Mixing interface and functional inheritance

[Kevin L]

Below is a code snippet that fails to compile under vc.net

class BaseInterface
{
public:
virtual void f() = 0;
};

class BaseImplementat ion : public BaseInterface
{
public:
void f(){}
};

class ExtraInterface : public BaseInterface
{
virtual void g() = 0;
};

class ExtraImplementa tion: public BaseImplementat ion, public
ExtraInterface
{
void g(){}
};

void func()
{
ExtraImplementa tion d;
}

The compiler complains that ExtraImplementa tion is an abstract class
and therefore cannot be initialized. Function f, despite being
defined in BaseImplementat ion, is undefined in ExtraImplementa tion.

I hope it's clear what I'm trying to do, i.e., maintaining two
inheritance lines that parallel each other. One line consists of
interfaces and the other implementation. Am I misguided in my
efforts?

Kevin

[Alf P. Steinbach]

Re: Mixing interface and functional inheritance

On 22 Sep 2003 17:39:26 -0700, inkdust@hotmail .com (Kevin L) wrote:
[color=blue]
>Below is a code snippet that fails to compile under vc.net
>
>class BaseInterface
>{
>public:
> virtual void f() = 0;
>};
>
>class BaseImplementat ion : public BaseInterface
>{
>public:
> void f(){}
>};
>
>class ExtraInterface : public BaseInterface
>{[/color]

-- public:

[color=blue]
> virtual void g() = 0;
>};
>
>class ExtraImplementa tion: public BaseImplementat ion, public
>ExtraInterfa ce
>{[/color]

-- public:

[color=blue]
> void g(){}
>};
>
>void func()
>{
> ExtraImplementa tion d;
>}
>
>The compiler complains that ExtraImplementa tion is an abstract class
>and therefore cannot be initialized. Function f, despite being
>defined in BaseImplementat ion, is undefined in ExtraImplementa tion.[/color]

The compiler is correct.

In Java you can inherit a function spec from an interface, and an
implementation of that function from a concrete class. In C++ the
two inheritance chains are distinct, functions in one don't map
onto functions in the other, unless you use virtual inheritance,
which is both messy and a tad inefficient.

Instead of using virtual inheritance (the tech level solution) I
recommend _factoring out_ the ExtraInterface, like this:


class ExtraInterface
{
public:
virtual void g() = 0;
};


Now isn't that nice?

[color=blue]
>I hope it's clear what I'm trying to do, i.e., maintaining two
>inheritance lines that parallel each other. One line consists of
>interfaces and the other implementation.[/color]

You can do that as shown above.

I do not recommend the virtual inheritance alternative.


[color=blue]
>Am I misguided in my efforts?[/color]

Not enough information to say, but probably not; however, it seems
you come from a Java or perhaps C# background, and need to ditch
much of the low-level details from the previous language, while
holding on the high-level concepts (which you'll have to _implement_,
some times by adhering strictly to some convention, in C++).

[Gianni Mariani]

Re: Mixing interface and functional inheritance

Kevin L wrote:[color=blue]
> Below is a code snippet that fails to compile under vc.net
>
> class BaseInterface
> {
> public:
> virtual void f() = 0;
> };
>
> class BaseImplementat ion : public BaseInterface
> {
> public:
> void f(){}
> };
>
> class ExtraInterface : public BaseInterface
> {
> virtual void g() = 0;
> };
>
> class ExtraImplementa tion: public BaseImplementat ion, public
> ExtraInterface
> {
> void g(){}
> };
>
> void func()
> {
> ExtraImplementa tion d;
> }
>
> The compiler complains that ExtraImplementa tion is an abstract class
> and therefore cannot be initialized. Function f, despite being
> defined in BaseImplementat ion, is undefined in ExtraImplementa tion.
>
> I hope it's clear what I'm trying to do, i.e., maintaining two
> inheritance lines that parallel each other. One line consists of
> interfaces and the other implementation. Am I misguided in my
> efforts?[/color]

The virtual inheritance mechanism is what is intended to solve the problem.

class BaseInterface
{
public:
virtual void f() = 0;
};

class BaseImplementat ion : public *virtual* BaseInterface
{
public:
void f(){}
};

class ExtraInterface : public *virtual* BaseInterface
{
virtual void g() = 0;
};

class ExtraImplementa tion: public BaseImplementat ion, public
ExtraInterface
{
void g(){}
};

void func()
{
ExtraImplementa tion d;
}

[Kevin L]

