Universal aggregate initialization by variadic templates - c++

This is an interesting problem I'm thinking about a time ago.
Given a struct with an underlying aggregate:
#include <array>
template <typename T, size_t N>
struct A
{
constexpr A() = default;
template <typename ... Ts>
constexpr A(const T& value, const Ts& ... values); // magic
std::array<T, N> arr; // aggregate
};
How would you implement variadic template constructor A(const T& value, const Ts& ... values) to
accept both values of type T and another A<T, N>
properly initialize the underlying aggregate based on the values represented by the passed arguments
respect the capacity of the aggregate
support C++14 constexpr rules and do not introduce any runtime overhead
Satisfying the above requirements, it is possible to do the following:
int main()
{
A<int, 3> x(1, 2, 3);
A<int, 2> y(1, 2);
A<int, 6> a(x, 1, 2, 3);
A<int, 6> b(1, x, 2, 3);
A<int, 6> c(1, 2, x, 3);
A<int, 6> d(1, 2, 3, x);
A<int, 6> e(x, x);
A<int, 6> f(y, y, y);
return 0;
}

Here's one approach that works, but could almost certainly be improved on.
We have a constructor for A that takes a parameter pack, converts each element into a tuple, concatenates the tuples together to make for one large tuple, and then simply uses aggregate initialization from that big tuple. All of the following can be constexpr, I just omitted it for brevity.
First we do the conversion:
template <class... Us>
A(Us const&... us)
: A(std::tuple_cat(as_tuple(us)...))
{ }
With:
// single argument
template <class U>
auto as_tuple(U const& u) {
return std::forward_as_tuple(u);
}
// aggregate argument
template <size_t M>
auto as_tuple(A<T, M> const& a) {
return as_tuple(a, std::make_index_sequence<M>{});
}
template <size_t M, size_t... Is>
auto as_tuple(A<T, M> const& a, std::index_sequence<Is...> ) {
return std::forward_as_tuple(std::get<Is>(a.arr)...);
}
And then we just initialize from there:
template <class... Us, class = std::enable_if_t<(sizeof...(Us) <= N)>>
A(std::tuple<Us...> const& t)
: A(t, std::index_sequence_for<Us...>{})
{ }
template <class... Us, size_t... Is>
A(std::tuple<Us...> const& t, std::index_sequence<Is...> )
: arr{{std::get<Is>(t)...}}
{ }
Demo

The answer by #Barry is certainly correct and acceptable. But it requires some C++14 library additions (which you could probably also write yourself in C++11), and overall requires some good tuple- and meta-programming fu.
Let's view multiple arguments a "range of ranges", where a range is just a pointer and a size. Scalar arguments are just a size-1 range, and A<T, N> arguments are size-N ranges.
template<class T>
struct Range
{
T const* data_;
std::size_t size_;
constexpr T const* begin() const noexcept { return data_; }
constexpr T const* end() const noexcept { return data_ + size_; }
constexpr std::size_t size() const noexcept { return size_; }
};
template<class T>
constexpr Range<T> as_range(T const& t)
{
return { &t, 1 };
}
template<class T, std::size_t N>
struct A;
template<class T, std::size_t N>
constexpr Range<T> as_range(A<T, N> const& a)
{
return { a.arr, N };
}
You can then simply do a double loop over all elements of all ranges
template <typename T, size_t N>
struct A
{
T arr[N]; // aggregate
constexpr A() = default;
template <typename U, typename... Us>
constexpr A(U const u, Us const&... us)
:
arr{}
{
Range<T> rngs[1 + sizeof...(Us)] { as_range(u), as_range(us)... };
auto i = 0;
for (auto const& r : rngs)
for (auto const& elem : r)
arr[i++] = elem;
assert(i == N);
}
};
Live Example working at compile-time (requires GCC >= 6.0 or Clang >= 3.4)
template <class T, size_t N>
void print(A<T, N> const& a) {
for (T const& t : a.arr) {
std::cout << t << ' ';
}
std::cout << '\n';
}
int main()
{
constexpr A<int, 3> x(1, 2, 3);
constexpr A<int, 2> y(1, 2);
constexpr A<int, 6> a(x, 1, 2, 3);
constexpr A<int, 6> b(1, x, 2, 3);
constexpr A<int, 6> c(1, 2, x, 3);
constexpr A<int, 6> d(1, 2, 3, x);
constexpr A<int, 6> e(x, x);
constexpr A<int, 6> f(y, y, y);
print(a); // 1 2 3 1 2 3
print(b); // 1 1 2 3 2 3
print(c); // 1 2 1 2 3 3
print(d); // 1 2 3 1 2 3
print(e); // 1 2 3 1 2 3
print(f); // 1 2 1 2 1 2
}

Related

How to make a function or constructor that accepts array of any rank

I have a wrapper struct around a vector and a shape, like this:
template <std::size_t N>
struct array_type {
std::array<std::size_t, N> shape;
std::vector<float> data;
};
I would like to be able to construct an array_type from any array float[N], float[N][M], etc.
Currently, I have a function for each rank, e.g.
template <std::size_t N>
array_type<1> _1d(const deel::formal::float_type (&values)[N]) {
return {{N},
std::vector<float_type>(
reinterpret_cast<const float_type*>(values),
reinterpret_cast<const float_type*>(values) + N)};
}
template <std::size_t N>
array_type<2> _2d(const deel::formal::float_type (&values)[N][M]) {
return {{N, M},
std::vector<float_type>(
reinterpret_cast<const float_type*>(values),
reinterpret_cast<const float_type*>(values) + N * M)};
}
I would like to write something like:
template <class Array>
array_type<std::rank_v<Array>> make_array(Array const&);
...but that does not work for initializer list:
auto arr1 = _1d({1, 2, 3}); // Ok
auto arr2 = make_array({1, 2, 3}); // Ko
Is there a way to have this make_array? Or (since I think it's not possible), to have at least something like make_array<1>({1, 2, 3}) where the rank is explicitly specified?
array_type::shape can be generated with std::extent with a bit of template meta programming:
template<typename Array, std::size_t... I>
auto extents_impl(const Array& a, std::index_sequence<I...>)
{
return std::array{std::extent_v<Array, I>...};
}
template<typename Array>
auto extents(const Array& a)
{
return extents_impl(a, std::make_index_sequence<std::rank_v<Array>>());
}
With this, make_array can be written as:
template <class Array, std::enable_if_t<std::is_same_v<std::remove_cv_t<std::remove_all_extents_t<Array>>, float>, int> = 0> // magic incantation :)
array_type<std::rank_v<Array>> make_array(Array const& a)
{
array_type<std::rank_v<Array>> ret {
.shape = extents(a),
.data = std::vector<float>(sizeof a / sizeof(float)),
};
std::memcpy(ret.data.data(), &a, sizeof a);
return ret;
}
I use memcpy to avoid potential pointer type aliasing violations and the technicality of iterating outside the bounds of sub array being UB according to strict interpretation of the standard.
