Another design question

[sockman]

Hi all,
I have a design issue and want to get some feedback on solutions.

I have two classes, let's call them A and B. They have very similar interfaces and are 98% polymorphic. It is in the 2% that they differ which is causing me doubts.

Ideally I would like to derive both these classes from a parent C. Like I say above this will work beautifully for 98% of the time. The issue is that there are a couple of functions A can perform which B can't and vice versa.

Do you think it's still good practice to have these objects inherit from a common class and then deal with the small differences via another solution (eg. Visitor pattern - thanks JohnW)?
Cheers,
BJW

[Amleto]

erm, so put the differences in the child/specialised class

[Lindley]

The question you need to ask is----are you ever going to treat these two types as if they're actually the same, or do they simply share a lot of common functionality?

If you may at some point want to access their common functionality through a common interface, then public inheritance is absolutely the way to go.

If you don't have that requirement, then you can still move the shared functionality to a third class, but you then have two options, and which one is better depends largely on the nature of the shared capability: either you can give each class (say A and B) a *member* of type C, or you can make both A and B privately inherit from type C. They're similar relationships, but provide slightly different benefits.

[sockman]

@Amleto
Sorry I probably wasn't specific enough about the problem. I want to have a list of objects of type C and treat them generically. Which I can't completely do since there are a couple of operations which can only be done on A or B.
Cheers,
BJW

[sockman]

@Lindley
Yes, I do plan to treat them as the same type of object - they basically are. It's just that in their specialisation there are a couple of mutually exclusive operations.
Cheers,
BJW

[Russco]

when you need to, downcast with dynamic_cast to access those methods that dont arise from the common base.

[sockman]

Thanks Russco,
That was my initial plan. I just thought I'd see if anyone had other ideas. The problem I have with that method is I have to query the object to see what type it is. I was hoping to avoid that.
Cheers,
BJW

[monarch_dodra]

Another solution is to give the A-only functions to B too, except that when called on B, they do nothing.

And vice versa.

[sockman]

Hi monarch_dodra,
I considered this as an option, but I have a hunch it's going to to get me into hot water because whenever I have to perform one of these mutually exculsive operations I will have to call both the operation for A and then for B. I can just see it all becoming a bit messy with the error checking and updating of my program's GUI.
Cheers,
BJW

[monarch_dodra]

Hi monarch_dodra,
I considered this as an option, but I have a hunch it's going to to get me into hot water because whenever I have to perform one of these mutually exculsive operations I will have to call both the operation for A and then for B. I can just see it all becoming a bit messy with the error checking and updating of my program's GUI.
Cheers,
BJW

Well, to be perfectly fair, we only have A and B to go by, so we're just listing possible solutions. It's up to you to figure out which one best suits your needs.

I'm intrigued by "mutually exculsive operations". These things only A can do, do they have some kind of relationship with the things only B can do?

Depending on your needs, it might be you don't even need a separate A and B class. You could have a single C class with a method called "doSpecialExclusive", and a bool to define if it should choose the A exclusion or the B exclusion. That's just another solution.

[Russco]

Another possibility. I let it leak for brevity, its long enough.

#include <iostream>
#include <vector>
#include <algorithm>
#include <functional>

using namespace std;

struct Base
{
	virtual void interface() = 0;
	virtual void commonfunc() = 0;

	virtual ~Base() {}

};

template <class Derived> struct Shim : Base
{
    virtual void interface()
    {
        static_cast<Derived*>(this) -> implementation();
    }
};
 
struct Derived1 : Shim <Derived1>
{
	virtual void commonfunc()
	{
		cout << " Virtual in Derived1 called " << endl;
	}

    void implementation()
	{
		cout << " Derived1::implementation called" << endl;
	}
};

struct Derived2 : Shim <Derived2>
{
	virtual void commonfunc()
	{
		cout << " Virtual in Derived2 called" << endl;
	}

    void implementation()
	{
		cout << " Derived2::implementation called" << endl;
	}
};

struct Derived3 : Derived2
{
	virtual void commonfunc()
	{
		cout << " Virtual in Derived3 called" << endl;
	}

	void implementation()
	{
		cout << " Derived3::implementation called - Oh dear " << endl;
	}

};

int main()
{
	vector <Base*> vec;
	vec.push_back( new Derived1 );
	vec.push_back( new Derived2 );
	vec.push_back( new Derived3 );
	for_each( vec.begin(), vec.end(),mem_fun( &Base::commonfunc ));
	for_each( vec.begin(), vec.end(),mem_fun( &Base::interface ));
	//ignore memory leak
	return 0;
}