Question

Let’s assume I have a class which manages some resource in RAII way:

class C
{
   HANDLE hResource_;

   // prevent sharing the ownership over the resource among multiple instances of C
   C(const C&);
   C& operator=(const C&);

public:
   C() : hResource_(INVALID_HANDLE){}

   C(int arg1, const std::string& arg2,...)
   {
      ...
      allocResource(arg1, arg2, ...);
      ...
   }

   ~C
   {
      ...
      FreeResource(hResource_);
      hResource_ = INVALID_HANDLE;
      ...
   }

   void allocResource(int arg1, const std::string& arg2, ...)
   {
      if(hResource_ == INVALID_HANDLE)
      {
          hResource_ = AllocateResource(arg1, arg2,...);
      }
   }

   HANDLE handle() {return hResource_;}
};

Its constructor takes some parameters needed for resource allocation and I am able to create an instance of it, use it and let it live within some scope:

// some global function 
void goo()
{
   C c(123, "test");
   UseResource(c.handle(),...);
   ... 
}

Let’s say I want now an instance of C to be a member of some class, and want to delay allocation of the resource that happens in C‘s c-tor. This requires C‘s default c-tor and some C‘s member function which performs resource allocation (e.g. allocResource() which calls AllocateResource()).

class A
{
   C c_;

public:
   void foo1()
   {
      ...
      c_.allocResource(123, "test"); 
      UseResource(c_.handle(),...);
      ...
   }   

   void foo2()
   {
      ...         
      UseResource(c_.handle(),...);
      ...
   }   
};

By using dedicated function, we are exposing C‘s internals in some way which I don’t like.

My question is: Is this approach a common way to enable lazy initialization? Are there any alternatives?

---

EDIT: This is a possible class design regarding (MSalters’) suggestions below:

class C
{
   HANDLE hResource_;

   // prevent sharing the ownership over the resource 
   // among multiple instances of C
   C(const C&);
   C& operator=(const C&);

public:      

   // prevent object creation if resource cannot be acquired
   C(int arg1, const std::string& arg2,...)
   {          
      hResource_ = AllocateResource(arg1, arg2,...);

      // assumption: AllocateResource() returns 
      // INVALID_HANDLE in case of failure
      if(hResource_ == INVALID_HANDLE)
         throw resource_acquisition_exception();
   }

   ~C
   {
      ...
      FreeResource(hResource_);
      hResource_ = INVALID_HANDLE;
      ...
   }

   HANDLE handle() {return hResource_;}
};

class A
{
   std::unique_ptr<C> c_;

public:
   void foo1()
   {
      try
      {
         ...
         c_ = std::unique_ptr<C>(new C(123, "test"));
         UseResource(c_->handle(),...);
         ...
      }
      catch(const resource_acquisition_exception& exc)
      {
         ...
      }
      catch(...)
      {
         ...
      }
   }   

   void foo2()
   {
      ...         
      UseResource(c_->handle(),...);
      ...
   }   
};

Answer 1

No, this is not a common way to do RAII. In fact, it’s not RAII at all. If you can’t allocate the necessary resources for a C, don’t create a C.