...but that does not work for initializer list:
Add one overload:
template <std::size_t N>
array_type<1> make_array(float const (&a)[N])
{
return make_array<float[N]>(a);
}
Alternatively, you can specify the array type in the call:
make_array<float[2][3]>({{1, 2, 3},{4, 5, 6}});
This requires no overloads.
T (&&)[N] is suggested when you want a function template to accept a braced-init-list and deduce its length, furthermore, you can use T (&&)[M][N], T (&&)[M][N][O] ... to deduce the lengths of each rank (sadly, there is nothing like T (&&)[N]...[Z] to accept array with arbitrary rank).
so you can provide overloading function like this:
// attention that `T (&&)[N][M]` is `Y (&&)[N]` where `Y` is `T[M]`.
template<typename T, size_t N>
using Array1D_t = T[N];
template<typename T, size_t M, size_t N>
using Array2D_t = Array1D_t<Array1D_t<T, N>, M>;
template<typename T, size_t M, size_t N, size_t O>
using Array3D_t = Array1D_t<Array2D_t<T, N, O>, M>;
// ...
// assume the max rank is X.
template<typename T, size_t N>
array_type<1> make_array(Array1D_t<T, N>&& v);
template<typename T, size_t M, size_t N>
array_type<2> make_array(Array2D_t<T, M, N>&& v);
template<typename T, size_t M, size_t N, size_t O>
array_type<3> make_array(Array3D_t<T, M, N, O>&& v);
// ...
and then, it will be okay for make_array({1, 2, 3}), make_array({{1, 2, 3}, {4, 5, 6}}) ..., except the rank is larger than X.
Is there a way to have this make_array?
I don't think so (but, to be honest, I'm not able to demonstrate it's impossible).
Or (since I think it's not possible), to have at least something like make_array<1>({1, 2, 3}) where the rank is explicitly specified?
I don't see a way also in this case.
But... if you accept that the sizes are explicitly specified... you can write a recursive custom type traits to construct the final type
// declararion and ground case
template <typename T, std::size_t ...>
struct get_ranked_array
{ using type = T; };
// recursive case
template <typename T, std::size_t Dim0, std::size_t ... Dims>
struct get_ranked_array<T, Dim0, Dims...>
: public get_ranked_array<T[Dim0], Dims...>
{ };
template <typename T, std::size_t ... Dims>
using get_ranked_array_t = typename get_ranked_array<T, Dims...>::type;
and the make function simply become
template <std::size_t ... Dims>
auto make_ranked_array (get_ranked_array_t<float, Dims...> const & values)
-> array_type<sizeof...(Dims)>
{
return {{Dims...},
std::vector<float>(
reinterpret_cast<float const *>(values),
reinterpret_cast<float const *>(values) + (Dims * ...))};
}
You can use it as follows
auto arr1 = make_ranked_array<3u>({1.0f, 2.0f, 3.0f});
auto arr2 = make_ranked_array<3u, 2u>({ {1.0f, 2.0f, 3.0f},
{4.0f, 5.0f, 6.0f} });
If you want, you can maintain the deduction of the last size, but I don't know if it's a good idea.
The following is a full compiling C++17 example
#include <array>
#include <vector>
template <std::size_t N>
struct array_type
{
std::array<std::size_t, N> shape;
std::vector<float> data;
};
// declararion and ground case
template <typename T, std::size_t ...>
struct get_ranked_array
{ using type = T; };
// recursive case
template <typename T, std::size_t Dim0, std::size_t ... Dims>
struct get_ranked_array<T, Dim0, Dims...>
: public get_ranked_array<T[Dim0], Dims...>
{ };
template <typename T, std::size_t ... Dims>
using get_ranked_array_t = typename get_ranked_array<T, Dims...>::type;
template <std::size_t ... Dims>
auto make_ranked_array (get_ranked_array_t<float, Dims...> const & values)
-> array_type<sizeof...(Dims)>
{
return {{Dims...},
std::vector<float>(
reinterpret_cast<float const *>(values),
reinterpret_cast<float const *>(values) + (Dims * ...))};
}
int main ()
{
auto arr1 = make_ranked_array<3u>({1.0f, 2.0f, 3.0f});
auto arr2 = make_ranked_array<3u, 2u>({ {1.0f, 2.0f, 3.0f},
{4.0f, 5.0f, 6.0f} });
static_assert ( std::is_same_v<decltype(arr1), array_type<1>> );
static_assert ( std::is_same_v<decltype(arr2), array_type<2>> );
}

Determine member offset of struct or tuple in template

I want to write a template function that writes tables to HDF5 files.
The signature should look similar to
template<typename record> void writeTable(const std::vector<record>& data);
where record is a struct, or
template<typename... elements>
void writeTable(const std::vector<std::tuple<elements...>>& data);
The actual implementation would have more parameters to determine the destionation, etc.
To write the data I need to define a HDF5 compound type, which contains the name and the offset of the members. Usually you would use the HOFFSET macro the get the field offset, but as I don't know the struct fields beforehand I can't do that.
What I tried so far was constructing a struct type from the typename pack. The naive implementation did not have standard layout, but the implementation here does. All that's left is get the offsets of the members. I would like to expand the parameter pack into an initializer list with the offsets:
#include <vector>
template<typename... members> struct record {};
template<typename member, typename... members> struct record<member, members...> :
record<members...> {
record(member m, members... ms) : record<members...>(ms...), tail(m) {}
member tail;
};
template<typename... Args> void
make_table(const std::string& name, const std::vector<record<Args...>>& data) {
using record_type = record<Args...>;
std::vector<size_t> offsets = { get_offset(record_type,Args)... };
}
int main() {
std::vector<record<int, float>> table = { {1, 1.0}, {2, 2.0} };
make_table("table", table);
}
Is there a possible implementation for get_offset? I would think not, because in the case of record<int, int> it would be ambiguous. Is there another way to do it?
Or is there any other way I could approach this problem?
