More Guile 1.8 code failing in 2.x

More Guile 1.8 code failing in 2.x

[Richard Shann]

We have one more construct being refused as we upgrade to guile 2.0 in
GNU/Denemo

(if (not (defined? 'ToggleTripleting::InsideTriplet))
    (define ToggleTripleting::InsideTriplet #t))

It is intended to set up a flag which toggles between true and false on
each call. If already set up, the flag is not altered.

Apparently (I haven't been able to check) Guile 2.0  gives an error.

#f definition in expression context, where definitions are not allowed, ((line . 2) (column . 4) (filename . #f)) (define ToggleTripleting::InsideTriplet #t) #f)

Can anyone suggest what could replace that? I would prefer to to replace
it with something that has the same semantics, rather than setting up
some alternative method of handling the situation - registering all such
variables on program startup, or some such - as at the moment this code
only evaluated if used and doesn't require any mechanism to get
initialized, other than this construct.

Richard Shann

Re: More Guile 1.8 code failing in 2.x

[mark]

Richard Shann <richard.shann@virgin.net> writes:

> We have one more construct being refused as we upgrade to guile 2.0 in
> GNU/Denemo
>
> (if (not (defined? 'ToggleTripleting::InsideTriplet))
>     (define ToggleTripleting::InsideTriplet #t))
>
> It is intended to set up a flag which toggles between true and false on
> each call. If already set up, the flag is not altered.
>
> Apparently (I haven't been able to check) Guile 2.0  gives an error.
>
> #f definition in expression context, where definitions are not
> allowed, ((line . 2) (column . 4) (filename . #f)) (define
> ToggleTripleting::InsideTriplet #t) #f)
>
> Can anyone suggest what could replace that? I would prefer to to replace
> it with something that has the same semantics, rather than setting up
> some alternative method of handling the situation - registering all such
> variables on program startup, or some such - as at the moment this code
> only evaluated if used and doesn't require any mechanism to get
> initialized, other than this construct.
>
> Richard Shann
>
>
>
>

Hi Richard,

You can probably just replace your code with

(define-once ToggleTripleting::InsideTriplet #t)

(define-once ...)  works like defvar in Common Lisp and will not
redefine a variable once it has been defined.  You can read about it at
the bottom of the node "Top Level Variable Definitions" in the Guile
manual.

I'm still learning the ins and outs of procedural macros, but I think
that if you need to include (define ...) in a conditional like that for
more complicated purposes, you can use syntax-case like this:

(define-syntax conditional-define
  (lambda (x)
    (syntax-case x ()
      ((conditional-define arg val)
       (let ((arg-datum (syntax->datum #'arg)))
	 (if (not (defined? arg-datum))
	     #'(define arg val) 
             #'*unspecified*))))))

-- 
Mark Witmer

Re: More Guile 1.8 code failing in 2.x

[Richard Shann]

On Wed, 2013-02-27 at 08:19 -0500, mark@markwitmer.com wrote:
> Richard Shann <richard.shann@virgin.net> writes:
> 
> > We have one more construct being refused as we upgrade to guile 2.0 in
> > GNU/Denemo
> >
> > (if (not (defined? 'ToggleTripleting::InsideTriplet))
> >     (define ToggleTripleting::InsideTriplet #t))
> >
> > It is intended to set up a flag which toggles between true and false on
> > each call. If already set up, the flag is not altered.
> >
> > Apparently (I haven't been able to check) Guile 2.0  gives an error.
> >
> > #f definition in expression context, where definitions are not
> > allowed, ((line . 2) (column . 4) (filename . #f)) (define
> > ToggleTripleting::InsideTriplet #t) #f)
> >
> > Can anyone suggest what could replace that? I would prefer to to replace
> > it with something that has the same semantics, rather than setting up
> > some alternative method of handling the situation - registering all such
> > variables on program startup, or some such - as at the moment this code
> > only evaluated if used and doesn't require any mechanism to get
> > initialized, other than this construct.
> >
> > Richard Shann
> >
> >
> >
> >
> 
> Hi Richard,
> 
> You can probably just replace your code with
> 
> (define-once ToggleTripleting::InsideTriplet #t)

Thanks for this - it seems define-once is not defined in guile 1.8
however, so while some systems only have guile 2.0 (Fedora) and others
only have guile 1.8 (Debian stable) I would need something more.
I wonder if I can test if define-once is defined and if not evaluate a
string which has the dis-allowed syntax in it?

