Re: [clisp-list] Post 2.33 CLISP behaviour and FFI

[Hoehle, Joerg-C. <Joe...@t-...>]

Nico wrote:
>I posted before on this list on the same subject. However, I=20
>investigated some more after that.
Thanks for taking the time to make a short example so I could look at =
it.

The parts relevant to the issue are
>(def-c-var sequencetab
and
>    (dolist (var '(sequencetab))
>      (set var (read stream)))))

IMHO this codes assumes undocumented behaviour of CLISP.

DEF-C-VAR introduces a global symbol macro.  Introducing a symbol macro =
changes the interpretation of programs.  E.g. the interpreter (and =
compiler) will assign a new meaning to an occurrence of the form =
SEQUENCETAB, and substitute the macro definition.

However, using SET (aka SETF SYMBOL-VALUE) "bypasses" interpretation of =
programs.  It accesses the value cell of a symbol.
Your program expects that cell to contain something related to the =
interpretation of programs.  But normally, such information is kept in =
what's called the environment.

>This 'works' with clisp 2.33.
Indeed, on 2005-01-01 Bruno Haible introduced a change which made =
global symbol macros orthogonal to symbol-value.

Before that, the symbol macro definition where found in a symbol's =
value cell.  SET and SYMBOL-VALUE where special-cased so as to perform =
the substitution.  In effect they were doing part of the interpreter's =
job.
Since the change, the substitution is kept in the interpreter's (or =
compiler's) environment -- where it arguably belongs.
SYMBOL-VALUE is left untouched, and freely available.
I don't know what motivated this change last year.


Great (disappointing?) news, now let's return to your problem:
>(defun load-seq (filename)
>  (with-open-file (stream filename :direction :input=20
>                   :if-does-not-exist :error)
>    (dolist (var '(sequencetab))
>      (set var (read stream)))))
>(load-seq "Mandseq-file")

It looks like your functions "expects" to alter the interpretation of =
the symbol SEQUENCETAB.  This is a job of the interpreter (or =
compiler), and SET or SYMBOL-VALUE will not manage to do so (anymore).
So you need to modify the environment.

Your initial example was:
>(defmacro def-i/o (writer-name reader-name (&rest vars))
>(def-i/o save-seq load-seq (sequencetab))

Since you statically know the names of the variables involved, you can =
have def-i/o generate the following snippet of code:
 ... (setq/setf sequencetab (read stream))
[As usual, use macroexpand-1 to test macro output.]

Then the interpreter (and compiler) will do the right thing, since they =
know (from their environment) that sequencetab is a foreign variable.
It's purely lexical, there's no need for SPECIAL declarations or =
SYMBOL-VALUE.

Regards,
	J=F6rg H=F6hle.