Calculating offsets is quite simple. Given a tuple with types T0, T1 ... TN. The offset of T0 is 0 (as long as you use alignas(T0) on your char array. The offset of T1 is the sizeof(T0) rounded up to alignof(T1).
In general, the offset of TB (which comes after TA) is round_up(offset_of<TA>() + sizeof(TA), alignof(TB)).
Calculating the offsets of elements in a std::tuple could be done like this:
constexpr size_t roundup(size_t num, size_t multiple) {
const size_t mod = num % multiple;
return mod == 0 ? num : num + multiple - mod;
}
template <size_t I, typename Tuple>
struct offset_of {
static constexpr size_t value = roundup(
offset_of<I - 1, Tuple>::value + sizeof(std::tuple_element_t<I - 1, Tuple>),
alignof(std::tuple_element_t<I, Tuple>)
);
};
template <typename Tuple>
struct offset_of<0, Tuple> {
static constexpr size_t value = 0;
};
template <size_t I, typename Tuple>
constexpr size_t offset_of_v = offset_of<I, Tuple>::value;
Here's a test suite. As you can see from the first test, the alignment of elements is taken into account.
static_assert(offset_of_v<1, std::tuple<char, long double>> == 16);
static_assert(offset_of_v<2, std::tuple<char, char, long double>> == 16);
static_assert(offset_of_v<3, std::tuple<char, char, char, long double>> == 16);
static_assert(offset_of_v<4, std::tuple<char, char, char, char, long double>> == 16);
static_assert(offset_of_v<0, std::tuple<int, double, int, char, short, long double>> == 0);
static_assert(offset_of_v<1, std::tuple<int, double, int, char, short, long double>> == 8);
static_assert(offset_of_v<2, std::tuple<int, double, int, char, short, long double>> == 16);
static_assert(offset_of_v<3, std::tuple<int, double, int, char, short, long double>> == 20);
static_assert(offset_of_v<4, std::tuple<int, double, int, char, short, long double>> == 22);
static_assert(offset_of_v<5, std::tuple<int, double, int, char, short, long double>> == 32);
I hardcoded the offsets in the above tests. The offsets are correct if the following tests succeed.
static_assert(sizeof(char) == 1 && alignof(char) == 1);
static_assert(sizeof(short) == 2 && alignof(short) == 2);
static_assert(sizeof(int) == 4 && alignof(int) == 4);
static_assert(sizeof(double) == 8 && alignof(double) == 8);
static_assert(sizeof(long double) == 16 && alignof(long double) == 16);
std::tuple seems to store it's elements sequentially (without sorting them to optimize padding). That's proven by the following tests. I don't think the standard requires std::tuple to be implemented this way so I don't think the following tests are guaranteed to succeed.
template <size_t I, typename Tuple>
size_t real_offset(const Tuple &tup) {
const char *base = reinterpret_cast<const char *>(&tup);
return reinterpret_cast<const char *>(&std::get<I>(tup)) - base;
}
int main(int argc, char **argv) {
using Tuple = std::tuple<int, double, int, char, short, long double>;
Tuple tup;
assert((offset_of_v<0, Tuple> == real_offset<0>(tup)));
assert((offset_of_v<1, Tuple> == real_offset<1>(tup)));
assert((offset_of_v<2, Tuple> == real_offset<2>(tup)));
assert((offset_of_v<3, Tuple> == real_offset<3>(tup)));
assert((offset_of_v<4, Tuple> == real_offset<4>(tup)));
assert((offset_of_v<5, Tuple> == real_offset<5>(tup)));
}
Now that I've gone to all of this effort, would that real_offset function suit your needs?
This is a minimal implementation of a tuple that accesses a char[] with offset_of. This is undefined behavior though because of the reinterpret_cast. Even though I'm constructing the object in the same bytes and accessing the object in the same bytes, it's still UB. See this answer for all the standardese. It will work on every compiler you can find but it's UB so just use it anyway. This tuple is standard layout (unlike std::tuple). If the elements of your tuple are all trivially copyable, you can remove the copy and move constructors and replace them with memcpy.
template <typename... Elems>
class tuple;
template <size_t I, typename Tuple>
struct tuple_element;
template <size_t I, typename... Elems>
struct tuple_element<I, tuple<Elems...>> {
using type = std::tuple_element_t<I, std::tuple<Elems...>>;
};
template <size_t I, typename Tuple>
using tuple_element_t = typename tuple_element<I, Tuple>::type;
template <typename Tuple>
struct tuple_size;
template <typename... Elems>
struct tuple_size<tuple<Elems...>> {
static constexpr size_t value = sizeof...(Elems);
};
template <typename Tuple>
constexpr size_t tuple_size_v = tuple_size<Tuple>::value;
constexpr size_t roundup(size_t num, size_t multiple) {
const size_t mod = num % multiple;
return mod == 0 ? num : num + multiple - mod;
}
template <size_t I, typename Tuple>
struct offset_of {
static constexpr size_t value = roundup(
offset_of<I - 1, Tuple>::value + sizeof(tuple_element_t<I - 1, Tuple>),
alignof(tuple_element_t<I, Tuple>)
);
};
template <typename Tuple>
struct offset_of<0, Tuple> {
static constexpr size_t value = 0;
};
template <size_t I, typename Tuple>
constexpr size_t offset_of_v = offset_of<I, Tuple>::value;
template <size_t I, typename Tuple>
auto &get(Tuple &tuple) noexcept {
return *reinterpret_cast<tuple_element_t<I, Tuple> *>(tuple.template addr<I>());
}
template <size_t I, typename Tuple>
const auto &get(const Tuple &tuple) noexcept {
return *reinterpret_cast<tuple_element_t<I, Tuple> *>(tuple.template addr<I>());
}
template <typename... Elems>
class tuple {
alignas(tuple_element_t<0, tuple>) char storage[offset_of_v<sizeof...(Elems), tuple<Elems..., char>>];
using idx_seq = std::make_index_sequence<sizeof...(Elems)>;
template <size_t I>
void *addr() {
return static_cast<void *>(&storage + offset_of_v<I, tuple>);
}
template <size_t I, typename Tuple>
friend auto &get(const Tuple &) noexcept;
template <size_t I, typename Tuple>
friend const auto &get(Tuple &) noexcept;
template <size_t... I>
void default_construct(std::index_sequence<I...>) {
(new (addr<I>()) Elems{}, ...);
}
template <size_t... I>
void destroy(std::index_sequence<I...>) {
(get<I>(*this).~Elems(), ...);
}
template <size_t... I>
void move_construct(tuple &&other, std::index_sequence<I...>) {
(new (addr<I>()) Elems{std::move(get<I>(other))}, ...);
}
template <size_t... I>
void copy_construct(const tuple &other, std::index_sequence<I...>) {
(new (addr<I>()) Elems{get<I>(other)}, ...);
}
template <size_t... I>
void move_assign(tuple &&other, std::index_sequence<I...>) {
