Overload by deriv class param; call w base class param

[ectoplasm]

Abstract base class A. Derived from A are class B and C, both concrete.

I hold a pointer to an object of class A, an instance which is either B
or C.

I have trouble with making the following work, while I think it should
be possible:

A *pA = new B; // One of...
A *pA = new C; // ...these two
DoSomething(pA); // Call overloaded method

This called method is overloaded, and looks like this:

void DoSomething(B *pB)
{
// Code to handle a B
}

void DoSomething(C *pC)
{
// Code to handle a C)
}

If this is not possible, what should I do? Is the following the only
way?

void DoSomething(A *pA)
{
if (pA->IsTypeOf(B))
// Code to handle a B
else if (pA->IsTypeOf(C))
// Code to handle a C
}

There must be something more "OO".

Thanks,
E.

 

[Alf P. Steinbach]

* ectoplasm:
Make DoSomething a virtual member function of class A.

 

[ectoplasm]

Alf P. Steinbach, thanks. But my example simplified things and I forgot
to mention: the methods I want to implement shall not be (virtual)
members of A. Are there other solutions?

 

[Alf P. Steinbach]

* ectoplasm:

Alf P. Steinbach, thanks. But my example simplified things and I forgot
to mention: the methods I want to implement shall not be (virtual)
members of A. Are there other solutions?

Depends what the problem is.

 

[ectoplasm]

The problem is, that MS VC++ gives a compile error

error C2664: cannot convert parameter 1 from 'A *' to 'B *'

on my call DoSomething(pA). Even though pA is a pointer to a B
instance.

Seems there is no way around this. A method can only take a parameter
of its type or a derived type. In other words, can only be an
(implicit) upcast. In my opinion, this is a failure of C++. Don't you
agree that what I want is not strange at all but is OO functionality
that naturally should be possible?

 

[Jack Klein]

Alf P. Steinbach, thanks. But my example simplified things and I forgot
to mention: the methods I want to implement shall not be (virtual)
members of A. Are there other solutions?

Why not?

 

[Jack Klein]

The problem is, that MS VC++ gives a compile error

error C2664: cannot convert parameter 1 from 'A *' to 'B *'

on my call DoSomething(pA). Even though pA is a pointer to a B
instance.

Seems there is no way around this. A method can only take a parameter
of its type or a derived type. In other words, can only be an
(implicit) upcast. In my opinion, this is a failure of C++. Don't you
agree that what I want is not strange at all but is OO functionality
that naturally should be possible?

No, it is not strange. C++ provides a simple method of doing it,
providing a virtual function in the base class. What is silly is your
arbitrary insistence that you must have functionality but won't use
the standard mechanism that provides it.

 

[Alf P. Steinbach]

* ectoplasm:

The problem is, that MS VC++ gives a compile error

error C2664: cannot convert parameter 1 from 'A *' to 'B *'

on my call DoSomething(pA). Even though pA is a pointer to a B
instance.

That is not the problem, it's the compiler telling you you've chosen a bad
way to solve the problem.

Seems there is no way around this. A method can only take a parameter
of its type or a derived type. In other words, can only be an
(implicit) upcast. In my opinion, this is a failure of C++. Don't you
agree that what I want is not strange at all but is OO functionality
that naturally should be possible?

What you seem to want is not OO, it's the opposite.

However, my earlier comment stands: make 'DoSomething' a virtual member
function of class A. You then wrote, "the [two] methods I want to implement
shall not be (virtual) members of A", but that's OK. Here's one way --
and note that it's just a technical solution, it _does not_ solve the real
problem, which is at the design level, not the language level, and unknown:

#include <iostream>
#include <ostream>

class B;
class C;

static void doSomething( B* )
{
std::cout << "doSomething( B* )" << std::endl;
}

static void doSomething( C* )
{
std::cout << "doSomething( C* )" << std::endl;
}

class A
{
private:
virtual void doSomething() = 0;
public:
static void doSomething( A* arg ) { arg->doSomething(); }
};

class B: public A
{
private:
virtual void DoSomething()
{
::doSomething( this );
}
};

class C: public A
{
private:
virtual void doSomething()
{
::doSomething( this );
}
};

void doSomething( A* arg )
{
A::doSomething( arg );
}

int main()
{
A* p = new C;
doSomething( p );
delete p;
}

 

[Zorro]

With reference to the line:

DoSomething(pA); // Call overloaded method

Let us say the compiler should do it. Then, which one of the two
overloads should it choose?

I am trying to say that the compiler cannot resolve the ambiguity, even
if it did implicit upcast of some sort.


Hope that helps.
Z.

 

[ectoplasm]

Even though it's been a few days, I still want to comment on this
because I think it still is like a "missing link" in C++ / OO; your
arguments are not convincing me.

You ask: why shall the method (DoSomething()) not be a virtual member
of A? Answer: the method I want to call is actually a constructor of a
dialog (CDialog class of MFC). You'll have to agree that it is not good
design to have dialogs (UI elements) inside of a class which abstracts
some entity (clear separation between data and visualisation). So I
want to be able to have a dialog for a B and C class instance; they
have a different appearance. Suppose the dialog class is simply named
Dialog:

class Dialog
{
Dialog (B *pB);
Dialog (C *pC);
~Dialog();
}

I want to hold a pointer to type A and when I construct the dialog, I
don't want to know whether it is really a B or C (derived) type. Only
at the moment a dialog needs to be shown, their difference becomes
apparent (inside of the dialog).

Then, Zorro, you say the compiler cannot know which overload to choose.
Right, but what about virtual calls, the vtable...? To make the above
work, the only thing the compiler might want to check if the method
DoSomething() has overrides for all types derived from A. Maybe also
have a default implementation at the base level (A).

Concluding, I wouldn't know why there cannot be overloads of (non
member) functions based on the parameter's common base class. You say
it is not OO, I say it definitely is. It is not a far-fetched idea and
you can easily find real world analogies for this situation.

 

[ectoplasm]

N.b.: the dialog constructor is analogous to the DoSomething() method;
I mixed them up in my last post. Sorry for that.

 

[Alf P. Steinbach]

* ectoplasm:

N.b.: the dialog constructor is analogous to the DoSomething() method;
I mixed them up in my last post. Sorry for that.

Well you have already got working code to solve your language-level problem,
so what is the fuzz all about?

 

[Zorro]

Then, Zorro, you say the compiler cannot know which overload to choose.

All I am saying is the following (using your class example).

Given that p is a pointer to base, what is expected from the line:

Dialog MyDiag(p);

Forget about the compiler. What would you do? Would you create an
instance using type B or C?

Regards,
(e-mail address removed)
http://www.zhmicro.com
http://distributed-software.blogspot.com