Question

To have a sort of duck typing, I do

template<bool b>
struct A{
  static template<typename V> f1(V*, [other params]);     
  static template<typename V> f2(V*, [other params]);            
};

template<> template<typename T>
void A<false>::f1(V*, [other params]){}

template<> template<typename T>
void A<true>::f1(V*, [other params]){
   ...some code...
}  

template<int flags>
struct V{
  void f(){
     A<flags&Some compile time conditions>::f1 (this,[params]);
     A<flags&Some compile time conditions>::f2 (this,[params]); 
  } 
};

Do you think there is a more elegant solution, which is not Template class, function specialization
(I do not want to add an extra param to the functions)

I would like to do something like

template<int X> struct C{
 void f(){std::cout<<"C::f"<<std::endl;};
};


template<> struct C<0>{
};


template<int X> struct D{
 C<X> c;

 template<bool b>
 void f();

 void g(){
  f<X!=0>();
 }

};

template<>
template<int X> 
void D<X>::f<true>{
c.f();
};

template<int X>  
 template<>
 void D<X>::f<false>{};


int main(){
 D<3> ch;
 ch.g();

 D<0> cn;
 cn.g();

}

but this is not valid code, and I get error: template-id ‘f’ used as a declarator.

Is there a way to specialize a template function by the value of its non-type template parameter?

Answer 1

template<>
template<int X> 
void D<X>::f<true>(){
c.f();
};

template<int X>  
 template<>
 void D<X>::f<false>(){};

That is illegal (all attempts are). When you specialize a member function template, it’s enclosing class has to be specialized as well.

However, you can overcome that easily by wrapping your function inside a templated struct that would take its template arguments. Something like

template <int X, bool B>
struct DXF;

template <int X>
struct DXF<X, true>
{
  static void f() { // B is true!
  }
};

template <int X>
struct DXF<X, false>
{
  static void f() { // B is false!
  }
};

and call it with DXF<X, (X!=0)>::f().

However, it seems that you just want to specialize for X==0. In that case, you can just specialize:

template <>
void D<0>::f() {}

Note that f in that case is not a member template.

---

Another option you can go for is overloading. You could wrap your int in a parameter list of some template, like this:

template<int X> struct D{
 C<X> c;

 void f(std::true_type*) { ... true code ... }
 void f(std::false_type_*) { ... false code ... }
 void g(){
  f((std::integral_constant<bool, X!=0>*)0);
 }

_{Note that true_type and false_type are just typedefs of std::integral_constant<bool, true> and false, resp.}