So this would risk turning a compile-time error into a runtime error.
Another ! makes sense to me here. Are there any cases where it doesn't work to auto-assign ! to all associated types of a ! trait impl? Associated constants might require some mechanism similar to `compile_error!()`.