Error with visitor to variant and returning different types

Question

#include <iostream>
#include <vector>
#include <variant>

template<class... Ts> struct overload : Ts... { using Ts::operator()...; };
template<class... Ts> overload(Ts...) -> overload<Ts...>;

class data_output
{
public:
    double create_data() { return 1.57;}
};

class data_output_two
{
public:
    int create_data() { return 66;}
};



int main()
{
    using my_type = std::variant<data_output, data_output_two>;

    std::vector<my_type> data_vec;

    auto my_lambda = overload{[](data_output& d) {return d.create_data();},
                              [](data_output_two& d) {return d.create_data();}};

    data_vec.emplace_back(data_output{});
    data_vec.emplace_back(data_output_two{});

    std::cout << std::visit(my_lambda, data_vec[0]) << "\n";
}

With this I get the error

/usr/include/c++/8.2.1/variant:836:43: error: 
invalid conversion from ‘std::__success_type<int>::type (*)(overload<main()::<lambda(data_output&)>, main()::<lambda(data_output_two&)> >&, std::variant<data_output, data_output_two>&)’ 
{aka ‘int (*)(overload<main()::<lambda(data_output&)>, main()::<lambda(data_output_two&)> >&, std::variant<data_output, data_output_two>&)’}
to ‘double (*)(overload<main()::<lambda(data_output&)>, main()::<lambda(data_output_two&)> >&, std::variant<data_output, data_output_two>&)’ [-fpermissive]
     { return _Array_type{&__visit_invoke}; }

If I try return an int from both data_output and data_output_two it compiles fine. What am I overlooking here? Here is a link to a godbolt

Edit:

I would like to get able to do std::visit(gen_visit, vec[0]) and get a double and then possibly do a std::visit(get_visit, vec[1]) and get a int. Is that possible?

Answer 1

Visit returns a single type.

Your code is asking it to return two different types.

It cannot do that.

Now, the single type it returns could be in turn a variant over two types. But tht mostly kicks the can down the road.

{[](data_output& d)->std::variant<int,double> {return d.create_data();},
                          [](data_output_two& d)->std::variant<int, double> {return d.create_data();}};

with this you'll have to revisit in order to print:

std::visit( [](auto&&x){std::cout <<  x << "\n";}, std::visit(my_lambda, data_vec[0]) );

Now you can write some metaprogramming to deduce the proper return type of the composit overload/visit.

But more practically simply do the processing in the first visit:

std::visit([&](auto&&x){ std::cout<<my_lambda(x)<<"\n";}, data_vec[0] );