(static_cast<void>(get<I>(*this) = std::move(get<I>(other))), ...);
}
template <size_t... I>
void copy_assign(const tuple &other, std::index_sequence<I...>) {
(static_cast<void>(get<I>(*this) = get<I>(other)), ...);
}
public:
tuple() noexcept((std::is_nothrow_default_constructible_v<Elems> && ...)) {
default_construct(idx_seq{});
}
~tuple() {
destroy(idx_seq{});
}
tuple(tuple &&other) noexcept((std::is_nothrow_move_constructible_v<Elems> && ...)) {
move_construct(other, idx_seq{});
}
tuple(const tuple &other) noexcept((std::is_nothrow_copy_constructible_v<Elems> && ...)) {
copy_construct(other, idx_seq{});
}
tuple &operator=(tuple &&other) noexcept((std::is_nothrow_move_assignable_v<Elems> && ...)) {
move_assign(other, idx_seq{});
return *this;
}
tuple &operator=(const tuple &other) noexcept((std::is_nothrow_copy_assignable_v<Elems> && ...)) {
copy_assign(other, idx_seq{});
return *this;
}
};
Alternatively, you could use this function:
template <size_t I, typename Tuple>
size_t member_offset() {
return reinterpret_cast<size_t>(&std::get<I>(*static_cast<Tuple *>(nullptr)));
}
template <typename Member, typename Class>
size_t member_offset(Member (Class::*ptr)) {
return reinterpret_cast<size_t>(&(static_cast<Class *>(nullptr)->*ptr));
}
template <auto MemPtr>
size_t member_offset() {
return member_offset(MemPtr);
}
Once again, this is undefined behavior (because of the nullptr dereference and the reinterpret_cast) but it will work as expected with every major compiler. The function cannot be constexpr (even though member offset is a compile-time calculation).
Not sure to understand what do you exactly want but... what about using recursion based on a index sequence (starting from C++14) something as follows?
#include <vector>
#include <utility>
#include <iostream>
template <typename... members>
struct record
{ };
template <typename member, typename... members>
struct record<member, members...> : record<members...>
{
record (member m, members... ms) : record<members...>(ms...), tail(m)
{ }
member tail;
};
template <std::size_t, typename, std::size_t = 0u>
struct get_offset;
template <std::size_t N, typename A0, typename ... As, std::size_t Off>
struct get_offset<N, record<A0, As...>, Off>
: public get_offset<N-1u, record<As...>, Off+sizeof(A0)>
{ };
template <typename A0, typename ... As, std::size_t Off>
struct get_offset<0u, record<A0, As...>, Off>
: public std::integral_constant<std::size_t, Off>
{ };
template <typename... Args, std::size_t ... Is>
auto make_table_helper (std::string const & name,
std::vector<record<Args...>> const & data,
std::index_sequence<Is...> const &)
{ return std::vector<std::size_t>{ get_offset<Is, record<Args...>>::value... }; }
template <typename... Args>
auto make_table (std::string const & name,
std::vector<record<Args...>> const & data)
{ return make_table_helper(name, data, std::index_sequence_for<Args...>{}); }
int main ()
{
std::vector<record<int, float>> table = { {1, 1.0}, {2, 2.0} };
auto v = make_table("table", table);
for ( auto const & o : v )
std::cout << o << ' ';
std::cout << std::endl;
}
Unfortunately isn't an efficient solution because the last value is calculated n-times.

C++ "forgetting" that variable is constexpr when used as function argument

I have the following code where I am irritated by the fact that compiler is unable to see that variable passed as argument to a function is constexpr so I must use arity 0 function instead of 1 argument function.
I know this is not a compiler bug, but I wonder if there are idioms that enable to workaround this problem.
#include <array>
#include <iostream>
static constexpr std::array<int, 5> arr{11, 22, 33, 44, 55};
template <typename C, typename P, typename Y>
static constexpr void copy_if(const C& rng, P p, Y yi3ld) {
for (const auto& elem: rng) {
if (p(elem)){
yi3ld(elem);
}
}
}
// template<std::size_t N>
static constexpr auto get_evens(/* const std::array<int, N>& arr */) {
constexpr auto is_even = [](const int i) constexpr {return i % 2 == 0;};
constexpr int cnt = [/* &arr, */&is_even]() constexpr {
int cnt = 0;
auto increment = [&cnt] (const auto&){cnt++;};
copy_if(arr, is_even, increment);
return cnt;
}();
std::array<int, cnt> result{};
int idx = 0;
copy_if(arr, is_even, [&result, &idx](const auto& val){ result[idx++] = val;});
return result;
}
int main() {
// constexpr std::array<int, 5> arr{11, 22, 33, 44, 55};
for (const int i:get_evens(/* arr */)) {
std::cout << i << " " << std::endl;
}
}
If it is not obvious what I want: I would like to change get_evens signature so that it is template templated on array size N and that it takes 1 argument of type const std::array<int, N>&.
The error message when I change arr to be an function argument isn't helpful:
prog.cc:25:21: note: initializer of 'cnt' is not a constant expression
prog.cc:19:19: note: declared here
constexpr int cnt = [&arr, &is_even]()constexpr {
A function argument is never a constant expression, even if a function is used in constexpr context:
constexpr int foo(int i)
{
// i is not a constexpr
return i + 1;
}
constexpr auto i = 1;
constexpr auto j = foo(i);
To mimic a constexpr argument, use a template parameter:
template<int i>
constexpr int foo()
{
// i is constexpr
return i + 1;
}
constexpr auto i = 1;
constexpr auto j = foo<i>();
A possible solution is to use std::integer_sequence to encode integers into a type:
#include <array>
#include <iostream>
#include <type_traits>
template<typename P, typename Y, int... elements>
constexpr void copy_if_impl(P p, Y yi3ld, std::integer_sequence<int, elements...>) {
((p(elements) && (yi3ld(elements), true)), ...);
}
template<typename arr_t, typename P, typename Y>
constexpr void copy_if(P p, Y yi3ld) {
copy_if_impl(p, yi3ld, arr_t{});
}
template<typename arr_t>
constexpr auto get_evens(){
constexpr auto is_even = [](const int i) constexpr { return i % 2 == 0; };
constexpr int cnt = [&is_even]() constexpr {
int cnt = 0;
auto increment = [&cnt](const auto&) { cnt++; };
copy_if<arr_t>(is_even, increment);
return cnt;
}();
std::array<int, cnt> result{};
int idx = 0;
copy_if<arr_t>(is_even, [&result, &idx](const auto& val) {
result[idx++] = val; });
return result;
}
int main()
{
using arr = std::integer_sequence<int, 11, 22, 33, 44, 55>;
for (const int i : get_evens<arr>()) {
std::cout << i << " " << std::endl;
}
}
Addition suggested by Constantinos Glynos.
