Objects are not unwrapped when returned through parameters

[mihe]

The unwrapping of objects, that typically happens when you return a godot-cpp Object* back to the engine, doesn't seem to happen when it's returned as part of a result struct that's passed as a parameter, such as the collider field on PhysicsServer3DExtensionRayResult that's passed to PhysicsDirectSpaceState3DExtension::_intersect_ray.

This leads to having to employ workarounds such as either manually unwrapping the object, by directly grabbing its internal _owner field, or calling the internal GDExtension function object_get_instance_from_id with the object's instance ID. In either case you're forced to also (incorrectly) cast the resulting pointer to a godot-cpp Object*, since the generated interface still expects that type.

Normally I'd include steps to reproduce the problem and a minimal repro project, but seeing as how this involves result structs for PhysicsDirectSpaceState3D you need quite a lot of boilerplate to even get to the point where you can call this. So I'm hoping the description will be sufficient for someone familiar with the space. If not, I'd be happy to elaborate further.

[fabriceci]

I experienced the same issue with the 2D equivalent, you have to use the mentioned workaround, because the cast returns a corrupted object.

[vnen]

I feel like the best solution is to not allow Object pointers in those kind of structs. Since those are passed straight away from Godot to GDExtension (and vice-versa) via pointers, relying on Objects will always have issues (objects in extension won't always match objects in engine, like it happens here in godot-cpp).

I don't think there's any easy way (if any) to automatically encode these, so you have to unravel the abstraction to properly work with the structs.

Another possible fix would be in the abstraction level so a pointer to a Godot Object is equivalent to a pointer of a Godot-cpp Object, but I can't think of a way how to do this.

[mihe]

I feel like the best solution is to not allow Object pointers in those kind of structs.

Seeing as how get_rest_info gets away with only returning the object's instance ID and not the actual pointer, I feel like the proper solution to this would be to simply remove the collider field from PhysicsServer3D::RayResult, PhysicsServer3D::ShapeResult and their 2D equivalents altogether, as it would make the APIs consistent between eachother. But I guess that isn't much of an option now with 4.0-stable out in the wild.

The more immediate solution would be to remove that field only for GDExtension and do the ObjectDB::get_instance call somewhere in PhysicsDirectSpaceState3DExtension as part of marshalling the result struct, similar to what's already happening with the exclude parameter for the physics queries.