bug#10974: guile-user@gnu.org

bug#10974: guile-user@gnu.org

[Alexei Matveev]

Hi, All,

For use from a Fortran program I am collecting API fixes for libguile.so
as wrapper functions for what is provided to C-programs as macros.
I noted that some of the macros are function-macros some are symbol
macros. An example of the latter is

  #define scm_to_int scm_to_int23

This is inconsistent and makes escaping such macros slightly more difficult.
(Honestly I dont know a way yet).

One guile developer on IRC said this is "probably a good thing to fix"
so I report
it here. I dont build Guile 2 myself as the installations I use are quite
dated/conservative.

As a background, Fortran allows you to declare "foreign" functions, for example,
like this:

type, public, bind(c) :: scm_t
   private
   integer(c_intptr_t) :: do_not_ever_use
end type scm_t

interface
  function scm_symbol_p (obj) result (yes) bind (c)
     type(scm_t), intent(in), value :: obj
    type(scm_t) :: yes
  end function scm_symbol_p
  ...
end interface

This makes use of library functions quite handy. But it does not provide a
way to access a C-macro, naturally.

Alexei

#include <libguile.h>

SCM guile_macro_scm_from_int (int i);
int guile_macro_scm_to_int (SCM obj);

int (scm_is_true) (SCM obj);
int (scm_is_symbol) (SCM obj);

int (scm_is_null) (SCM obj);
SCM (scm_eol) (void);

SCM guile_macro_scm_from_int (int i)
{
  return scm_from_int(i);
}

int guile_macro_scm_to_int (SCM obj)
{
  return scm_to_int(obj);
}

int (scm_is_true) (SCM obj)
{
  return scm_is_true(obj);
}

int (scm_is_symbol) (SCM obj)
{
  return scm_is_symbol(obj);
}

int (scm_is_null) (SCM obj)
{
  return scm_is_null(obj);
}

SCM (scm_eol) ()
{
  return SCM_EOL;
}

bug#10974: guile-user@gnu.org

[Ludovic Courtès]

Hi,

Sorry for the late reply.

Alexei Matveev <alexei.matveev@gmail.com> skribis:

> For use from a Fortran program I am collecting API fixes for libguile.so
> as wrapper functions for what is provided to C-programs as macros.
> I noted that some of the macros are function-macros some are symbol
> macros. An example of the latter is
>
>   #define scm_to_int scm_to_int23

The macros in numbers.h that are “symbol macros”, such as
‘scm_from_int’, allow users to write code like:

  &scm_from_int

This wouldn’t be possible if these were function macros.

Thus, I think things will have to remain this way.

What do you think?

Thanks,
Ludo’.

bug#10974: guile-user@gnu.org

[Alexei Matveev]

On Mon, Jul 2, 2012 at 11:38 AM, Ludovic Courtès <ludo@gnu.org> wrote:
>
>> For use from a Fortran program I am collecting API fixes for libguile.so
>> as wrapper functions for what is provided to C-programs as macros.
>> I noted that some of the macros are function-macros some are symbol
>> macros. An example of the latter is
>>
>>   #define scm_to_int scm_to_int23
>
> The macros in numbers.h that are “symbol macros”, such as
> ‘scm_from_int’, allow users to write code like:
>
>   &scm_from_int
>
> This wouldn’t be possible if these were function macros.
>
> Thus, I think things will have to remain this way.

Hi,

It's ok. You may close it.

I still think it could be less confusing if the libguile.so implemented/provided
functions as advertised in Guile API docs for the sake of interfacing to
languages other than C. And &scm_from_int wold also work if it were a real
function.

But there are many more macros, so such a link-time interface would be a
lot of work, I realize by now.

Alexei

bug#10974: guile-user@gnu.org

[Ludovic Courtès]

Hi,

Alexei Matveev <alexei.matveev@gmail.com> skribis:

> It's ok. You may close it.

Thanks.

> I still think it could be less confusing if the libguile.so implemented/provided
> functions as advertised in Guile API docs for the sake of interfacing to
> languages other than C. And &scm_from_int wold also work if it were a real
> function.
>
> But there are many more macros, so such a link-time interface would be a
> lot of work, I realize by now.

Yeah.  Though here, you could still write bindings for ‘scm_from_int32’
(the real function) instead of ‘scm_from_int’, for instance, no?

Thanks,
Ludo’.

bug#10974: guile-user@gnu.org

[Andy Wingo]

On Mon 02 Jul 2012 11:38, ludo@gnu.org (Ludovic Courtès) writes:

> The macros in numbers.h that are “symbol macros”, such as
> ‘scm_from_int’, allow users to write code like:
>
>   &scm_from_int
>
> This wouldn’t be possible if these were function macros.

Interesting, I hadn't thought of that.

> Thus, I think things will have to remain this way.

Would this be "fixed" if we changed these to be implemented as inline
functions?

Dunno, just thinking out loud.

Andy
-- 
http://wingolog.org/

bug#10974: guile-user@gnu.org

[Ludovic Courtès]

Hi,

Andy Wingo <wingo@pobox.com> skribis:

> Would this be "fixed" if we changed these to be implemented as inline
> functions?

In some way, though their address could still not be taken.

Thanks,
Ludo’.

bug#10974: guile-user@gnu.org

[Andy Wingo]

On Tue 03 Jul 2012 00:35, ludo@gnu.org (Ludovic Courtès) writes:

> Andy Wingo <wingo@pobox.com> skribis:
>
>> Would this be "fixed" if we changed these to be implemented as inline
>> functions?
>
> In some way, though their address could still not be taken.

Sorry to be obtuse, but with our inline functions, we also residualize a
concrete instantiation (via the inline.c madness).  Does that not work?

A
-- 
http://wingolog.org/

bug#10974: guile-user@gnu.org

[Ludovic Courtès]

Hi,

Andy Wingo <wingo@pobox.com> skribis:

> On Tue 03 Jul 2012 00:35, ludo@gnu.org (Ludovic Courtès) writes:
>
>> Andy Wingo <wingo@pobox.com> skribis:
>>
>>> Would this be "fixed" if we changed these to be implemented as inline
>>> functions?
>>
>> In some way, though their address could still not be taken.
>
> Sorry to be obtuse, but with our inline functions, we also residualize a
> concrete instantiation (via the inline.c madness).  Does that not work?

Yes, right, so we could just inline them all.

Ludo’.