Is it possible to concatenate two strings of type `const char *` at compile time?

Here is a quick compile time string class:

template<std::size_t N>
struct ct_str
{
    char state[N+1] = {0};
    constexpr ct_str( char const(&arr)[N+1] )
    {
        for (std::size_t i = 0; i < N; ++i)
            state[i] = arr[i];
    }
    constexpr char operator[](std::size_t i) const { return state[i]; } 
    constexpr char& operator[](std::size_t i) { return state[i]; } 

    constexpr explicit operator char const*() const { return state; }
    constexpr char const* data() const { return state; }
    constexpr std::size_t size() const { return N; }
    constexpr char const* begin() const { return state; }
    constexpr char const* end() const { return begin()+size(); }

    constexpr ct_str() = default;
    constexpr ct_str( ct_str const& ) = default;
    constexpr ct_str& operator=( ct_str const& ) = default;

    template<std::size_t M>
    friend constexpr ct_str<N+M> operator+( ct_str lhs, ct_str<M> rhs )
    {
        ct_str<N+M> retval;
        for (std::size_t i = 0; i < N; ++i)
            retval[i] = lhs[i];
        for (std::size_t i = 0; i < M; ++i)
            retval[N+i] = rhs[i];
        return retval;
    }

    friend constexpr bool operator==( ct_str lhs, ct_str rhs )
    {
        for (std::size_t i = 0; i < N; ++i)
            if (lhs[i] != rhs[i]) return false;
        return true;
    }
    friend constexpr bool operator!=( ct_str lhs, ct_str rhs )
    {
        for (std::size_t i = 0; i < N; ++i)
            if (lhs[i] != rhs[i]) return true;
        return false;
    }
    template<std::size_t M, std::enable_if_t< M!=N, bool > = true>
    friend constexpr bool operator!=( ct_str lhs, ct_str<M> rhs ) { return true; }
    template<std::size_t M, std::enable_if_t< M!=N, bool > = true>
    friend bool operator==( ct_str, ct_str<M> ) { return false; }
};

template<std::size_t N>
ct_str( char const(&)[N] )->ct_str<N-1>;

you can use it like this:

template <class T>
constexpr auto get_arithmetic_size()
{
    if constexpr (sizeof(T)==1)
        return ct_str{"1"};
    if constexpr (sizeof(T)==2)
        return ct_str{"2"};
    if constexpr (sizeof(T)==4)
        return ct_str{"4"};
    if constexpr (sizeof(T)==8)
        return ct_str{"8"};
    if constexpr (sizeof(T)==16)
        return ct_str{"16"};
}

template <class T, std::enable_if_t<std::is_arithmetic<T>{}, bool> = true>
constexpr auto make_type_name()
{
    if constexpr (std::is_signed<T>{})
        return ct_str{"int"} + get_arithmetic_size<T>();
    else
        return ct_str{"uint"} + get_arithmetic_size<T>();
}

which leads to statements like:

static_assert(make_type_name<int>() == make_type_name<int32_t>());

passing.

Live example.

Now one annoying thing is that the length of the buffer is in the type system. You could add a length field, and make N be "buffer size", and modify ct_str to only copy up to length and leave the trailing bytes as 0. Then override common_type to return the max N of both sides.

That would permit you do pass ct_str{"uint"} and ct_str{"int"} in the same type of value and make the implementation code a bit less annoying.

template<std::size_t N>
struct ct_str
{
    char state[N+1] = {0};

    template<std::size_t M, std::enable_if_t< (M<=N+1), bool > = true>
    constexpr ct_str( char const(&arr)[M] ):
        ct_str( arr, std::make_index_sequence<M>{} )
    {}
    template<std::size_t M, std::enable_if_t< (M<N), bool > = true >
    constexpr ct_str( ct_str<M> const& o ):
        ct_str( o, std::make_index_sequence<M>{} )
    {}
private:
    template<std::size_t M, std::size_t...Is>
    constexpr ct_str( char const(&arr)[M], std::index_sequence<Is...> ):
        state{ arr[Is]... }
    {}
    template<std::size_t M, std::size_t...Is>
    constexpr ct_str( ct_str<M> const& o, std::index_sequence<Is...> ):
        state{ o[Is]... }
    {}
public:
    constexpr char operator[](std::size_t i) const { return state[i]; } 
    constexpr char& operator[](std::size_t i) { return state[i]; } 

