Expose is_class_ptr to GDNative for dynamic casts

[sheepandshepherd]

Casting objects of non-exposed classes (BulletPhysicsDirectSpaceState etc) always fails in GDNative C++. #27936 proposed exposing all internal classes to fix the type tag system used for casting, but a simpler solution would be for the bindings to reimplement casts the same way the engine does.

This PR exposes is_class_ptr and class pointers only for exposed classes - nothing from internal classes needs to be accessible from GDNative, since is_class_ptr only requires the target class's pointer. (These do still work when Godot is built using dynamic_cast instead.)

Supersedes #27936 and godotengine/godot-headers#44

Relevant to #28195, #28574, and godotengine/godot-cpp#175, but the fix will be done in the godot-cpp repo using these 2 functions.

[akien-mga]

CC @godotengine/gdnative

[touilleMan]

What about godot_object_is_instance or godot_object_is_instance_of instead ?

The _ptr suffix in godot_object_is_class_ptr seems to imply we are checking 1rst parameter is somehow a alternative representation of the class pointer provided as 2nd parameter.
Renaming to godot_object_is_class doesn't seem right to me either given we an object is an instance of a class and not a directly a class.

[sheepandshepherd]

Is it fine to use a different name than the engine itself uses (just is_class_ptr) for GDNative? If so, I do agree godot_object_is_instance_of is better.

(is_class exists as well - it does the same thing as is_class_ptr, but with String comparisons instead of pointer comparisons.)

[touilleMan]

Is it fine to use a different name than the engine itself uses (just is_class_ptr) for GDNative ?

Ha ! I wasn't aware of this !

I guess my knowledge on the project naming convention falls short here, so I guess sticking to godot_object_is_class_ptr is fine.

is_class exists as well

I think it would be useful to also expose godot_object_is_class then ;-)

[sheepandshepherd]

is_class is exposed to scripts already, so we could even use it to fix casts without this PR. I'm a bit reluctant to have each cast doing multiple String comparisons - it doesn't use StringName internally for some reason - but that may just be premature optimization.

If we do still want to add a non-String-based safe cast, maybe this naming is clearer?

godot_object *godot_object_cast_to(const godot_object *p_object, void *p_class_tag);

It does the same thing, just returning p_object or NULL instead of true or false. This way, language binding authors don't need to know the details about is_class_ptr - the function will only ever be used to reimplement Object::cast_to.

[akien-mga]

Having access to Object::cast_to directly instead of is_class_ptr seems pretty good to me. If we're checking that the object is a class instance via a dynamic cast, we might as well get the casted object instead of just a boolean.

[sheepandshepherd]

The branch has been updated with the improved names and godot_string_name. The C++ binding change is ready as well - I'll do some more testing before making that PR tomorrow.

[akien-mga]

Code changes and exposed API seem good to me.

[touilleMan]

Good job ! 😃

[akien-mga]

Thanks!