Re: On the deprecated auto-loading of compiled-code modules

[Matthias Koeppe]

Marius Vollmer <address@hidden> writes:

> Matthias Koeppe <address@hidden> writes:
> 
> > 2) What is the "correct" way to register modules in the static case? I
> >    somehow need to expose the C-level initialization function to the
> >    yet-to-be-started interpreter; this is what
> >    scm_register_modules_xxx was for.
> 
> Ideally, I would like you to still use `dynamic-call' and
> `dynamic-link', even when doing static linking.  Guile is using
> libltdl for doing the actual linking, and libltdl has support for
> statically linking dynamic libtool libraries. [...]
>
> In effect, I want Guile to hand over almost all dynamic linking
> issures to libtool, including the handling of libraries that are not
> really dynamically linked.

Of course this is possible.  But I think that a real static linking
method, avoiding libtool's "export-dynamic" feature, should be
supported.

> If you don't want to go this untrodden route just now, there are two
> options I can see:
> 
> - Write a Scheme file for your module (like you would for dynamic
>   linking without explicit support), but don't use `dynamic-call' in
>   it.  Call the initialization procedure directly.  Like so:
> 
>   In C:
> 
>     init_module_bindings ()
>     {
>       #include "foo.x"
>     }
> 
>     init_module_initializer ()
>     {
>       scm_make_gsubr ("init-my-module", 0, 0, 0,
>                       (SCM (*)())init_module_bindings);
>     }
> 
>   In Scheme:
> 
>     (define-module (my module))
> 
>     (init-my-module)

But who will call init_module_initializer?  The Guile core modules are
all explicitly initialized in `scm_init_guile_1'.  I can't plug in my
initializers here (and I don't want to).  I use `scm_boot_guile' to
initialize Guile, and if I understand it right, I can't call
`scm_make_gsubr' when Guile has not been initialized yet. (Please
don't suggest I use `scm_init_guile' instead of `scm_boot_guile'.)

> - Explicitely create and select your module in the initialization routine

Same problem here.

-- 
Matthias Köppe -- http://www.math.uni-magdeburg.de/~mkoeppe