(if (defined? 'define-once)
	(define-once ToggleTripleting::InsideTriplet #t)
	(eval-string "(if (not (defined? 'ToggleTripleting::InsideTriplet))
		     	(define ToggleTripleting::InsideTriplet #t))"))

or will the guile 2.0 compiler try and compile the string being passed to eval-string?

Richard


> 
> (define-once ...)  works like defvar in Common Lisp and will not
> redefine a variable once it has been defined.  You can read about it at
> the bottom of the node "Top Level Variable Definitions" in the Guile
> manual.
> 
> I'm still learning the ins and outs of procedural macros, but I think
> that if you need to include (define ...) in a conditional like that for
> more complicated purposes, you can use syntax-case like this:
> 
> (define-syntax conditional-define
>   (lambda (x)
>     (syntax-case x ()
>       ((conditional-define arg val)
>        (let ((arg-datum (syntax->datum #'arg)))
> 	 (if (not (defined? arg-datum))
> 	     #'(define arg val) 
>              #'*unspecified*))))))
> 

Re: More Guile 1.8 code failing in 2.x

[Ludovic Courtès]

Richard Shann <richard.shann@virgin.net> skribis:

> (if (defined? 'define-once)
> 	(define-once ToggleTripleting::InsideTriplet #t)
> 	(eval-string "(if (not (defined? 'ToggleTripleting::InsideTriplet))
> 		     	(define ToggleTripleting::InsideTriplet #t))"))

Rather something along these lines:

  (cond-expand ((not guile-2)
                (define-macro (define-once var expr)
                  `(if (not (defined? ',var))
                       (define ,var ,expr))))
               (guile-2 #t))

This will do nothing on Guile 2.0, and will define ‘define-once’ on 1.8.

HTH,
Ludo’.

Re: More Guile 1.8 code failing in 2.x

[Mike Gran]

> From: Richard Shann <richard.shann@virgin.net>
> 
> Thanks for this - it seems define-once is not defined in guile 1.8
> however, so while some systems only have guile 2.0 (Fedora) and others
> only have guile 1.8 (Debian stable) I would need something more.

Just as an aside... Fedora doesn't have Guile 2.0 at all, except in its
experimental Rawhide distro.  It still ships with 1.8.  Still about
half of the Guile-using programs in Fedora can't compile with 2.0, so
Fedora will be shipping with both guile-1.8 and 2.0 in the next version,
Fedora 19.  Maintainers will be able to link their packages to
either one. The executable for Guile 2.0 will likely be called
/usr/bin/guile2.
 
-Mike

Re: More Guile 1.8 code failing in 2.x

[Stefan Israelsson Tampe]

Can you wrap the code in a cond-expand form no?

/Stefan

Re: More Guile 1.8 code failing in 2.x

[Andy Wingo]

Hi Richard,

On Wed 27 Feb 2013 15:22, Richard Shann <richard.shann@virgin.net> writes:

> On Wed, 2013-02-27 at 08:19 -0500, mark@markwitmer.com wrote:
>> Richard Shann <richard.shann@virgin.net> writes:
>> 
>> > (if (not (defined? 'ToggleTripleting::InsideTriplet))
>> >     (define ToggleTripleting::InsideTriplet #t))

I'm sorry we were not able to make this case work in Guile 2; it's quite
complicated actually.

>> You can probably just replace your code with
>> 
>> (define-once ToggleTripleting::InsideTriplet #t)
>
> Thanks for this - it seems define-once is not defined in guile 1.8
> however

You could define it in a prelude somewhere:

  (cond-expand
   (guile-2) ; nothing
   (else ; guile < 2.0
    (define-macro (define-once sym exp)
      `(if (not (defined? ',sym)) (define ,sym ,exp)))))

Then you could change all uses of this idiom to use define-once.

