Policies can be used to customize almost any aspect of the implementation, as well as to alter the class interface. However, there are unique challenges that arise when we attempt to customize class constructors using policies.
As an example, let's consider another limitation of our current smart pointer. As it stands so far, the object owned by the smart pointer is always deleted when the smart pointer is deleted. If the smart pointer supports release, then we can call the release() member function and be wholly responsible for the deletion of the object. But how are we going to ensure this deletion? The most likely way is, we will let another smart pointer own it:
SmartPtr<C> p1(new C);SmartPtr<C> p2(&*p1); // ...
