Re: [AD] A small bugfix in allegro_exit

[Eric Botcazou <ebotcazou@xxxxxxxxxx>]

> Speaking of which, does anyone know the rationale to use
> constructor routines with GCC ? It means stuff is called before
> allegro_init is called, which is not very friendly.

They make it possible to avoid fake dependencies between modules, for example 
the graphics module and the mouse module. Because set_gfx_mode() changes the 
mouse range, it used to contain a call to set_mouse_etc(), which causes the 
mouse module to be linked to the executable whenever the graphics module is, 
even though the mouse module is not used at all.

Laurence implemented a registration scheme that lets each module register its 
interface if it is present, so there is now an indirection in set_gfx_mode() 
and linking the graphics module doesn't cause the mouse module to be linked 
any longer.

But, in order for a module to register itself, its registration routine (for 
example _mouse_constructor in mouse.c) must be called from the core library 
so normally there is again a linking dependency, this time between the core 
library and the mouse module (see the beginning of install_allegro), 
_except_ if this registration routine is called through a constructor 
because there is no explicit call to constructors in the code.

> Also, it means that the image loader/saver list, for instance,
> is cleaned to the start state (eg, PCX/TGA/BMP/LBM), and not
> entirely.

Could you elaborate a bit?

-- 
Eric Botcazou