3

I have a base class that I made a template because I want to vary the type it takes for several functions, but I want to derive from these templated base classes. I want to store a vector of these classes. My idea was to create a non-templated base class above everything in the hierarchy, and use double dispatching to figure out the type. Am I doing this the "right way"?

Here's a code snippet of the scenario:

class FooBase
{
public:
    virtual void Accept( Visitor &v );
};

template<class T>
class Foo : public FooBase
{
public:
    virtual void DoThing( const T & );
    virtual void Accept( Visitor &v) = 0;

};

template<>
class Foo<Bar> : public FooBase
{
public:
    virtual void Accept( Visitor &v )
    {
        v.HandleBar( *this );
    }
};

template<>
class Foo<Baz> : public FooBase
{
public:
    virtual void Accept( Visitor &v )
    {
        v.HandleBaz( *this );
    }
};

// and many derived classes from Foo, Foo

Then in another class

class Visitor
{
public:
    virtual void HandleBar( Foo<Bar> &f ) = 0;
    virtual void HandleBaz( Foo<Baz> &f ) = 0;
};

class Manager : public Visitor
{
public:
    void AddFoo( FooBase& f )
    {
        a.push_back( f );
    }

    void RunAll()
    {
        for ( std::vector<std::shared_ptr<FooBase> >::iterator it = a.begin(); it != a.end(); ++it )
        {
            (*it)->Accept( *this );
            // do common action that doesn't depend on types
        }
    }

    virtual void HandleBar( Foo<Bar> &f )
    {
         Bar item = GetBarItemFunction(); // not shown
         f.DoThing( item );
    }
    virtual void HandleBaz( Foo<Baz> &f )
    {
         Baz item = GetBazItemFunction(); // not shown
         f.DoThing( item );
    }

private:
    std::vector<std::shared_ptr<FooBase> > a;
};

I just don't know if this is the "best" way to do it. I could use dynamic_casting, but that feels dirty. So is this a solid solution to the situation? Please advise (I hope I didn't leave any glaring syntax errors in the example)

(EDIT removed, was stupid error on my part)

Matt
  • 5,478
  • 9
  • 56
  • 95

1 Answers1

3

I think you've almost got it. I would write the visitor class like: 

class Visitor
{
public:
    virtual void HandleFoo( Foo<Bar> &f ) = 0;
    virtual void HandleFoo( Foo<Baz> &f ) = 0;
    //default implementation for unknown Foo types:
    virtual void HandleFoo( FooBase &f ) = 0; 
};

Now you don't need to specialize your templated Foo class and you can just write the following to work with all class T's your application may need. The correct overloaded HandleFoo function will be chosen based on the template type used in Foo. You will still have to add methods to your visitor class to avoid the default behavior being invoked. 

template<class T>
class Foo : public FooBase
{
public:
    virtual void DoThing( const T & );
    virtual void Accept( Visitor &v) {
        v.HandleFoo( *this );
    };
};
hifier
  • 457
  • 3
  • 9
  • wow I was really struggling with specialization (was trying reuse common code, etc.) Your suggestion is exactly what I need! Thanks! – Matt Apr 26 '11 at 04:21
  • Another question may be do you really need the visitor pattern here? can you just templatize the functions themselves? (instead of the whole class) – hifier Apr 26 '11 at 04:33
  • but how would I know whether to call `GetBarItemFunction()` or `GetBazItemFunction()`? That is the one point in the code where the different template specializations of Foo diverge... – Matt Apr 26 '11 at 04:54
  • I'm not sure if that will work for your case or not. It really depends if the template type conceptually contributes to Foo's type or if it only applies to certain actions. This is obviously dumb'd down so I you'll have to make that decision. One thing that could help is to think about what state Foo is maintaining -- is this state dependent on the template type? If so, then I think this is getting there, if not, then I think there might be a different way to go about it. – hifier Apr 26 '11 at 05:16