Initialize std::array with a range (pair of iterat

问题:

How can I initialize an std::array from a range (as defined by a pair of iterators)?

Something like this:

vector<T> v;
...
// I know v has exactly N elements (e.g. I just called v.resize(N))
// Now I want a initialized with those elements
array<T, N> a(???);  // what to put here?

I thought array would have a constructor taking a pair of iterators, so that I could do array<T, N> a(v.begin(), v.end()), but it appears to have no constructors at all!

I know I can copy the vector into the array, but I'd rather initialize the array with the vector contents directly, without default-constructing it first. How can I?

回答1:

With random access iterators, and assuming a certain size at compile-time, you can use a pack of indices to do so:

template <std::size_t... Indices>
struct indices {
    using next = indices<Indices..., sizeof...(Indices)>;
};
template <std::size_t N>
struct build_indices {
    using type = typename build_indices<N-1>::type::next;
};
template <>
struct build_indices<0> {
    using type = indices<>;
};
template <std::size_t N>
using BuildIndices = typename build_indices<N>::type;

template <typename Iterator>
using ValueType = typename std::iterator_traits<Iterator>::value_type;

// internal overload with indices tag
template <std::size_t... I, typename RandomAccessIterator,
          typename Array = std::array<ValueType<RandomAccessIterator>, sizeof...(I)>>
Array make_array(RandomAccessIterator first, indices<I...>) {
    return Array { { first[I]... } };
}

// externally visible interface
template <std::size_t N, typename RandomAccessIterator>
std::array<ValueType<RandomAccessIterator>, N>
make_array(RandomAccessIterator first, RandomAccessIterator last) {
    // last is not relevant if we're assuming the size is N
    // I'll assert it is correct anyway
    assert(last - first == N); 
    return make_array(first, BuildIndices<N> {});
}

// usage
auto a = make_array<N>(v.begin(), v.end());

This assumes a compiler capable of eliding the intermediate copies. I think that assumption is not a big stretch.

Actually, it can be done with input iterators as well, since the computation of each element in a braced-init-list is sequenced before the computation of the next element (§8.5.4/4).

// internal overload with indices tag
template <std::size_t... I, typename InputIterator,
          typename Array = std::array<ValueType<InputIterator>, sizeof...(I)>>
Array make_array(InputIterator first, indices<I...>) {
    return Array { { (void(I), *first++)... } };
}    

Since *first++ doesn't have any I in it, we need a dummy I to provoke the pack expansion. Comma operator to the rescue, with void() to silence warnings about lack of effects, and also preventing overloaded commas.

回答2:

Like you have noticed, std::array has no constructors at all (except for the compiler generated default constructor).

This was done on purpose, so it can be statically initialized just like a C array. If you want to fill the array without a static initializer, you will have to copy your data.

回答3:

You can use BOOST_PP_ENUM as:

include <boost/preprocessor/repetition/enum.hpp>

#define INIT(z, i, v) v[i] 

std::vector<int> v;

//fill v with at least 5 items 

std::array<int,5> a = { BOOST_PP_ENUM(5, INIT, v) };  //MAGIC

Here, the last line is expanded as:

std::array<int,5> a = {v[0], v[1], v[2], v[3], v[4]}; //EXPANDED LINE

which is what you want.