Avoid code duplication in a pair of almost identical base > derived classes

Question

I have a pair of base/derived classes which are almost identical, but not quite.

I could simply copy all the code of Base1 > Derived1 to create Base2 > Derived2, but that would be ugly, and would require making almost any modification twice.

Question: How can I share as much code as possible between the two pairs, to avoid code duplication?

---

I tried to create a small toy example that has most features of the actual problem. I want to avoid having duplicate code for the identical part of the interface of D1 and D2. If you want to see more of the actual problem, scroll to the end of the question.

#include <iostream>
using namespace std;


//////////// 1st PAIR ////////////

class B1 {
protected:
    string name;
public:
    B1() : name("B1") { } // constructors are different between B1 and B2

    void speak() { cout << name << endl; } // identical between B1 and B2
};

template<typename T>
class D1 : public B1 {
    T x; // identical between D1 and D2
public:
    D1(const T &a) { x = a + name.size(); } // refers to base class member

    int getX() { return x; } // identical between D1 and D2
    int nameLength() { return name.size(); } // accesses member of B, identical between D1 and D2

    // differences between D1 and D2 follow:
    int add(int i, int j) { return i+j; } // different signature between D1 and D2
    void more() {} // not present in D1
};


//////////// 2nd PAIR ////////////

class B2 {
protected:
    string name;
public:
    B2() : name("B2") { }

    void speak() { cout << name << endl; }
};

template<typename T>
class D2 : public B2 {
    T x; // identical between D1 and D2
public:
    D2(const T &a) { x = a + name.size(); }

    int getX() { return x; } // identical between D1 and D2
    int nameLength() { return name.size(); } // accesses member of B, identical between D1 and D2

    int add(int i, int j, int k) { return i+j+k; } // different signature between D1 and D2
};


// this is just to test that the program compiles and works
int main() {
    D1<int> d1(5);
    D2<long> d2(6l);
    cout << d1.getX();
    cout << d1.nameLength();
    return 0;
}

---

The interface of B1 and B2 can be shared by making them inherit from a class BInterface.

It was suggested to me to use multiple inheritance to be able to do the same with D1 and D2, through an additional base class DInterface. Furthermore, it was suggested I try to use the curiously recurring template pattern to allow this additional base class to access the members of D1 and D2. My attempt at doing this follows. I find it a bit complicated, and I would like to know if this is a reasonable approach, and whether there is a better way to do the same.

#include <iostream>
using namespace std;

//////////// COMMON INTERFACES ////////////

class BInterface {    
protected:
    string name;

    BInterface(const string &n) : name(n) { }

public:

    void speak() { cout << name << endl; } 
};

template<typename D>
class DInterface {
private:
    D &derived() { return *static_cast<D *>(this); }

protected:
    DInterface() {}

public:
    int getX() { return derived().x; }
    int nameLength() { return derived().name.size(); }
};


//////////// 1st PAIR ////////////

class B1 : public BInterface {
public:
    B1() : BInterface("B1") { } // constructors are different between B1 and B2
};

template<typename T>
class D1 : public B1, public DInterface< D1<T> > {
    friend class DInterface< D1<T> >;
    T x; // identical between D1 and D2
public:
    D1(const T &a) { x = a + name.size(); } // refers to base class member

    int add(int i, int j) { return i+j; } // different signature between D1 and D2

    void more() {} // not present in D1
};


//////////// 2nd PAIR ////////////

class B2 : public BInterface {
public:
    B2() : BInterface("B2") { }
};

template<typename T>
class D2 : public B2, public DInterface< D2<T> > {
    friend class DInterface< D2<T> >;
    T x; // identical between D1 and D2
public:
    D2(const T &a) { x = a + name.size(); }

    int add(int i, int j, int k) { return i+j+k; } // different signature between D1 and D2
};


// this is just to test that the program compiles and works
int main() {
    D1<int> d1(5);
    D2<long> d2(6l);
    cout << d1.getX();
    cout << d1.nameLength();
    return 0;
}

---

Since several people have commented that this is too broad, and that the context from my actual problem is lost, below I will describe the actual problem:

Mathematica has a C extension API. Certain data types, such as dense or sparse arrays or images can be manipulated in C. I am working on a much easier to use C++ interface. The system also includes in interface generator: a lot of glue code is automatically generated based on a symbolic representation of a C++ class interface in Mathematica. Here's an older version of the system.

I am now working on handling images. Mathematica has Image and Image3D, distinct expressions for 2D and 3D images. Image can also have different pixel types, such as byte, 16-bit, floating point, etc.

