Question

I have general question regarding the use of pointers vs. references in this particular scenario.

Let’s say that I have a function that is going to do some computation and store the value inside an object for later use by the caller. I can implement this by using either pointers or references.

Although, I would prefer using references because I trying avoiding pointers as much as possible, are there any pros/cons of one approach over the other.

The code using Pointers would be as follows:

Node*& computeNode() {  
  // Do some computation before creating a node object.  
  Node* newNode = new Node;  
  newNode->member1 = xyz;  
  newNode->member2 = abc;  
  // and so on ...  
  return newNode;  
}

The code using references could do something like this:

void computeNode(Node& newNode) {  
   // Do some computation before assigning values to the node object.  
   newNode.member1 = xyz;  
   newNode.member2 = abc;  
   // and so on.  
}

The differences that I can see are as follows:

1. When using the pointer method, the newNode object is allocated on the Heap. So, unless I call delete on it, it is not going to get deleted. However, in the reference method, whether newNode is allocated on the Heap/Stack depends on what the caller did to create the newNode object.

2. Whenever we use references, the number of arguments needed to pass to the function increases by at least 1. This is fine, only I find it a bit counter-intuitive to pass the return object also to a function call unless I name the function in such a way that it becomes obvious to the API user.

3. By using references, I can simulate the return of multiple objects. In the pointer method, I think I will have to wrap all the objects in another structure (like a pair class) and then return it. That increases the overhead.

However, I do not know if usually one is preferred over the other. And if there are any function naming conventions in C++ that let the developer know that he is supposed to pass the return object also as an argument.

Answer 1

The second approach is probably preferable because there is no possibility of a memory leak, in the event you forget to delete the returned pointer.

It’s usually good practice to code in such a way that each function or object which allocates heap memory also deallocates that memory. Your first example violates that practice, making it the function caller’s responsibility to deallocate the memory. This makes memory leaks more likely, because now every time the function is called there is another opportunity to forget to delete the returned pointer.

You may also want to consider returning the object by value (which will return a copy of the object) in cases where the size of the object is not that large. Even though this will require a copy to be created, if the object is not so large it won’t impact performance. (This method will become a lot more attractive in the future with C++0x move semantics.)