views:

41

answers:

2

Background: I'm using a delegation technique to abstract access to arbitrary object methods, but I'm having some issues where the linker is concerned. Consider the following class, ContextNode.

template <class ObjectType, class GetType, class SetType>
class ContextNode: public ContextNodeBase {
  public:
    ContextNode(ObjectType* target,
                GetType (ObjectType::*getter)(void),
                void (ObjectType::*setter)(const SetType& ref)
    ): _target(target), _getter(getter), _setter(setter) { }

    virtual ~ContextNode(void) { }

    virtual void r(Datum& value) {
      value = (_target->*_getter)();
      return;
    }

    virtual void w(const Datum& value) {
      (_target->*_setter)(value);
      return;
    }

  private:
    ObjectType* _target;
    GetType (ObjectType::*_getter)(void);
    void (ObjectType::*_setter)(const SetType& ref);
};

The implementation of Datum is irrelevant. Also consider the trivial class Thing.

class Thing {
  public:
    Thing(void);
    ~Thing(void);

    int   getValue(void)         { return _value; }
    void  setValue(const int& x) { _value = x; }
  private:
    int _value;
};

The problem: I can construct instantiations of ContextNode like so.

Thing* thing = new Thing();
ContextNode<Thing,int,int>* cn = new ContextNode<Thing,int,int>(thing, &Thing::getValue, &Thing::setValue);

This works well for my needs. I run into issues though, with overloaded methods. Assume I had written instead:

class Thing {
  public:
    Thing(void);
    ~Thing(void);

    int   value(void)         { return _value; }
    void  value(const int& x) { _value = x; }
  private:
    int _value;
};

Thing* thing = new Thing();
ContextNode<Thing,int,int>* cn = new ContextNode<Thing,int,int>(thing, &Thing::value, &Thing::value);

This fails to link. The issue, I believe, is that the linker is attempting name-based resolution only, thus I see <unresolved overloaded function type> errors.

My question: is there some syntax sugar to explicitly specify to which of several overloaded methods I'm referring? I can't imagine that such a silly quirk would break such an elegant solution. I've been able to find nothing online, nor on the C++ FAQ, nor right here on SO regarding the topic.

What's the fix, or am I hosed?

+2  A: 

You can use a cast to disambiguate an overloaded function name:

(int (Thing::*)(void))(&Thing::value)
(void (Thing::*)(const int&))(&Thing::value)
James McNellis
Great idea, but `error: address of overloaded function with no contextual type information`
Chris
Oh, and specifically, that error originates from the casting of the getter, not the setter.
Chris
Bah - my fault. I had some const-ness declarations wrong. Thanks!
Chris
+1  A: 

It might also be possible to provide utility functions that select one or another of the overloads.

//enjoy the syntax: function accepting and returning a pointer to member function
template <class Object, class T>
T (Object::*as_getter(T (Object::*f)()))()  
{
    return f;
}

template <class Object, class T>
void (Object::*as_setter(void (Object::*f)(T)))(T)
{
    return f;
}

class Thing {
  public:
    Thing(void);
    ~Thing(void);

    int   value(void)         { return _value; }
    void  value(const int& x) { _value = x; }
  private:
    int _value;
};

template <class F>
void foo(F f)
{}

int main()
{
    foo(as_getter(&Thing::value)); //selects methods of kind X Y::zzz()
    foo(as_setter(&Thing::value)); //selects methods of kind void Y::zzz(X)
}

However, getters normally are const methods, so as_getter might also want to deal with that.

Also, if possible, avoid needless overloading, particularly if you want to work with function pointers. IMO, a getter and a setter do sufficiently different things to deserve a different name.

UncleBens
Oh, man - great idea! And if I inline them, once the template types get resolved the compiler should basically able to optimize the call right out, leaving virtually no additional runtime overhead. Awesome abstraction! +1!
Chris