The C API uses a single representation for all of these, including 2D and 3D images, called MImage (a pointer type, multiple MImages may point to the same image in memory).

It is convenient for have separate classes for 2D and 3D images in C++, and also to template these on the pixel type. These correspond to the D1 and D2 classes above. However, in some cases, it is useful to operate with "generic" images that may have any pixel type (pixels can't be accessed in this case, but we can do other things with the images too). This is why I also have the base classes B1 and B2.

Here's the implementation of 2D image references so far (this is not done and it will change). I still need to add 3D images, which will share a lot of code.

Answer 1

Where you want to inherit for code reuse you can use private inheritance. With private inheritance the derived classes are blocked from being cast to their base classes.

#include <string>
#include <iostream>

class super
{
  std::string name_;
public:
  super( std::string n ): name_(n) {}
  virtual ~super(){}
  std::string name() const { return this->name_; }
  void name( std::string n ) { this->name_ = n; }
};

class base1: private super
{
  int vertices_;
public:
  base1( std::string n, int v ): super( n ), vertices_( v ) {}
  virtual ~base1() {}
  using super::name; // make both name methods accessible
  int vertices() const { return this->vertices_; }
  void vertices( int v ) { this->vertices_ = v; }
};

class base2: private super
{
  std::string surname_;
public:
  base2( std::string n, std::string s ): super( n ), surname_( s ) {}
  virtual ~base2() {}

  // to make only one name method accessible
  std::string name() const { return this->super::name(); }
  std::string surname() const { return this->surname_; }
};

// class derived1: public base1 { ... };
// class derived2: public base2 { ... };

int main()
{
base1 v1( "triangle", 3 );
base2 v2( "john", "doe" );

std::cout << "base1: " << v1.name() << " " << v1.vertices() << "\n";
std::cout << "base2: " << v2.name() << " " << v2.surname() << "\n";
v1.name( "square" );
v1.vertices( 4 );
std::cout << "base1: " << v1.name() << " " << v1.vertices() << "\n";

//v2.name( "jane" ); // illegal code
//super *p1 = &v1; // illegal code
//super *p2 = &v2; // illegal code
//derived1 d1(...);
//derived2 d2(...);
//base1 *p1 = &d1; // allowed
//base2 *p2 = &d2; // allowed
//derived1 *p1 = dynamic_cast< derived1* >((super*)&d2); // Not allowed
return 0;
}

With private inheritance you cannot directly access any base class methods outside the derived class. You have two options to allow this: (1) In base1 we use a public using statement to make the two name methods accessible. (2) In base2 we only want one of the name functions so we write a stub method that calls the super class method (NOTE: as this is inline it should result in the same assembly code as the using method).

Answer 2

This solution factors out the concepts of having a name and having a value through a base class that has a name.

If the individual components of the derived classes do not depend on each other then this kind of inheritance composition is relatively easy to maintain.

If the concerns of the base classes are interdependent then you'll have to use CRTP and marshal calls via the derived class.

#include <iostream>
using namespace std;

// factor out common parts

struct NamedThing
{
    NamedThing(std::string &&name) : name(std::move(name)) {}
    NamedThing(std::string const& name) : name(name) {}

    void speak() { cout << name << endl; }
    std::size_t nameLength() const { return name.size(); }
private:
    std::string name;
};

template<class T, class Base>
struct NamedValue : public Base
{
    T x; // identical between D1 and D2

public:
    NamedValue(T const& v) 
    : Base()
    , x(this->nameLength())
    {}

    T getX() { return x; } // identical between D1 and D2

};

//////////// 1st PAIR ////////////

class B1 : public NamedThing
{
public:
    B1() : NamedThing("B1") { } // constructors are different between B1 and B2
};

template<typename T>
class D1 : public NamedValue<T, B1> {
    using inherited = NamedValue<T, B1>;
public:
    D1(const T &a)
    : inherited(a)
    { 
    } 

    // differences between D1 and D2 follow:
    int add(int i, int j) { return i+j; } // different signature between D1 and D2
    void more() {} // not present in D1
};


//////////// 2nd PAIR ////////////

class B2 : public NamedThing
{
public:
    B2() : NamedThing("B2") { }
};

template<typename T>
class D2 : public NamedValue<T, B2> {
    using inherited = NamedValue<T, B2>;
public:
    D2(const T &a)
    : inherited(a)
    { 
    }

    int add(int i, int j, int k) { return i+j+k; } // different signature between D1 and D2
};


// this is just to test that the program compiles and works
int main() {
    D1<int> d1(5);
    D2<long> d2(6l);
    cout << d1.getX();
    cout << d1.nameLength();
    return 0;
}