[Gcl-devel] Re: GCL compiler bug: load-time-value and macroexpansion

[Camm Maguire]

Greetings!  I'm not very happy about the (not (eq 'lambda (car
form))), but it might be worth testing this:

(defun wrap-literals (form &aux fd)
  (cond ((and (consp form) (eq (car form) 'quote))
         (let ((x (cadr form)))
           (if (and (symbolp x)
                    (eq :external (cadr (multiple-value-list (find-symbol 
(symbol-name x) 'lisp)))))
               form
             `(load-time-value (si::nani ,(si::address x))))))
        ((and (consp form) (symbolp (car form)) (not (eq 'lambda (car form))) 
(setq fd (macro-function (car form))))
         (wrap-literals (cmp-expand-macro fd (car form) (cdr form))))
        ((consp form)
         (cons (wrap-literals (car form)) (wrap-literals (cdr form))))
        ((symbolp form)
         (unless (symbol-package form)
           (unless *tmp-pack*
             (setq *tmp-pack* (make-package (symbol-name (gensym)))))
           (import form *tmp-pack*))
         form)
        ((or (rationalp form) (characterp form))
         form)
        (`(load-time-value (si::nani ,(si::address form))))))

Take care,


Matt Kaufmann <address@hidden> writes:

> By the way, the times for the ACL2 regression suite are virtually
> identical before and after the following change (added to a compiled
> ACL2 source file):
> 
> #+(and gcl (not ansi-cl)) (defun compiler::wrap-literals (x) x)
> 
> ; Before above addition:
> 12990.367u 274.629s 3:46:08.34 97.7%  0+0k 0+0io 5pf+0w
> ; After above addition:
> 12987.607u 275.321s 3:45:30.28 98.0%  0+0k 0+0io 0pf+0w
> 
> I've saved a copy of the development sources that I used, so that I
> can test an alternate wrap-literals that you send me.
> 
> -- Matt
>    Sender: address@hidden
>    Cc: address@hidden
>    From: Camm Maguire <address@hidden>
>    Date: 30 Apr 2007 13:35:32 -0400
>    X-SpamAssassin-Status: No, hits=-2.5 required=5.0
>    X-UTCS-Spam-Status: No, hits=-310 required=200
> 
>    Greetings!  This should work.  Would you be willing to test an
>    alternate wrap-literals if I get one together in the near-future?
> 
>    Take care,
> 
>    Matt Kaufmann <address@hidden> writes:
> 
>    > Thank you, Camm.  Unfortunately, after (setq compiler::*keep-gaz* t),
>    > then all the gazonk*.lsp files are left around.  So I'm wondering if
>    > it would safe to do the following instead:
>    > 
>    > #+(and gcl (not ansi-cl)) (defun compiler::wrap-literals (x) x)
>    > #+(and gcl (not ansi-cl)) (compile 'compiler::wrap-literals)
>    > 
>    > A small test suggests that this may work, though I have no idea really
>    > what I'm doing.  Should I expect the above solution to be OK?
>    > 
>    > Thanks --
>    > -- Matt
>    >    Sender: address@hidden
>    >    Cc: address@hidden
>    >    From: Camm Maguire <address@hidden>
>    >    Date: 30 Apr 2007 12:16:06 -0400
>    >    X-SpamAssassin-Status: No, hits=-2.5 required=5.0
>    >    X-UTCS-Spam-Status: No, hits=-310 required=200
>    > 
>    >    Greetings, and thanks so much for this report!
>    > 
>    >    The issue in brief stems from ansification -- compile'ed forms must
>    >    refer to the exact object literally referred to in the form, not a
>    >    copy, so the traditional GCL print and compile-file won't work.  The
>    >    function is compiler::wrap-literals, which you can trace if
>    >    interested.  There is obviously a bug here -- most likely
>    >    wrap-literals should do some selective macro-expansion, perhaps along
>    >    the lines of compiler::portable-source in 2.7.0.  I will see if I can
>    >    come up with a solution which also retains our current (2.7.0)
>    >    compatibility with the ansi tests for compile.  If you have any
>    >    suggestions, they are of course most appreciated.  The tests in
>    >    question as run thus:
>    > 
>    >    cd ansi-tests
>    >    ../unixport/saved_ansi_gcl
>    >    >(load "gclload1")
