Re: [Axiom-developer] clisp and sbcl

[Waldek Hebisch]

> On Mon, 28 May 2007, Waldek Hebisch wrote:
> 
> [...]
> 
> | The problem is more subtle: the code above (printed by sbcl)
> | references |#1|.  Apparently sbcl complier does not see that
> | |#1| is defined, but it is.  
> | 
> | In more detail, Axiom uses Lisp eval to replace |#1| by aproproate
> | value.  Relevant part of backtrace is:
> | 
> | 17: (SB-INT:SIMPLE-EVAL-IN-LEXENV (EQCAR |#1| 0) #<NULL-LEXENV>)
> | 18: (|evalSharpOne|
> |      (EQCAR |#1| 0)
> |      (0 (1 #(#(|cos| 1 (#)) ((# 0 . 1)) 2 4864) (1 0 . -1)) 0 . 1))
> | 19: (|coerceUnion2Branch|
> |      ((|Union| (|Expression| (|Integer|)) (|List| (|Expression| 
> (|Integer|))))
> |       WRAPPED 0 (1 #(#(|cos| 1 (#)) ((# 0 . 1)) 2 4864) (1 0 . -1)) 0 . 1))
> | 
> | 
> | |evalSharpOne| is defined as:
> | 
> | (defun |evalSharpOne| (x \#1) (declare (special \#1)) (EVAL x))
> |
                              ^                      ^
                              |                      |
                        definition              declaration

> | so at the time of eval |#1| has special binding.  My understanding is
> | that such bindings are visible to eval.
> 
> The above shows that |#1| is *declared*.  But whereis it defined?
> 

Using parameters to establish special binding is very frequent idiom
in Axiom code.  I guess that from Ansi point of view '(EQCAR |#1| 0)'
may be incorrect, because it lacks special declaration.  Ansi would
like something like:

(defun |evalSharpOne| (x |#1|) (declare (special |#1|))
   (EVAL `(let () (declare (special |#1|) ,x))))

Or maybe:

(defun |evalSharpOne| (x y)
   (EVAL `(let ((|#1| ,y)) ,x)))

The second one does not create special binding, which would be more
tidy if x does not call functions referencing special binding of |#1|.

BTW.  This remaind me of your change replacing global SETQ by DEFPARAMETER.
IIUC 

(defparameter x y)

is equivalent to

(defvar x)
(setf x y)

It looks for me that replacing SETQ by DEFPARAMETER defeats much of the
purpose of DEFPARAMETER/DEFVAR: we silence _all_ warnings about assignments
to undefined variables, loosing them also when variable name is
spelled incorrectly.  I would prefer to have in a _single_ place explicit
DEFVAR/DEFPARAMETER declaration and in all other places use SETQ.

-- 
                              Waldek Hebisch
address@hidden