Did I understand correctly the point of Scott Meyers' example of std::weak_ptr?

Question

Effective Modern C++ (page 136) uses the following example to motivate std::weak_ptr. A cache is defined as an unordered map with weak pointers to objects as values. Whenever clients of this cache request an object (by key), the corresponding weak pointer is looked up and lock() is invoked on it. If the resulting std::shared_ptr is not null, it is returned. Otherwise, the object is re-loaded from an external database, entered into the cache and std::shared_ptr to it is returned.

Now the question: one might think that it would be possible to implement this without std::weak_ptr, but to store strong shared pointers as cache values instead. If use_count() of a strong pointer is equal to one, that means that all the client's pointers have been destroyed. Is the whole point of this example that using std::weak_ptr allows us to save memory by actually deleting the objects?

Answer 1

"Is the whole point of this example that using std::weak_ptr allows us to save memory by actually deleting the objects?"

Yes. Otherwise, the cache would keep the pointed-to resources allocated indefinitely. The space required for a control block, if allocated separately, is generally much less than the sort of resource that might require caching (e.g. textures, mesh data, even localized strings).

While it is possible to implement the same functionality using atomics and checking for a reference count of 1, this would require a custom handle implementation. std::weak_ptr and std::shared_ptr make the whole thing much less of an ordeal to prototype.