@pcwalton @Manishearth @asajeffrey I recall discussing this topic in Berlin, but I can't find the notes I thought I took. Attempting to reconstruct from memory:

• there exist DOM objects that have Drop impls for various reasons
• there was agreement that it should be possible to run some code in response to a GC finalizer (eg. webgl objects need to be able to unreference their graphics resources)
• the easiest way to ban Drop impls is to make the #[dom_struct] macro also add an empty Drop impl, which will conflict with any manual implementation
• for any code which does require a valid Drop, we can create inner structs that only hold the data required for that implementation and nothing more

Does this make sense? My memory is haziest when it comes to the last point.