    constexpr explicit operator char const*() const { return state; }
    constexpr char const* data() const { return state; }
    constexpr std::size_t size() const {
        std::size_t retval = 0;
        while(state[retval]) {
            ++retval;
        }
        return retval;
    }
    constexpr char const* begin() const { return state; }
    constexpr char const* end() const { return begin()+size(); }

    constexpr ct_str() = default;
    constexpr ct_str( ct_str const& ) = default;
    constexpr ct_str& operator=( ct_str const& ) = default;

    template<std::size_t M>
    friend constexpr ct_str<N+M> operator+( ct_str lhs, ct_str<M> rhs )
    {
        ct_str<N+M> retval;
        for (std::size_t i = 0; i < lhs.size(); ++i)
            retval[i] = lhs[i];
        for (std::size_t i = 0; i < rhs.size(); ++i)
            retval[lhs.size()+i] = rhs[i];
        return retval;
    }

    template<std::size_t M>
    friend constexpr bool operator==( ct_str lhs, ct_str<M> rhs )
    {
        if (lhs.size() != rhs.size()) return false;
        for (std::size_t i = 0; i < lhs.size(); ++i)
            if (lhs[i] != rhs[i]) return false;
        return true;
    }
    template<std::size_t M>
    friend constexpr bool operator!=( ct_str lhs, ct_str<M> rhs )
    {
        if (lhs.size() != rhs.size()) return true;
        for (std::size_t i = 0; i < lhs.size(); ++i)
            if (lhs[i] != rhs[i]) return true;
        return false;
    }
};

template<std::size_t N>
ct_str( char const(&)[N] )->ct_str<N-1>;

The function implementations now become:

template <class T>
constexpr ct_str<2> get_arithmetic_size()
{
    switch (sizeof(T)) {
        case 1: return "1";
        case 2: return "2";
        case 4: return "4";
        case 8: return "8";
        case 16: return "16";
    }

}

template <class T, std::enable_if_t<std::is_arithmetic<T>{}, bool> = true>
constexpr auto make_type_name()
{
    constexpr auto base = std::is_signed<T>{}?ct_str{"int"}:ct_str{"uint"};
    return base + get_arithmetic_size<T>();
}

which is a lot more natural to write.

Live example.

No, it's impossible. You can implement something like below (it's C++14).

#include <cmath>
#include <cstring>
#include <iostream>
#include <type_traits>

constexpr const char* name[] = {
  "uint1", "uint2", "uint4", "uint8", "uint16",
  "int1",  "int2",  "int4",  "int8",  "int16"
};

template <class T>
constexpr std::enable_if_t<std::is_arithmetic<T>::value, const char *> make_type_name() {
  return name[std::is_signed<T>::value * 5 +
    static_cast<int>(std::log(sizeof(T)) / std::log(2) + 0.5)];
}

static_assert(std::strcmp(make_type_name<int>(), make_type_name<int32_t>()) == 0);

int main() {
  std::cout << make_type_name<int>();
  return 0;
}

https://ideone.com/BaADaM

If you don't like using <cmath>, you may replace std::log:

#include <cstring>
#include <iostream>
#include <type_traits>

constexpr const char* name[] = {
  "uint1", "uint2", "uint4", "uint8", "uint16",
  "int1",  "int2",  "int4",  "int8",  "int16"
};

constexpr size_t log2(size_t n) {
  return (n<2) ? 0 : 1 + log2(n/2);
}

template <class T>
constexpr std::enable_if_t<std::is_arithmetic<T>::value, const char *> make_type_name() {
  return name[std::is_signed<T>::value * 5 + log2(sizeof(T))];
}

static_assert(std::strcmp(make_type_name<int>(), make_type_name<int32_t>()) == 0);

int main() {
  std::cout << make_type_name<int>();
  return 0;
}