While it's usually best to use pointers for member variables rather than storing them by value, sometimes it's better to store the member as a whole object, that is, to store it by value. Here are some situations where this is true:
- When the header file for the member's class has to be included anywayperhaps to satisfy some other part of the same class definition.
- When the member is a built-in type, like an integer.
- When the member is small.
- If you want to save memory from heap manager overhead, as when many objects of the class that holds this member are likely to be in existence at the same time. Note that typical heaps require at least a dozen bytes in overhead per object allocated, which may actually be larger than the memory actually used as payload in the heap block.
- In order to improve performance in allocating the object, as when objects of the class will be allocated and destroyed frequently.
- In special situations when the object is meant for use when it is unsafe to access the heap and must be allocated purely on the stack or into memory previously allocated by the user using the placement new syntax. An example of this would be C++ objects constructed at interrupt time in the classic Mac OS.
An important instance in which you cannot
store a member by value is when the member is being used polymorphically (especially when the base class contains a pure virtual functionthen the class declaration will fail to compile). You will usually want a pointer to the member's base class in this case.