>    >    >(load "compile")
>    >    >(load "compile-file")
>    >    >(rt:do-tests)
>    > 
>    >    There is an immediate work-around.  Set the variable
>    >    compiler::*keep-gaz* to t -- this avoids wrap-literals and behaves as
>    >    the traditional compile via print/compile-file did.  The idea is that
>    >    there are certain packages in the ansi build, notably pcl, which
>    >    compile functions which need to be linked later in gazonk files at the
>    >    raw build stage.  Even though pcl uses compile here, literal object
>    >    reference is impossible as the running image at compile time is gone.
>    >    So qualitatively if one needs to keep the gazonk files around, they
>    >    better not refer to objects only available in the compiling image.  
>    > 
>    >    This exception in all likelihood should not be there eventually, but I
>    >    can't at the moment envision a bridge between ansi compile and
>    >    traditional gcl compile without one.
>    > 
>    >    Comments/suggestions as always most welcome.
>    > 
>    >    Take care,
>    > 
>    >    Matt Kaufmann <address@hidden> writes:
>    > 
>    >    > Hello --
>    >    > 
>    >    > It appears that the GCL compiler (at least: version 2.6.7 CLtL1, and
>    >    > also version 2.7.0 ANSI as of about 11/27/06) is laying down calls 
> of
>    >    > lisp::load-time-value that are interfering with macro expansion.
>    >    > Below is an example exhibiting the problem.
>    >    > 
>    >    > Is there any simple workaround, such as (setq *some-compiler-switch*
>    >    > nil)?  By the way, the actual (much bigger) failure I had, from 
> which
>    >    > the example below is extracted, was only an explicit error when
>    >    > calling COMPILE as shown below.  When I put the function into a 
> file,
>    >    > I didn't see any problem with COMPILE-FILE, but I found bizarre and
>    >    > somewhat nondeterministic behavior that went away when I avoided
>    >    > compiling that function by loading the .lisp file instead.
>    >    > 
>    >    > .....
>    >    > 
>    >    > >(defmacro my-mac (b)
>    >    >        (list 'list
>    >    >           (if (and (consp b)
>    >    >                    (stringp (car b)))
>    >    >               (list 'quote b)
>    >    >             b)))
>    >    > 
>    >    > MY-MAC
>    >    > 
>    >    > >(defun foo ()
>    >    >    (my-mac ("Guards")))
>    >    > 
>    >    > FOO
>    >    > 
>    >    > >(foo)
>    >    > 
>    >    > (("Guards"))
>    >    > 
>    >    > >(compile 'foo)
>    >    > 
>    >    > Compiling gazonk4.lsp.
>    >    > ; (DEFUN FOO ...) is being compiled.
>    >    > ;;; The function (LOAD-TIME-VALUE (SYSTEM:NANI 139732192)) is 
> illegal.
>    >    > No FASL generated.
>    >    > 
>    >    > Error: Cannot open the file NIL..
>    >    > Fast links are on: do (si::use-fast-links nil) for debugging
>    >    > Error signalled by LET.
>    >    > Broken at LOAD.  Type :H for Help.
>    >    > >>(quit)
>    >    > sundance:~> cat gazonk4.lsp
>    >    > 
>    >    > (lisp::defun user::foo lisp::nil (user::my-mac 
> ((lisp::load-time-value (system::nani 139732192)))))sundance:~> 
>    >    > 
>    >    > Thanks --
>    >    > -- Matt
>    >    > 
>    >    > 
>    >    > 
>    > 
>    >    -- 
>    >    Camm Maguire                                          address@hidden
>    >    
> ==========================================================================
>    >    "The earth is but one country, and mankind its citizens."  --  
> Baha'u'llah
>    > 
>    > 
>    > 
> 
>    -- 
>    Camm Maguire                                               address@hidden
>    ==========================================================================
>    "The earth is but one country, and mankind its citizens."  --  Baha'u'llah
> 
> 
> 

-- 
Camm Maguire                                            address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah