Question

I have some class structure for the project I build for my company. At some point, I’ve seen a “usually unwanted” slicing of derived object could actually make my code efficient. Please observe:

class Base {
  private:
    int myType; //some constant here
  public:
    Base(): myType(1)
    { 
       if(someTest()) 
       { 
         myTest = 0; // signifies that this object is not working well for me
       }  // if the test succeeds, now I know this is a TYPE1 object
    }

    virtual bool someTest() { /* does make some validation tests */ } 
    int GetMyType() { return myType; }

    // some more virtual functions follow
}

class Derived: public Base {
  public:
    Derived() : Base() 
    { 
      if( someTest() /*calls the derived*/ ) 
      { 
        myType =2; // signifies that this is a TYPE2 object
        /* Do something more*/ 
      }  
    }

    virtual bool someTest() { /* does make OTHER validation tests */ } 
}

Now, if someTest() at Derived fails, this shows that my object is actually of Base type (myType = 1) or an faulty object (myType =0). Since these stuff are at constructors, and since I cannot use exception handling (embedded system); I thought of this:

{ 
  // this is the object holder test block
  // there is some latch - Base* - above = let's call it "ObjectHolder"
  extern Base* ObjectHolder;


  Derived * deriv = new Derived();
  int dType = deriv->GetMyType() ;
  if( dType == 1) // if test at base succeded but the one in derived failed
     Base = & static_cast<Base> (*deriv); // specifically slice the object
  else if(dType =2) 
     Base = deriv;
  else
     delete deriv; // this is a TYPE0 - nonworking- object so don't latch it

  // use other implemented methods to do work ...

}

Now, why have I got such a design? Well, while I was designing the classes, since the methods of inner workings of “Base” (type 1) and “Derived” (type 2) classes differ (and there is a possibilty of type 3,4,5…. objects that are also different in inside) and since I didn’t want to make and “if-then-else” check in every single method; I thought I’d make such a design and make the differing methods virtual so they could be called correctly (with thanks to polymorphism) whilst the common methods can be at some Base class.

But now, firstly, I’m not sure if that weird syntax ( & static_cast < Base > *deriv ) is correct (tested and seemed like working but I just want to make sure this is not because of luck); and secondly, I’m not sure if this is the “suggested method” to achive this – I kinda suspect that this syntax can cause memory leaks or something…

Changed code a little bit (also corrected syntax) to make the problem clearer.

Now, since the &(static_cast(*deriv)) is a faulty approach, I was thinking if I can make a “copy constructor” which bypasses the check at the Base class (which is actually the reason I try these stuff – I don’t want that test to be run in some cases). Like below:

class Base { 
    // other stuff is like above

    // cc:
    Base(const Base &other) : myType(other.myType)
    {
         // no test is performed here!
    }

    // like other one above
}

With this written, I think I can do that now at the test block:

{ 
   // the code above

   if( dType == 1) // if test at base succeded but the one in derived failed
     Base = new Base(*deriv); // specifically slice the object with cc

    // code goes on ... 
}

What about this?

Answer 1

Surely, if you want to create two types of objects, the correct way is to have a factory function that contains the “if (sometest)” and then creates the correct object type:

 Base* factory(...) 
 {
     if(sometest())
     {
         return new Derived;
     }
     else
     {
          return new Base;
     }
 }

That way, you don’t unnecessarily create the derived type of object when you actually want a Base type object. Bear in mind also that if you slice the object, your vtable
associated with Derived objects that have been sliced will still be a Derived vtable.

Edit: Here’s some code to show the problem with slicing:

#include <iostream>

using namespace std;

class Base
{
public:
    virtual void func() { cout << "Base" << endl; }
};


class Derived : public Base
{
public:
    virtual void func() { cout << "Derived" << endl; }
};


int main()
{
    Base *list[10];
    for(int i = 0; i < 10; i++)
    {
        if (i % 2)
        {
            list[i] = new Base;
        }
        else
        {
            list[i] = new Derived;
        }
    }

    list[2] = &*(static_cast<Base*>(list[2]));

    for(int i = 0; i < 10; i++)
    {
        list[i]->func();
    }
    return 0;
}

Output:

Derived
Base
Derived     <- should have changed to "Base" if questio was "right".
Base
Derived
Base
Derived
Base
Derived
Base

Using the syntax for “slicing” as given in the original question:
&(static_cast<Base>(*list[2])); (as per adapted to the code posted here), givesn an error in both clang++ and g++ compiler. The essence of the message is the same. This is clang’s variant:

error: taking the address of a temporary object of type 'Base'