From Effective Modern C++ book by Scott Meyers, item 15, p.98:
constexpr functions can be used in contexts that demand compile-time constants. If the values of the arguments you pass to a constexpr function in such a context are known during compilation, the result will be computed during compilation. If any of the arguments’ values is not known during compilation, your code will be rejected.
When a constexpr function is called with one or more values that are not known during compilation, it acts like a normal function, computing its result at runtime. This means you don’t need two functions to perform the same operation, one for compile-time constants and one for all other values. The constexpr function does it all.
The other answer has a correct work around but I think the reasoning has nothing to do with parameters but instead to do with the lambda capture here:
constexpr int cnt = [/* &arr, */&is_even]()
Indeed we can test the various scenarios with this code:
#include <array>
#include <iostream>
template <size_t N>
constexpr int foo(const std::array<int, N>& arr) {
return [&arr] () { return arr.size(); }();
}
template <size_t N>
constexpr int bar(const std::array<int, N>& arr) {
int res{};
for (auto i : arr) {
res++;
}
return res;
}
template <size_t N>
constexpr int baz(const std::array<int, N>& arr) {
constexpr int test = [&arr] () constexpr {
return bar(arr);
}();
return test;
}
int main() {
constexpr std::array<int, 5> arr{11, 22, 33, 44, 55};
constexpr std::array<int, foo(arr)> test{};
constexpr std::array<int, bar(arr)> test2{};
constexpr std::array<int, baz(arr)> test3{};
}
Note that the line where test3 is initialized fails to compile. This, however, compiles just fine:
template <size_t N>
constexpr int baz(const std::array<int, N>& arr) {
return bar(arr);
}
So, what's the problem here? Well lambdas are really just glorified functors, and internally it'll look something like this:
struct constexpr_functor {
const std::array<int, 5>& arr;
constexpr constexpr_functor(const std::array<int, 5>& test)
: arr(test) { }
constexpr int operator()() const {
return bar(arr);
}
};
// ...
constexpr constexpr_functor t{arr};
constexpr std::array<int, t()> test3{};
Notice now that we get an error message showing the real problem:
test.cpp:36:33: note: reference to 'arr' is not a constant expression
test.cpp:33:34: note: declared here
constexpr std::array<int, 5> arr{11, 22, 33, 44, 55};
The other answer quotes Scotts Meyer's book but misinterprets the quotes. The book actually shows several examples of parameters being used in constexpr situations, but the quotes are simply saying that if you pass a non-constexpr parameter, the function can run at compile-time.
Following the Evg's suggestion, so passing the numbers as template parameters of a std::integer_sequence, but passing the integer sequence as argument of the get_evens() function, and not as template parameter, you can use the numbers directly inside get_evens().
I mean... you can simplify the get_evens() as follows (EDIT: further simplified following a suggestion from Evg (Thanks!))
template <typename T, T ... Ts>
constexpr auto get_evens (std::integer_sequence<T, Ts...> const &)
{
std::array<T, (std::size_t(!(Ts & T{1})) + ...)> result{};
std::size_t idx = 0;
((void)(Ts & 1 || (result[idx++] = Ts, true)), ...);
return result;
}
and you can use it this way
int main()
{
using arr = std::integer_sequence<int, 11, 22, 33, 44, 55>;
for ( const int i : get_evens(arr{}) )
std::cout << i << " " << std::endl;
}
#include <array>
#include <iostream>
static constexpr std::array<int, 5> arr{11, 22, 33, 44, 55};
template <typename C, typename P, typename T>
static constexpr void invoke_if(const C& rng, P p, T target) {
for (const auto& elem: rng) {
if (p(elem)){
target(elem);
}
}
}
constexpr bool is_even(int i) {
return i % 2 == 0;
}
template<std::size_t N>
constexpr std::size_t count_evens(const std::array<int, N>& arr)
{
std::size_t cnt = 0;
invoke_if(arr, is_even, [&cnt](auto&&){++cnt;});
return cnt;
}
template<std::size_t cnt, std::size_t N>
static constexpr auto get_evens(const std::array<int, N>& arr) {
std::array<int, cnt> result{};
int idx = 0;
invoke_if(arr, is_even, [&result, &idx](const auto& val){ result[idx++] = val;});
return result;
}
int main() {
// constexpr std::array<int, 5> arr{11, 22, 33, 44, 55};
for (const int i:get_evens<count_evens(arr)>(arr)) {
std::cout << i << " " << std::endl;
}
}
this works in g++, but in clang we get a problem because the begin on an array isn't properly constexpr with at least one library. Or maybe g++ violates the standard and clang does not.
