Question

I have the following problem (simplified version):

Interface X;
class Y : public X { ... }
class Z : public X { ... }

class A
{
    map<int, X> m;
    init() { X bla = new Y or Z; map.insert(...) }
}

I would like to specify either Y or Z at creation time of A, thus getting a different functionality based on type specified. Currently I could “templatize” A, but that would mean EVERY function I define for A will have to have the template specified, even though I need to use it only in one method. Also, I am using g++ 4.2.4, which means I can’t separate my implementation of functions from the header file.

Answer 1

… even though I need to use it only in one method …

What is usually done in this case is moving the common functionality, which is not templatized, to a base class, and defining a templatized A that inherits the base functionality. I.e.:

class A_base {
    // common things go here, they can be implemented in a .cpp file.
};

// This is the part that depends on T
template<class T>
class A : public A_base
{
    map<int, T> m;
    void init() { T *bla = new T; map.insert(...) }
};

This is a common technique to reduce code bloat. In case that A_base needs to call a method of A, you can make it virtual.