Confusion regarding lifetimes

[JuanDiegoMontoya]

(I'm using v5.2.0)

I have some code that creates a sphere shape, then uses it in creating one or more bodies. There are two ways of creating a sphere shape as far as I can tell.

1:

auto sphereSettings = JPH::SphereShapeSettings(1);
sphereSettings.SetEmbedded();
auto sphere = sphereSettings.Create().Get(); // `sphere` is Ref<Shape>

2:

auto sphere = JPH::SphereShape(1); // `sphere` is on the stack

However, when sphere goes out of scope in the second version, an assert in ~RefTarget() is tripped. It doesn't seem to matter whether I call sphere.SetEmbedded();.

The assert no longer trips if I new the SphereShape (including if it's wrapped in a JPH::Ref).

I guess I have two main questions:

1. Is it possible to create a shape on the stack?
   • The docs say Simple shapes like spheres and boxes can be constructed immediately by simply new-ing them.. which implies that they cannot live on the stack. It's not clear why that is.
2. When am I supposed to call SetEmbedded? What happens if I call SetEmbedded unnecessarily or if I forget to call it (I predict that the former is a leak while the latter is a crash)?
   • The docs say Mark this class as embedded, this means the type can be used in a compound or constructed on the stack..
   • The hello world example has a comment next to a usage of SetEmbedded: A ref counted object on the stack (base class RefTarget) should be marked as such to prevent it from being freed when its reference count goes to 0.
   • It's not clear to me why it wouldn't work for shapes as they are refcounted objects.

[jrouwe]

You cannot create a shape on the stack if you want to attach that shape to a body. The body doesn't make a copy of the shape, it takes a reference. So unless you create and destroy a body within the scope where sphere exists, you'll get crashes.

That said, you can do things like:

SphereShape sphere(1.0f);
sphere.SetEmbedded();
physics_system->GetNarrowPhaseQuery().CollideShape(&sphere, ...);

because then sphere lives long enough for the narrow phase query to be done with it and it is more efficient than:

SphereShape *sphere = new SphereShape(1.0f);
physics_system->GetNarrowPhaseQuery().CollideShape(sphere, ...);
delete sphere;

[JuanDiegoMontoya]

I see now. I had a broken preconception of what shapes did internally, but this makes sense.

Thanks!