template<auto t0, auto...ts>
struct ct_array:
std::array<decltype(t0) const, 1+sizeof...(ts)>,
std::integer_sequence<decltype(t0), t0, ts...>
{
ct_array():std::array<decltype(t0) const, 1+sizeof...(ts)>{{t0, ts...}} {};
};
template<class target, auto X>
struct push;
template<auto X>
struct push<void, X>{using type=ct_array<X>;};
template<auto...elems, auto X>
struct push<ct_array<elems...>, X>{using type=ct_array<elems...,X>;};
template<class target, auto X>
using push_t= typename push<target, X>::type;
template<class target>
struct pop;
template<auto x>
struct pop<ct_array<x>>{using type=void;};
template<auto x0, auto...xs>
struct pop<ct_array<x0, xs...>>{using type=ct_array<xs...>;};
template<class target>
using pop_t=typename pop<target>::type;
template<class lhs, class rhs, class F, class=void>
struct transcribe;
template<class lhs, class rhs, class F>
using transcribe_t = typename transcribe<lhs, rhs, F>::type;
template<auto l0, auto...lhs, class rhs, class F>
struct transcribe<ct_array<l0, lhs...>, rhs, F,
std::enable_if_t<F{}(l0) && sizeof...(lhs)>
>:
transcribe<pop_t<ct_array<l0, lhs...>>, push_t<rhs, l0>, F>
{};
template<auto l0, auto...lhs, class rhs, class F>
struct transcribe<ct_array<l0, lhs...>, rhs, F,
std::enable_if_t<!F{}(l0) && sizeof...(lhs)>
>:
transcribe<pop_t<ct_array<l0, lhs...>>, rhs, F>
{};
template<auto lhs, class rhs, class F>
struct transcribe<ct_array<lhs>, rhs, F, void>
{
using type=std::conditional_t< F{}(lhs), push_t<rhs, lhs>, rhs >;
};
template<class lhs, class F>
using filter_t = transcribe_t<lhs, void, F>;
// C++20
//auto is_even = [](auto i)->bool{ return !(i%2); };
struct is_even_t {
template<class T>
constexpr bool operator()(T i)const{ return !(i%2); }
};
constexpr is_even_t is_even{};
template<auto...is>
static constexpr auto get_evens(ct_array<is...>) {
return filter_t< ct_array<is...>, decltype(is_even) >{};
}
Live example.
Test code:
auto arr = ct_array<11, 22, 33, 44, 55>{};
for (const int i : get_evens(arr)) {
std::cout << i << " " << std::endl;
}

Make longer std::array accessible as if it's shorter

I am implementing my static multi-dimentional vector class. I am using std::array as the underlying data type.
template <typename T, std::size_t N>
class Vector {
private:
std::array<T, N> data;
};
I want to make my class downwards-compatible, so I am writing this:
template <typename T, std::size_t N>
class Vector : public Vector<T, N-1>{
private:
std::array<T, N> data;
};
template <typename T>
class Vector<T, 0> {};
My goal is that when one instance is used in downwards-compatible mode, its underlying data should be able to be reliably accessed:
template<typename T, std::size_t N>
T& Vector<T, N>::operator[](int i) {
// Do boundary checking here
return this->data[i];
}
void foo(Vector<int, 3>& arg) {
arg[1] = 10;
}
Vector<int, 5> b;
foo(b);
// Now b[1] should be 10
There are two points here:
Vector<T, 5> should be accepted by foo(), Vector<T, 2> should be rejected.
Changes to b[0] through b[2] in foo() should pertain. b[3] and b[4] should not be accessible in foo().
How can I achieve that?
How about a simple read wrapper around std::array<> itself?
template<typename T, std::size_t N>
struct ArrayReader {
public:
// Intentionally implicit.
template<std::size_t SRC_LEN>
ArrayReader(std::array<T, SRC_LEN> const& src)
: data_(src.data()) {
static_assert(SRC_LEN >= N);
}
private:
T const* data_;
};
void foo(ArrayReader<float, 3>);
void bar() {
std::array<float, 4> a;
std::array<float, 2> b;
foo(a);
foo(b); //BOOM!
}
Of course, you can easily substitute std::array for your own type, this is just an example of the principle.
Have array keep the data, and then create additional non-owning class, e.g. array_view that will keep only a pointer. It will have generic constructor that accepts the array, and will have a static_assert to check the sizes.
Here's how I would approach this:
template <class Container, std::size_t size>
struct range_view
{
range_view(Container * p): container(p) { assert(size <= p->size()); }
auto & operator[](std::size_t i) { return (*container)[i]; }
private:
Container * container;
};
Then you simply define foo as:
template <class C>
void foo(range_view<C, 3> c)
{
c[1] = 1;
}
Here's something that is closest to what I think you would need.
Make Vector a viewer/user of the data, not the owner of the data.
#include <array>
template <typename T, std::size_t N, std::size_t I>
class Vector : public Vector<T, N, I-1>
{
public:
Vector(std::array<T, N>& arr) : Vector<T, N, I-1>(arr), arr_(arr) {}
T& operator[](int i) {
return arr_[i];
}
private:
std::array<T, N>& arr_;
};
template <typename T, std::size_t N>
class Vector<T, N, 0ul>
{
public:
Vector(std::array<T, N>& arr) : arr_(arr) {}
private:
std::array<T, N>& arr_;
};
void foo(Vector<int, 5, 3>& arg) {
arg[1] = 10;
// Can't find a way to make this a compile time error.
arg[3] = 10;
}
#include <iostream>
int main()
{
std::array<int, 5> arr;
Vector<int, 5, 5> b(arr);
foo(b);
std::cout << b[1] << std::endl;
}
Here's a demonstration of how to implement the Vector class that you tried in your question. At each level you only store 1 value instead of an array and that way when you compose all your N Arrays together you get space for N values. Of course then calling operator[] gets tricky, which is the meat of what I wanted to demonstrate.
#include <utility>
template <class T, std::size_t N>
struct Array : Array<T, N-1>
{
T & operator[](std::size_t i)
{
return const_cast<T&>((*const_cast<const Array*>(this))[i]);
}
const T & operator[](std::size_t i) const
{
return Get(i, std::make_index_sequence<N>());
}
template <std::size_t i>
const T & Geti() const
{
return static_cast<const Array<T, i+1>&>(*this).GetValue();
}
const T & GetValue() const { return value; }
template <std::size_t ... indices>
const T & Get(std::size_t i, std::integer_sequence<std::size_t, indices...>) const
{
using X = decltype(&Array::Geti<0>);
X getters[] = { &Array::Geti<indices>... };
return (this->*getters[i])();
}
template <std::size_t i, class = typename std::enable_if<(i <= N)>::type>
operator Array<T, i>&() { return (Array<T, i>&)*this; }
private:
T value;
};
template <class T>
struct Array<T, 0>{};
void foo(Array<float, 3> & a) { a[1] = 10; }
int main()
{
Array<float, 10> a;
foo(a);
}

Concatenating a sequence of std::arrays

Consider the following: (Wandbox)
#include <array>
#include <algorithm>
#include <iostream>
template<typename T, int N, int M>
auto concat(const std::array<T, N>& ar1, const std::array<T, M>& ar2)
{
std::array<T, N+M> result;
std::copy (ar1.cbegin(), ar1.cend(), result.begin());
std::copy (ar2.cbegin(), ar2.cend(), result.begin() + N);
return result;
}
int main()
{
std::array<int, 3> ar1 = {1, 2, 3};
std::array<int, 2> ar2 = {4, 5};
auto result = concat<int, 3, 2>(ar1, ar2);
for (auto& x : result)
std::cout << x << " ";
std::cout << std::endl;
return 0;
}
Given a sequence of std::array<T, length1>, std::array<T, length2>, ..., std::array<T, lengthK>, how can I generalize the above code and write a function which concatenates the sequence into an std::array<T, sum(lengths)>?
