Re: [AD] 4.3 graphic drivers

[Elias Pschernig <elias@xxxxxxxxxx>]

On Fri, 2005-11-11 at 21:54 +0000, Peter Hull wrote:

> I've been playing around with an implementation of the 'versioned structures' 
> option, which I think is similar to Peter's. 
> In the case where a function's prototype is altered, and an old plugin tries 
> to register a vtable with a newer Allegro, I think it's possible for the 
> 'register' function to build a compatible vtable by allocating extra private 
> fields at the end, in which it will store the old function pointer. Assuming 
> that every vtable function will take an object pointer as its first argument, 
> you can get the vtable and then cast it to the version that contains the 
> private fields. However, the code looks rather confusing... something like 
> this:
> 
> struct private_vt {
>   VTABLE_VER_2 vt;
>   int (*old_somefunction)(OBJECT*, int);
> };
> 
> and the adaptor like this:
> int somefunction_v1_to_v2_adaptor(OBJECT* o, float arg) {
>   struct private_vt* pvt2=(struct private_vt*) o->vtable;
>   return (pvt2->oldsomefunction)(o, (int) arg);
> }
> 
> where the API change was from int somefunction(int) in v1 to int 
> somefunction(float) in v2. The definition of OBJECT is
> typedef struct {
>  void* vtable;
>  int field1, field2, ..
> } OBJECT;
> 
> Does that make sense? I need to polish up my code a bit and check it works, 
> then I'll post it, which might make everything clearer.
> 

Makes sense. But do we need the casting? We could explicitly specify the
vtable, like so (using the above example):

struct OBJECT { VTABLE_v9 *vtable; ...} // OBJECT is e.g. GFX_DRIVER or BITMAP

struct VTABLE_v9 { int (*somefunction)(OBJECT *, float); VTABLE_v8 *old;}

struct VTABLE_v8 { int (*somefunction)(OBJECT *, int); VTABLE_v7 *old;}

int somefunction_v9_with_v8_adaptor(OBJECT *o, float arg)
{
	VTABLE_v8 *old = o->old;
	return old->somefunction(old, (int)arg);
}

Of course, we would need as many adaptors as vtable versions, so I'm
actually getting afraid of the whole adaptors idea. Simply detecting a
version mismatch and failing to load an outdated driver would make this
much simpler..

-- 
Elias Pschernig