tags:

views:

177

answers:

3

Hi,

I have a template class, C_Foo<T>, which is specialised in a number of ways.

struct Bar_Base { ... };
struct Bar_1 : public Bar_Base { ... };
struct Bar_2 : public Bar_Base { ... };
struct Bar_3 : public Bar_Base { ... };

class C_Foo<T> { ... };

class C_Foo_1 : public C_Foo<Bar_1> { ... };
class C_Foo_2 : public C_Foo<Bar_2> { ... };
class C_Foo_3 : public C_Foo<Bar_3> { ... };

And instantiations as follows:

C_Foo_1      foo1;
C_Foo_2      foo2;
C_Foo_3      foo3;

I have a set of common operations, all of which are defined on C_Foo, that I want to perform on foo1, foo2, and foo3. I've tried the following:

vector<C_Foo *>  v;
v.push_back(&foo1);
v.push_back(&foo2);
v.push_back(&foo3);

But I get compile errors, presumably because the compiler isn't sure how to go from a C_Foo_1 to a C_Foo.

Is it possible to do something like this? I want to be able to loop through foo1 .. fooN and perform the same operations on all of them, without having to copy and paste boilerplate code like so:

foo1.do_stuff();
foo2.do_stuff();
foo3.do_stuff();

Thanks for your help.

+4  A: 

The problem is that C_Foo cannot be instantiated because it requires template parameters.

You can make another base class, and have your set of common operations within that:

class C_FooBase { ... };

template<typename T>
class C_Foo<T> : public C_FooBase { ... };

class C_Foo_1 : public C_Foo<Bar_1> { ... };
class C_Foo_2 : public C_Foo<Bar_2> { ... };
class C_Foo_3 : public C_Foo<Bar_3> { ... };
strager
+6  A: 

You can do that, if the function does not depend on the template parameter:

// note: not a template
class C_Foo_Common {
public:
    virtual void do_stuff() = 0;
};

template<typename T>
class C_Foo : public C_Foo_Common { 
    virtual void do_stuff() {
        // do stuff...
    }
};

vector<C_Foo_Common *>  v;
v.push_back(&foo1);
v.push_back(&foo2);
v.push_back(&foo3);
// now, you can iterate and call do_stuff on them.

But if the function in C_Foo_Common needs to know the type T (for example to have another return type that depends on T), then that's not possible anymore. C_Foo<Bar_1> is a different type than C_Foo<Bar_2>. You can use discriminated unions instead. Those keep track about what is stored in them and are completely generic:

typedef boost::variant<
    C_Foo<Bar_1>*, C_Foo<Bar_2>*, C_Foo<Bar_3>*
> variant_type;
vector<variant_type> v;
v.push_back(&foo1);
v.push_back(&foo2);
v.push_back(&foo3);

The variant knows what it stores, and can call functions overloaded on the types of what can be stored in it. Read the documentation of boost::variant for more information on how to get at what the variants contain.

Johannes Schaub - litb
+1  A: 

This is because C_Foo<T> does not exist as a class. C_Foo<Bar1> is the real class and has nothing to do with a C_Foo<Bar2>, these are totally different classes. A template is a tool for the compiler to duplicate code, it is not a real class like generics can be in Java or C#.

If you look again at your code by thinking that C_Foo<Bar1>, C_Foo<Bar2> and C_Foo<Bar3> are totally different classes and have nothing in common, then you understand why you can't do what you want the way you want.

Like strager said, just use a real base class as the common class

Vincent Robert