It would be nice if there is a way to write a reusable function which reduces a similar sequence of template classes using a given binary operation, e.g., use concat in the example above, rather than writing a special method (which would have to be re-written each time the binary op changes).
(IIUC, the relevant Standard Library algorithms (accumulate, reduce) only work in case the class of the result of the binary operation is always the same.)
Here is a simple C++17 solution via fold expressions:
#include <array>
#include <algorithm>
template <typename Type, std::size_t... sizes>
auto concatenate(const std::array<Type, sizes>&... arrays)
{
std::array<Type, (sizes + ...)> result;
std::size_t index{};
((std::copy_n(arrays.begin(), sizes, result.begin() + index), index += sizes), ...);
return result;
}
Example of using:
const std::array<int, 3> array1 = {{1, 2, 3}};
const std::array<int, 2> array2 = {{4, 5}};
const std::array<int, 4> array3 = {{6, 7, 8, 9}};
const auto result = concatenate(array1, array2, array3);
Live demo
You may do the following:
template <typename F, typename T, typename T2>
auto func(F f, T&& t, T2&& t2)
{
return f(std::forward<T>(t), std::forward<T2>(t2));
}
template <typename F, typename T, typename T2, typename ... Ts>
auto func(F f, T&& t, T2&& t2, Ts&&...args)
{
return func(f, f(std::forward<T>(t), std::forward<T2>(t2)), std::forward<Ts>(args)...);
}
With usage
struct concatenater
{
template<typename T, std::size_t N, std::size_t M>
auto operator()(const std::array<T, N>& ar1, const std::array<T, M>& ar2) const
{
std::array<T, N+M> result;
std::copy (ar1.cbegin(), ar1.cend(), result.begin());
std::copy (ar2.cbegin(), ar2.cend(), result.begin() + N);
return result;
}
};
and
auto result = func(concatenater{}, ar1, ar2, ar3, ar4);
C++14 Demo
C++11 Demo
Given a sequence of std::array<T, length1>, std::array<T, length2>, ..., std::array<T, lengthK>, how can I write a function which concatenates the sequence into an std::array<T, sum(lengths)>?
Here's a C++17 solution. It can very probably be shortened and improved, working on it.
template <std::size_t Last = 0, typename TF, typename TArray, typename... TRest>
constexpr auto with_acc_sizes(TF&& f, const TArray& array, const TRest&... rest)
{
f(array, std::integral_constant<std::size_t, Last>{});
if constexpr(sizeof...(TRest) != 0)
{
with_acc_sizes<Last + std::tuple_size_v<TArray>>(f, rest...);
}
}
template<typename T, std::size_t... Sizes>
constexpr auto concat(const std::array<T, Sizes>&... arrays)
{
std::array<T, (Sizes + ...)> result{};
with_acc_sizes([&](const auto& arr, auto offset)
{
std::copy(arr.begin(), arr.end(), result.begin() + offset);
}, arrays...);
return result;
}
Usage:
std::array<int, 3> ar1 = {1, 2, 3};
std::array<int, 2> ar2 = {4, 5};
std::array<int, 3> ar3 = {6, 7, 8};
auto result = concat(ar1, ar2, ar3);
live wandbox example
Works with both g++7 and clang++5.
My first line of reasoning would be to consider converting the array to a tuple of references (a tie), manipulate with tuple_cat and then perform whatever operation is necessary to build the final array (i.e. either move or copy - depending on the arguments originally passed in):
#include <array>
#include <iostream>
namespace detail {
template<class Array, std::size_t...Is>
auto array_as_tie(Array &a, std::index_sequence<Is...>) {
return std::tie(a[Is]...);
};
template<class T, class Tuple, std::size_t...Is>
auto copy_to_array(Tuple &t, std::index_sequence<Is...>) {
return std::array<T, sizeof...(Is)>
{
std::get<Is>(t)...
};
};
template<class T, class Tuple, std::size_t...Is>
auto move_to_array(Tuple &t, std::index_sequence<Is...>) {
return std::array<T, sizeof...(Is)>
{
std::move(std::get<Is>(t))...
};
};
}
template<class T, std::size_t N>
auto array_as_tie(std::array<T, N> &a) {
return detail::array_as_tie(a, std::make_index_sequence<N>());
};
// various overloads for different combinations of lvalues and rvalues - needs some work
// for instance, does not handle mixed lvalues and rvalues yet
template<class T, std::size_t N1, std::size_t N2>
auto array_cat(std::array<T, N1> &a1, std::array<T, N2> &a2) {
auto tied = std::tuple_cat(array_as_tie(a1), array_as_tie(a2));
return detail::copy_to_array<T>(tied, std::make_index_sequence<N1 + N2>());
};
template<class T, std::size_t N1, std::size_t N2>
auto array_cat(std::array<T, N1> &&a1, std::array<T, N2> &&a2) {
auto tied = std::tuple_cat(array_as_tie(a1), array_as_tie(a2));
return detail::move_to_array<T>(tied, std::make_index_sequence<N1 + N2>());
};
int main() {
std::array<int, 3> ar1 = {1, 2, 3};
std::array<int, 2> ar2 = {4, 5};
auto result = array_cat(ar1, ar2);
for (auto &x : result)
std::cout << x << " ";
std::cout << std::endl;
// move-construction
auto r2 = array_cat(std::array<std::string, 2> {"a", "b"},
std::array<std::string, 2>{"asdfghjk", "qwertyui"});
std::cout << "string result:\n";
for (auto &&x : r2)
std::cout << x << " ";
std::cout << std::endl;
return 0;
}
expected results:
1 2 3 4 5
string result:
a b asdfghjk qwertyui
A more concise evolution of #Constructor's C++17 solution with the added benefit that Type is not required to be default constructible
template <typename Type, std::size_t... sizes>
constexpr auto concatenate(const std::array<Type, sizes>&... arrays)
{
return std::apply(
[] (auto... elems) -> std::array<Type, (sizes + ...)> { return {{ elems... }}; },
std::tuple_cat(std::tuple_cat(arrays)...));
}
C++14.
