Juanma Barranquero <lekktu@gmail.com> writes:

> Additionally to Eli's comments, the functions init_*_functions are
> defined after being used to initialize the struct, so it fails:
>
> You'll have to move them around.

Thanks.  Fixed in the attached patch, which also addresses Eli's other
comments.  Please take a look again.

> Anyway, what I was doing to move Vlibrary_cache out of w32 is affected
> by your patch, so I'd be happy if you could also fix that. The main
> issue is making sure that dynamic loading can be used before dumping,
> but the dumped Vlibrary_cache and image_types do not get dumped with
> entries that would not be relevant afterwards (that's could be an
> issue on Windows).

As far as I can tell, my patch should not change anything in this
regard, except that when you move Vlibrary_cache out of w32, you can
also get rid of the #ifdef NTGUI in define_image_type.


Eli Zaretskii <eliz@gnu.org> writes:

> Since LIBRARIES could now be Qnil, but this function cannot tolerate
> that, there should be an assertion to that effect, either here or in
> every type->init function.

Passing a nil LIBRARIES argument causes nil to be passed to
w32_delayed_load, which AFAICT is safe; w32_delayed_load deals with nil
just fine, as a no-op.