Map of template functions

Question

I want to make a std::map with std::string keys that contains functions. These functions should be templates in order to work with different numeric values:

template <typename T> T (*pfunc)(T,T);

The map declarations looks like this:

std::map<std::string, pfunc> funcMap;

I know I need a template argument list for pfunc but I don't really know how to do it.

[EDIT]

Taking in count the comments. I would like to be able to: Create a function template as:

template <typename T> 
T myMax(T x, T y) 
{ 
   return (x > y)? x: y; 
} 

or

template <typename T> 
T myMin(T x, T y) 
{ 
   return (x < y)? x: y; 
} 

• Then add it to my map: funcMap["myMax"] = myMax; or funcMap["myMin"] = myMin;
• And use it as:

funcMap["myMax"]<int>(3,4);
funcMap["myMax"]<float>(3.1,4.5);
funcMap["myMin"]<int>(3,4);
funcMap["myMin"]<float>(3.1,4.5);

• Is this posible? any better ideas?

Answer 1

I think the best you will get is

template <typename T> 
T myMax(T x, T y) 
{ 
   return (x > y)? x: y; 
} 

template <typename T> 
T myMin(T x, T y) 
{ 
   return (x < y)? x: y; 
} 

template <typename T> using pfunc = T(*)(T, T);

template <typename T> std::map<std::string, pfunc<T>> funcMap = { 
    { "myMax", myMax }, 
    { "myMin", myMin } 
};

You will need to define all your function templates before defining funcMap, or limit yourself to a pre-determined set of types. I don't know of a way of populating infinite funcMap instantiations with infinite function template instantiations after it's definition.

Answer 2

You can use a generic lambda:

[](auto x, auto y) { return myMax(x, y); }

However, this is not a function pointer, but rather an object.

Answer 3

As we don't know much about your final goal, we can't say if there is an easy workaround to your problem. When you are in front of a kind of unsolvable problem, you can follow the pragmatic programmer principle called "Cutting the Gordian Knot" by asking yourself:

• Is there an easier way?
• Am I solving the right problem?
• Why is this a problem?
• What makes it hard?
• Do I have to do it this way?
• Does it have to be done at all?

That's said, I think Caleth solution is the most straightforward. If you want a solution with a single map containing the whole overload set, here is a proof of concept (WARNING: It has many flaws and works only in your case).

First, you need some helpers to detect if a class has the right function call operator overload or not:

namespace helpers {
    // simplify is_detected pattern (see https://en.cppreference.com/w/cpp/experimental/is_detected)
    template <typename Dummy, template <typename...> typename Op, typename... Args>
    struct is_detected : std::false_type {};

    template <template <typename...> typename Op, typename... Args>
    struct is_detected<std::void_t<Op<Args...>>, Op, Args...> : std::true_type {};

    template <template <typename...> typename Op, typename... Args>
    constexpr bool is_detected_v = is_detected<void, Op, Args...>::value;

    // Check if a class has an overloaded function call operator with some params
    template <typename T, typename... Args>
    using has_fcall_t = decltype(std::declval<T>()(std::declval<Args>()...));

    template <typename T, typename... Args>
    constexpr bool has_fcall_v = is_detected_v<has_fcall_t, T, Args...>;
}

Then, you define your basic numerical operations:

template <typename T>
struct Additionner {
    T operator()(T a, T b) {
        return a + b;
    }
};

template <typename T>
struct Multiplier{
    T operator()(T a, T b) {
        return a * b;
    }
};

template <typename T>
struct Incrementor {
    T operator()(T a) {
        return a++;
    }
};

The next step is to gather all the specialization of an operation you are interested in in a single class:

// Used to store many overloads for the same operations
template <typename... Bases>
struct NumOverloader : Bases...
{
    using Bases::operator()...;
};

Internally, we use a std::variant to simulate a kind of heterogeneous map. We wrap it into a class to provide a straightforward interface to use:

// wrapper around a variant that expose a universal function call operator
template <typename... Ts>
class NumDispatcher {
public:
    NumDispatcher() = default;

    template <typename T> // Fine tuning needed (see https://mpark.github.io/programming/2014/06/07/beware-of-perfect-forwarding-constructors/)
    NumDispatcher(T&& t) : m_impl(std::forward<T>(t)){
    }

    // visit the variant
    template <typename... Args>
    auto operator()(Args... args) {
        using type = std::common_type_t<Args...>;
        type t{};
        std::visit([&](auto&& visited) {
            using vtype = std::decay_t<decltype(visited)>;
            if constexpr(helpers::has_fcall_v<vtype, Args...>)
                t = std::forward<vtype>(visited)(args...);
            else
                throw std::runtime_error("bad op args");
        }, m_impl);
        return t;
    }
private:
    using Impl = std::variant<Ts...>;
    Impl m_impl;
};

The final step is to define your mapping:

// Here you need to know at compile-time your overloads
using MyIncrementors = NumOverloader<Incrementor<int>, Incrementor<unsigned>>;
using MyAdditionners = NumOverloader<Additionner<int>, Additionner<double>>;
using MyMultipliers = NumOverloader<Multiplier<int>, Multiplier<double>>;

using MyValueType = NumDispatcher<MyIncrementors, MyAdditionners, MyMultipliers>;
using MyMap = std::map<std::string, MyValueType>;

And then, you can play with it:

Num::MyMap m;
m["add"] = Num::MyAdditionners{};
m["mul"] = Num::MyMultipliers{};
m["inc"] = Num::MyIncrementors{};

auto d = m["add"](2.4, 3.4);
std::cout << d << std::endl;
// auto d2 = m["add"](1.3f, 2); // throw no overload match
std::cout << m["inc"](1) << std::endl;
//std::cout << m["inc"](1,1) << std::endl; // throw no overload match
std::cout << m["mul"](3, 2) << std::endl;

DEMO HERE.

Regards.