template<std::size_t I>
using index_t=std::integral_constant<std::size_t, I>;
template<std::size_t I>
constexpr index_t<I> index{};
template<std::size_t...Is>
auto index_over(std::index_sequence<Is...>){
return [](auto&&f)->decltype(auto){
return decltype(f)(f)( index<Is>... );
};
}
template<std::size_t N>
auto index_upto(index_t<N>={}){
return index_over(std::make_index_sequence<N>{});
}
this lets us expand parameter packs inline.
template<std::size_t, class T>
using indexed_type=T;
template<class T>
std::decay_t<T> concat_arrays( T&& in ){ return std::forward<T>(in); }
template<class T, std::size_t N0, std::size_t N1 >
std::array<T, N0+N1>
concat_arrays( std::array<T,N0> arr0, std::array<T,N1> arr1 ){
auto idx0 = index_upto<N0>();
auto idx1 = index_upto<N1>();
return idx0( [&](auto...I0s){
return idx1( [&](auto...I1s)->std::array<T, N0+N1>{
return {{
arr0[I0s]...,
arr1[I1s]...
}};
})
});
}
which gets us to two. For N, the easy way is:
template<class T, std::size_t N0, std::size_t N1, std::size_t...Ns >
auto concat_arrays( std::array<T,N0> arr0, std::array<T,N1> arr1, std::array<T, Ns>... arrs ){
return concat_arrays( std::move(arr0), concat_arrays( std::move(arr1), std::move(arrs)... ) );
}
but it should be possible without recursion.
Code not tested.
C++17 solution that is constexpr and works correctly with moveably-only types.
template<class Array>
inline constexpr auto array_size = std::tuple_size_v<std::remove_reference_t<Array>>;
template<typename... Ts>
constexpr auto make_array(Ts&&... values)
{
using T = std::common_type_t<Ts...>;
return std::array<T, sizeof...(Ts)>{static_cast<T>(std::forward<Ts>(values))...};
}
namespace detail
{
template<typename Arr1, typename Arr2, std::size_t... is1, std::size_t... is2>
constexpr auto array_cat(Arr1&& arr1, Arr2&& arr2, std::index_sequence<is1...>, std::index_sequence<is2...>)
{
return make_array(std::get<is1>(std::forward<Arr1>(arr1))...,
std::get<is2>(std::forward<Arr2>(arr2))...);
}
}
template<typename Arr, typename... Arrs>
constexpr auto array_cat(Arr&& arr, Arrs&&... arrs)
{
if constexpr (sizeof...(Arrs) == 0)
return std::forward<Arr>(arr);
else if constexpr (sizeof...(Arrs) == 1)
return detail::array_cat(std::forward<Arr>(arr), std::forward<Arrs>(arrs)...,
std::make_index_sequence<array_size<Arr>>{},
std::make_index_sequence<array_size<Arrs...>>{});
else
return array_cat(std::forward<Arr>(arr), array_cat(std::forward<Arrs>(arrs)...));
}
Strictly C++11; not as readable as #Jarod42's, but potentially much more efficient with many arrays if the call-tree isn't fully flattened (in terms of inlining) since only one result object exists rather than multiple temporary, progressively-growing result objects:
namespace detail {
template<std::size_t...>
struct sum_sizes_;
template<std::size_t Acc>
struct sum_sizes_<Acc> : std::integral_constant<std::size_t, Acc> { };
template<std::size_t Acc, std::size_t N, std::size_t... Ns>
struct sum_sizes_<Acc, N, Ns...> : sum_sizes_<Acc + N, Ns...> { };
template<typename... As>
using sum_sizes_t = typename sum_sizes_<
0, std::tuple_size<typename std::decay<As>::type>{}...
>::type;
template<std::size_t O, typename A, typename R>
void transfer(R& ret, typename std::remove_reference<A>::type const& a) {
std::copy(a.begin(), a.end(), ret.begin() + O);
}
template<std::size_t O, typename A, typename R>
void transfer(R& ret, typename std::remove_reference<A>::type&& a) {
std::move(a.begin(), a.end(), ret.begin() + O);
}
template<std::size_t, typename R>
void concat(R const&) { }
template<std::size_t O, typename R, typename A, typename... As>
void concat(R& ret, A&& a, As&&... as) {
transfer<O, A>(ret, std::forward<A>(a));
concat<(O + sum_sizes_t<A>{})>(ret, std::forward<As>(as)...);
}
}
template<typename... As, typename std::enable_if<(sizeof...(As) >= 2), int>::type = 0>
auto concat(As&&... as)
-> std::array<
typename std::common_type<typename std::decay<As>::type::value_type...>::type,
detail::sum_sizes_t<As...>{}
> {
decltype(concat(std::forward<As>(as)...)) ret;
detail::concat<0>(ret, std::forward<As>(as)...);
return ret;
}
Online Demo
Note that this also forwards properly by using the std::move algorithm for rvalues rather than std::copy.
This doesn't generalize, but takes advantage of the fact that if we splat two arrays inside a set of braces, we can use that to initialize a new array.
I'm not sure how useful generalizing is, in any case. Given a bunch of arrays of mismatched sizes, just what else is there to do with them but join them all together?
#include <array>
#include <iostream>
#include <utility>
template<typename T, std::size_t L, std::size_t... Ls,
std::size_t R, std::size_t... Rs>
constexpr std::array<T, L + R> concat_aux(const std::array<T, L>& l, std::index_sequence<Ls...>,
const std::array<T, R>& r, std::index_sequence<Rs...>) {
return std::array<T, L + R> { std::get<Ls>(l)..., std::get<Rs>(r)... };
}
template<typename T, std::size_t L, std::size_t R>
constexpr std::array<T, L + R> concat(const std::array<T, L>& l, const std::array<T, R>& r) {
return concat_aux(l, std::make_index_sequence<L>{},
r, std::make_index_sequence<R>{});
}
template<typename T, std::size_t L, std::size_t R, std::size_t... Sizes>
constexpr auto concat(const std::array<T, L>& l,
const std::array<T, R>& r,
const std::array<T, Sizes>&... arrays) {
return concat(concat(l, r), arrays...);
}
int main() {
std::array<int, 5> a1{1, 2, 3, 4, 5};
std::array<int, 3> a2{6, 7, 8};
std::array<int, 2> a3{9, 10};
for (const auto& elem : concat(a1, a2, a3)) {
std::cout << elem << " ";
}
}