[Chicken-hackers] Re: swig

[Tony Sidaway]

Well I editedthe swig-produced example.scm file as follows:

removed the line "(declare (unit example))"

Surround the rest of the file with the following:

(module example
(export)
(import scheme chicken easyffi)

...

)

Now I can load that module but I don't know what to do to access the
symbols defined in swig_example_init in the file example_wrap.c, which
is compiled into the shared library and (presumably) called when the
module in example.scm is loaded.

It isn't enough for module example to export, say, My-Variable,
because it knows nothing about that binding.  Does example_wrap.c need
to do something to export the symbols so that example.scm can import
them and re-export?

On 11/11/09, Tony Sidaway <address@hidden> wrote:
> Is the chicken module of swig usable?
>
> I just ran swig 1.3.40 on the chicken/simple example bundled with swig
> using the makefile that comes with it.  It created a file "simple.so"
> but warned:
>
> Warning: library unit `example' compiled in dynamic mode
> Warning: global variable `swig-init-return' is never used
>
> When run in csi (4.2.0), the provided test script tries to load
> simple.so but barfs:
>
> Error: (load) unable to load compiled module
> "simple.so"
> #f
>
> The source file for that appears to be a library unit--a file with
> (declare (unit example)) at its head.
>
> I seem to recall something like this working at one time.
>
> Presumably the file should now be a chicken 4 module with declared
> exports.  That would require an update to swig's chicken module.
>
> I did try recompiling the Scheme file using Chicken 3.4.0 and running
> the script in a Chicken 3.4.0 REPL but that didn't help.
>
> Looking at the source file all it does it this:
>
> (define swig-init (##core#primitive "swig_example_init"))
> (define swig-init-return (swig-init))
>
> swig_example_init, the Scheme primitive, is defined and exported by
> example_wrap.c. It seems to do what I'd expect: calls some SWIG
> initialization functions, and then this:
>
>     a =
> C_alloc(2*4+C_SIZEOF_INTERNED_SYMBOL(11)+C_SIZEOF_INTERNED_SYMBOL(4)+C_SIZEOF_INTERNED_SYMBOL(3)+C_SIZEOF_INTERNED_SYMBOL(8));
>
>
>     sym = C_intern (&a, 11, "My-variable");
>     C_mutate ((C_word*)sym+1, (*a=C_CLOSURE_TYPE|1,
> a[1]=(C_word)_wrap_My_variable, tmp=(C_word)a, a+=2, tmp));
>     sym = C_intern (&a, 4, "fact");
>     C_mutate ((C_word*)sym+1, (*a=C_CLOSURE_TYPE|1,
> a[1]=(C_word)_wrap_fact, tmp=(C_word)a, a+=2, tmp));
>     sym = C_intern (&a, 3, "mod");
>     C_mutate ((C_word*)sym+1, (*a=C_CLOSURE_TYPE|1,
> a[1]=(C_word)_wrap_mod, tmp=(C_word)a, a+=2, tmp));
>     sym = C_intern (&a, 8, "get-time");
>     C_mutate ((C_word*)sym+1, (*a=C_CLOSURE_TYPE|1,
> a[1]=(C_word)_wrap_get_time, tmp=(C_word)a, a+=2, tmp));
>     C_kontinue (continuation, ret);
>
> In other words, pop some references on the stack and call its
> continuation.  "My_variable" is a C variable that is supposed to be
> accessible from Chicken, and the others are all C functions that are
> supposed to interface to Chicken.
>
> So it seems to be basically sane, but somewhere along the line it's
> been affected by the dreaded bitrot.
>
> Would be nice to be able to get it running again.
>