Re: [Sbcl-devel] muffle-conditions, or "[?]"

[Nikodemus S. <tsi...@cc...>]

On Tue, 20 Apr 2004, Christophe Rhodes wrote:

> So, firstly, is this over-speculative generality, or is my analysis of
> the need correct?  (How do other Lisps overcome compile-time package
> lock checking locally, if indeed they do?)

As far as I can tell, they don't. Both CMUCL and Allegro use
WITHOUT-PACKAGE-LOCKS, which expands into:

 (let ((*enable-package-locks* nil))
   ...)

I really like the declaration protocol, but I just started thinking:
is this really a half-way house?

As you hint, it would be good for users to be able make meaningfull
declarations in for use in their own compile-time protocols, and it might
make more sense to provide documented environment then shoehorn everything
into restarts.

Then users could do:

 (define-compiler-macro bar (&whole form &env env x)
   (if (subtypep (variable-type x env) '(integer 0 0))
       1.0
       (let ((decl (cdr (assoc 'unoptimizable-bar (declarations env)))))
         (if (and (consp decl) (eq 'use-random (car decl)))
             `(random ,(cdr decl))
              form))))

 (defun foo (x)
   (declare (unoptimizable-bar (use-random 1.0))
   (bar x))

For package locks the protocol would become eg:

 (in-package "Foo")

 (defmacro foo () ...)

 (defun make-macrolet (&body body)
   `(locally
         (declare (disable-package-locks "Foo"))
       (macrolet ((foo () ...))
         (declare (enable-package-locks "Foo"))
         ,@body)))

I haven't really thought this option through yet, esp. not the
way it will interact with the in-severe-need-of-more-specification
compile vs. runtimeness of package locks.

Cheers,

 -- Nikodemus