables (actually
> > variable bindings, but we can leave that for later).  ilocs always
> > look up their values in the data environment of their closure, and
> > variables are just pointers to a storage location.  An extloc would
> > be like an iloc (that is, a frame offset and distance in a frame),
> > but would be stored in a cell where the second word pointed to the
> > data environment to look it up in (the first word would contain the
> > frame offset and distance).
> 
> What use would be an extloc?  Do you want to export lexical variables?
     Yes
> Or would they be useful for hygienic macro expansion? 
      Not exactly, but the process of writing my own version of syntax-case
has made me think of variables in this way.
> Can you give
> example code that illustrates their use?
     Ok, I'll make up some incredibly bad notation
(exporting-lambda () 
   (define-exportable foo  'whatever)
   (define-exportable-syntax (lambda (x) (syntax-case x () ((_ f ...) (syntax 
'xyzzy)))))
   ....)
       If we have extlocs, this can be just a plain lambda that happens
to let import take references to some of its variables.  Then we can
use the keyword "module" instead.

> > Likewise an extvarref would contain both the pointer to the binding
> > and some way of identifying what top-level environment it's from.
> 
> If we need to, we could find the originating module for a imported
> variable also from the module that the current code is executed in.
> This would be slower, but we would not need the extra pointer in
> extvarrefs or variables.

     Part of the disagreement is that I don't think we should
"evaluate in" modules. Partly this is a terminology issue:  I think
"module" should be reserved for separately compilable units of
code that won't be invisibly dependent on whatever code happens
to be currently loaded ("invisibly" is because imports are clearly
visible to the module-using programmer, so are not prohibited).
    We should have some other terminology to describe other types
of namespaces.  Being able to import/export variables is just one
knob on a namespace/environment/whatever.
    And then there's Dirk's view of modules as parameterizable to
get a handle on too.

> > This would be a convenient way to represent imported variables.  The
> > only concrete reason I can offer for including a pointer (of some
> > form) to the top level environment of an external variable (actually
> > any top-level variable) is for ones that aren't bound at the time of
> > parsing (recalling that correct macro expansion can't be lazy in the
> > presence of side-effects, as some of the syntax-case examples I've
> > been posting should demonstrate).
> 
> What would you do with that pointer?

     I'm thinking about it.

Lynn