Note that Guile 2.'s `define-once' is slightly different; it always
declares a local definition, and it uses the existing value only if
there is already a local definition.  In this way imported values get
overridden by define-once.

Here is a Guile 2-compatible version:

  (cond-expand
   (guile-2) ; nothing
   (else ; guile < 2.0
    (define-macro (define-once sym exp)
      `(define ,sym (if (module-locally-bound? ',sym) ,sym ,val)))))

Finally in Guile 2 there is also the `define!' procedure, which makes
top-level definitions programmatically.  Use it if you need it.  But if
you can, better to rewrite to use standard Scheme.

In general I recommend adding shims to Denemo to make it
forward-compatible -- like adding define-once if needed so that it uses
the Guile 2.0 idioms, even on 1.8.

Good luck, and let us know how it goes.

Andy
-- 
http://wingolog.org/

Re: More Guile 1.8 code failing in 2.x

[Andy Wingo]

On Wed 27 Feb 2013 16:08, Andy Wingo <wingo@pobox.com> writes:

> Here is a Guile 2-compatible version:
>
>   (cond-expand
>    (guile-2) ; nothing
>    (else ; guile < 2.0
>     (define-macro (define-once sym exp)
>       `(define ,sym (if (module-locally-bound? ',sym) ,sym ,val)))))

Sorry, thinko+typo.  Should be:

  (cond-expand
   (guile-2) ; nothing
   (else ; guile < 2.0
    (define-macro (define-once sym exp)
      `(define ,sym
         (if (module-locally-bound? (current-module) ',sym)
             ,sym
             ,exp)))))

-- 
http://wingolog.org/

Re: More Guile 1.8 code failing in 2.x

[Richard Shann]

On Wed, 2013-02-27 at 06:48 -0800, Mike Gran wrote:
> > From: Richard Shann <richard.shann@virgin.net>
> > 
> > Thanks for this - it seems define-once is not defined in guile 1.8
> > however, so while some systems only have guile 2.0 (Fedora) and others
> > only have guile 1.8 (Debian stable) I would need something more.
> 
> Just as an aside... Fedora doesn't have Guile 2.0 at all, except in its
> experimental Rawhide distro.
Yes, it was the maintainer of Denemo for Fedora that raised a bug about
the lack of guile 2.0 support. I guess he was unable to test it for the
upcoming release...

Richard

>   It still ships with 1.8.  Still about
> half of the Guile-using programs in Fedora can't compile with 2.0, so
> Fedora will be shipping with both guile-1.8 and 2.0 in the next version,
> Fedora 19.  Maintainers will be able to link their packages to
> either one. The executable for Guile 2.0 will likely be called
> /usr/bin/guile2.
>  
> -Mike

Re: More Guile 1.8 code failing in 2.x

[Richard Shann]

On Wed, 2013-02-27 at 16:13 +0100, Andy Wingo wrote:
> On Wed 27 Feb 2013 16:08, Andy Wingo <wingo@pobox.com> writes:
> 
> > Here is a Guile 2-compatible version:
> >
> >   (cond-expand
> >    (guile-2) ; nothing
> >    (else ; guile < 2.0
> >     (define-macro (define-once sym exp)
> >       `(define ,sym (if (module-locally-bound? ',sym) ,sym ,val)))))
> 
> Sorry, thinko+typo.  Should be:
> 
>   (cond-expand
>    (guile-2) ; nothing
>    (else ; guile < 2.0
>     (define-macro (define-once sym exp)
>       `(define ,sym
>          (if (module-locally-bound? (current-module) ',sym)
>              ,sym
>              ,exp)))))
> 

This works a treat (well, of course I didn't actually get to test on
guile 2.0 myself, but I will doubtless hear about it if it should fail
there :)

Thank you all who responded.

Richard