std_func_replacement.cpp

#include <iostream>
#include <assert.h>
#include <vector>

template<class T>
struct Func{};

// std::func replacement without dynamic memory allocation (captures are limited to 48 bytes)
// this could be extended to support captures >48 bytes... of course, with a bit more of logic
// and some state.
template<class R, class... Args>
struct Func<R(Args...)> {

    template<class U>
    Func(U &&callable) { set(std::move(callable)); }

    template<class U>
    Func& operator=(U &&callable) { set(std::move(callable)); return *this; }

    Func(Func&& other) noexcept {
        swap(std::move(other));
    }

    Func& operator=(Func &&other) noexcept {
        swap(std::move(other));
    }

    ~Func() {
        if (_destroy) _destroy(_space);
    }

    R operator()(Args... args) {
        return _func(_space, std::forward<Args>(args)...);
    }

    
    template<class U>
    void set(U &&callable) {
        assert(sizeof(U) <= sizeof(_space));
        new (_space)U(std::move(callable));
        _func = [](void *ptr, Args... args) -> R { return (*reinterpret_cast<U*>(ptr))(std::forward<Args>(args)...); };
        _destroy = [](void *ptr) { reinterpret_cast<U*>(ptr)->~U();};
    };

    void swap(Func &&other) {
        std::swap(_space, other._space);
        std::swap(_func, other._func);
        std::swap(_destroy, other._destroy);
    }

    R (*_func)(void *, Args...) = nullptr;
    void (*_destroy)(void *) = nullptr;
    uint8_t _space[64-sizeof(_func)-sizeof(_destroy)];
};

int main ()
{
    int f = 5.0f;
    std::vector<Func<float(int,int)>> callbacks;
    callbacks.push_back([](int a, int b) -> float { return a+b;});
    callbacks.push_back([f](int a, int b) -> float { return f+a+b;});
    for(auto &&f: callbacks) {
        printf("%f\n", f(1,2));
    }
    
}

Based on some conversations, this is an example of the same concept but for Callbacks that doesn't need to be stored

#include <stdio.h>
#include <utility>

template<class T> struct Callback;

template<class R, class... Args>
struct Callback<R(Args...)> {
    Callback() {}
    
    template<class F>
    Callback(F *f) {
        _func = [](void *data, Args... args) -> R {
            F *f = (F*)data;
            return (*f)(args...);
        };
        data = (void*)f;
    }

    template<class F>
    Callback(F &&func) : Callback(&func) {}

    R operator()(Args... args) {
        return _func(data, std::forward<Args>(args)...);
    }

    R (*_func)(void *, Args...) = nullptr;
    void *data = nullptr;
};

__attribute__((noinline))
void test(Callback<void(int)> callback, int times) {
    for (int i = 0; i < times; i++) {
        callback(i);
    }
}

int main(int argc, char **argv) {
    float f = 2.5f;
    auto lambda = [&](int i) { printf("%s %f\n", argv[i], f); };
    test(&lambda, argc);

    // TEST 2
    test([&](int i) { printf("%s %f\n", argv[i], f);} , argc);
}