templated class type conversion

[kezman]

Hey, I've created a templated Vector class and I want to be able to overload addition/multiplication operators etc. But I'm having a problem when combining different datatypes, for example Vector<double> * Vector<int>. The best solution I could think of was the following:

template <typename T, typename U>
inline friend double operator*(const Vector<T>& lhs, const Vector<U>& rhs)
{ return (lhs[0]*rhs[0] + lhs[1]*rhs[1]); }

This works for the usual case of having a T=double and U=int, or T=int and U=double. But it's not as general as I'd like. Is there a better solution to this (i'd assume) common problem?

[dwks]

Well, it's hard to know what type you'd want to end up with, which I suppose is your problem. I'd probably add that as a templatized parameter as well. Or: standard containers like std::vector provide a value_type typedef, so that you can use e.g.

#include <iostream>
#include <algorithm>
#include <vector>

template <typename VectorType>
inline typename VectorType::value_type dotProduct(const VectorType &one, const VectorType &two) {
    typename VectorType::value_type value = 0;
    for(typename VectorType::size_type i = 0;
        i < std::min(one.size(), two.size()); i ++) {
        
        value += one[i] * two[i];
    }
    return value;
}

int main() {
    std::vector<double> a, b;
    
    a.push_back(1);
    a.push_back(3);
    
    b.push_back(1);
    b.push_back(0.5);
    
    std::cout << dotProduct(a, b) << std::endl;
    
    return 0;
}

That's pretty ugly. But you get the idea. And if you wanted to be really general you could get even uglier:

#include <iostream>
#include <algorithm>
#include <vector>

template <typename TypeOne, typename TypeTwo, typename ResultType>
inline ResultType dotProduct(const TypeOne &one, const TypeTwo &two) {
    ResultType value = 0;
    for(std::size_t i = 0; i < std::min(one.size(), two.size()); i ++) {
        value += one[i] * two[i];
    }
    return value;
}

int main() {
    std::vector<double> a, b;
    
    a.push_back(1);
    a.push_back(3);
    
    b.push_back(1);
    b.push_back(0.5);
    
    std::cout << dotProduct<std::vector<double>, std::vector<double>, double>(a, b) << std::endl;
    
    return 0;
}

I'd definitely suggest that you let the types be inferred whenever possible, e.g. let the return type be the value_type of the first parameter or something. Also you'll notice that my code works for any data type (std::vector, maybe your Vector) that supports an [] operator and .size() and has a value_type. You can make your life easier if you assume a Vector.

Finally: I wouldn't overload a * operator if it has complicated semantics like copying the value type of the first argument. I'd probably call it by a custom name. But that's up to you.

[brewbuck]

This works for the usual case of having a T=double and U=int, or T=int and U=double. But it's not as general as I'd like. Is there a better solution to this (i'd assume) common problem?

In what way is it not general enough? This should work for any types T and U as far as I can tell.

[dwks]

I think the main problem is that it returns double. If you multiply a Vector<int> by a Vector<int> that might not be what you expect.

[phantomotap]

O_o

So you say you need a facility to selectively enlarge (as according to the implicit casting hierarchy) a set of types obtained from template parameters? There's a `Boost' for that!




Seriously, you can do this yourself building on the traits mechanism, but you better be ready for a lot tedious code. It is all easy code, but you can expect a lot of it.

*shrug*

Or just use `double' and demand that any types `T' and `U' be convertible to and from double.

Soma

[Elysia]

How about some C++0x magic?

#include <vector>

template <typename T, typename U>
auto operator * (const std::vector<T>& lhs, const std::vector<U>& rhs) -> decltype(lhs[0]*rhs[0] + lhs[1]*rhs[1])
{
	return (lhs[0]*rhs[0] + lhs[1]*rhs[1]); 
}

int main()
{
	std::vector<double> a;
	std::vector<int> b;
	a.push_back(1);
	a.push_back(1);
	b.push_back(2);
	b.push_back(2);
	double d = a*b;
	return 0;
}

[kezman]

Thanks for the suggestions guys. The C++0x auto feature is looking pretty attractive right now. Is it safe start using C++0x? Im assuming it'll be backwards compatible with the previous standard, so I'll be able to pick and choose which features I want to use within my code?

[Elysia]

Yeah, it's backwards compatible. We have two big compilers that supports these features, which are Visual C++ (2010) and GCC (4.x and up).