Re: Mixing interface and functional inheritance

alfps@start.no (Alf P. Steinbach) wrote in message news:<3f6f98aa. 907775187@News. CIS.DFN.DE>...[color=blue]
> On 22 Sep 2003 17:39:26 -0700, inkdust@hotmail .com (Kevin L) wrote:
>[color=green]
> >Below is a code snippet that fails to compile under vc.net
> >
> >class BaseInterface
> >{
> >public:
> > virtual void f() = 0;
> >};
> >
> >class BaseImplementat ion : public BaseInterface
> >{
> >public:
> > void f(){}
> >};
> >
> >class ExtraInterface : public BaseInterface
> >{[/color]
>
> -- public:
>
>[color=green]
> > virtual void g() = 0;
> >};
> >
> >class ExtraImplementa tion: public BaseImplementat ion, public
> >ExtraInterfa ce
> >{[/color]
>
> -- public:
>
>[color=green]
> > void g(){}
> >};
> >
> >void func()
> >{
> > ExtraImplementa tion d;
> >}
> >[/color][/color]
....[color=blue]
>
> Instead of using virtual inheritance (the tech level solution) I
> recommend _factoring out_ the ExtraInterface, like this:
>
>
> class ExtraInterface
> {
> public:
> virtual void g() = 0;
> };
>
>[/color]
....
Ah, but now I can't pass an object of ExtraImplementa tion as
ExtraInterface and have the BaseInterface functions available without
casting.

Now if I stick to strict abstract classes for the interfaces (no data
members), I can sidestep much of the messiness involved in virtual
inheritance, correct? The inheritance tree I am building is meant for
exception classes. I expect them to be constructed and used quite
infrequently. Right now I'm leaning toward the virtual inheritance
solution. Any reason not to?

Thanks for the reply.

Kevin

[Alf P. Steinbach]

Re: Mixing interface and functional inheritance

On 23 Sep 2003 00:58:28 -0700, inkdust@hotmail .com (Kevin L) wrote:
[color=blue]
>alfps@start. no (Alf P. Steinbach) wrote in message news:<3f6f98aa. 907775187@News. CIS.DFN.DE>...[color=green]
>> On 22 Sep 2003 17:39:26 -0700, inkdust@hotmail .com (Kevin L) wrote:
>>[color=darkred]
>> >Below is a code snippet that fails to compile under vc.net
>> >
>> >class BaseInterface
>> >{
>> >public:
>> > virtual void f() = 0;
>> >};
>> >
>> >class BaseImplementat ion : public BaseInterface
>> >{
>> >public:
>> > void f(){}
>> >};
>> >
>> >class ExtraInterface : public BaseInterface
>> >{[/color]
>>
>> -- public:
>>
>>[color=darkred]
>> > virtual void g() = 0;
>> >};
>> >
>> >class ExtraImplementa tion: public BaseImplementat ion, public
>> >ExtraInterfa ce
>> >{[/color]
>>
>> -- public:
>>
>>[color=darkred]
>> > void g(){}
>> >};
>> >
>> >void func()
>> >{
>> > ExtraImplementa tion d;
>> >}
>> >[/color][/color]
>...[color=green]
>>
>> Instead of using virtual inheritance (the tech level solution) I
>> recommend _factoring out_ the ExtraInterface, like this:
>>
>>
>> class ExtraInterface
>> {
>> public:
>> virtual void g() = 0;
>> };
>>
>>[/color]
>...
>Ah, but now I can't pass an object of ExtraImplementa tion as
>ExtraInterfa ce and have the BaseInterface functions available without
>casting.[/color]

First, you _can_ hardwire the relationship, via an accessor method
in ExtraInterface.

But, second, why would you want to do that?


[color=blue]
>Now if I stick to strict abstract classes for the interfaces (no data
>members), I can sidestep much of the messiness involved in virtual
>inheritance, correct? The inheritance tree I am building is meant for
>exception classes. I expect them to be constructed and used quite
>infrequently . Right now I'm leaning toward the virtual inheritance
>solution. Any reason not to?[/color]

1 In the most derived class you may have to call the constructor of
the root class.

2 Virtual inheritance mixed with non-virtual inheritance is very
messy. You're in for a maintainance nightmare.

3 It's a tad inefficient, also.