// Copyright 2015, Tobias Hermann and the FunctionalPlus contributors.
// https://github.com/Dobiasd/FunctionalPlus
// Distributed under the Boost Software License, Version 1.0.
// (See accompanying file LICENSE_1_0.txt or copy at
//  http://www.boost.org/LICENSE_1_0.txt)

#pragma once

#include <fplus/container_common.hpp>
#include <fplus/container_properties.hpp>
#include <fplus/generate.hpp>
#include <fplus/pairs.hpp>
#include <fplus/numeric.hpp>
#include <fplus/search.hpp>

#include <fplus/internal/invoke.hpp>
#include <fplus/internal/split.hpp>

namespace fplus
{

// API search type: group_by : (((a, a) -> Bool), [a]) -> [[a]]
// fwd bind count: 1
// Arrange the elements into groups using a given predicate.
// Only groups of consecutive elements are formed.
// For a version scanning the whole container see group_globally_by.
// group_by((==), [1,2,2,2,3,2,2,4,5,5]) == [[1],[2,2,2],[3],[2,2],[4],[5,5]]
// BinaryPredicate p is a (not neccessarily transitive) connectivity check.
// O(n)
template <typename BinaryPredicate, typename ContainerIn,
        typename ContainerOut = typename std::vector<ContainerIn>>
ContainerOut group_by(BinaryPredicate p, const ContainerIn& xs)
{
    // ContainerOut is not deduced to
    // SameContNewType(ContainerIn, ContainerIn)
    // here, since ContainerIn could be a std::string.
    internal::check_binary_predicate_for_container<BinaryPredicate, ContainerIn>();
    static_assert(std::is_same<ContainerIn,
        typename ContainerOut::value_type>::value,
        "Containers do not match.");
    ContainerOut result;
    if (is_empty(xs))
        return result;
    typedef typename ContainerOut::value_type InnerContainerOut;
    *internal::get_back_inserter(result) = InnerContainerOut(1, xs.front());
    for (auto it = ++std::begin(xs); it != std::end(xs); ++it)
    {
        if (internal::invoke(p, result.back().back(), *it))
            *internal::get_back_inserter(result.back()) = *it;
        else
            *internal::get_back_inserter(result) = InnerContainerOut(1, *it);
    }
    return result;
}

// API search type: group_on : ((a -> b), [a]) -> [[a]]
// fwd bind count: 1
// Arrange elements equal after applying a transformer into groups.
// Only groups of consecutive elements are formed.
// For a version scanning the whole container see group_globally_on.
// group_on((mod 10), [12,22,34]) == [[12,22],[34]]
// O(n)
template <typename F, typename ContainerIn>
auto group_on(F f, const ContainerIn& xs)
{
    return group_by(is_equal_by(f), xs);
}

// API search type: group_on_labeled : ((a -> b), [a]) -> [(b, [a])]
// fwd bind count: 1
// Arrange elements equal after applying a transformer into groups,
// adding the transformation result as a label to the group.
// Only groups of consecutive elements are formed.
// For a version scanning the whole container see group_globally_on_labeled.
// group_on_labeled((mod 10), [12,22,34]) == [(2,[12,22]), (4,[34])]
// O(n)
template <typename F, typename ContainerIn>
auto group_on_labeled(F f, const ContainerIn& xs)
{
    const auto group = [](auto f1, const auto& xs1)
    {
        return group_by(f1, xs1);
    };

    return internal::group_on_labeled_impl(group, f, xs);
}

// API search type: group : [a] -> [[a]]
// fwd bind count: 0
// Arrange equal elements into groups.
// Only groups of consecutive elements are formed.
// For a version scanning the whole container see group_globally.
// group([1,2,2,2,3,2,2,4,5,5]) == [[1],[2,2,2],[3],[2,2],[4],[5,5]]
// O(n)
template <typename ContainerIn,
        typename ContainerOut = typename std::vector<ContainerIn>>
ContainerOut group(const ContainerIn& xs)
{
    static_assert(std::is_same<ContainerIn,
        typename ContainerOut::value_type>::value,
        "Containers do not match.");
    typedef typename ContainerIn::value_type T;
    auto pred = [](const T& x, const T& y) { return x == y; };
    return group_by<decltype(pred), ContainerIn, ContainerOut>(pred, xs);
}

// API search type: group_globally_by : (((a, a) -> Bool), [a]) -> [[a]]
// fwd bind count: 1
// Arrange equal elements into groups.
// group_globally_by((==), [1,2,2,2,3,2,2,4,5,5])
// == [[1],[2,2,2,2,2],[3],[4],[5,5]]
// BinaryPredicate p is a
// transitive (whenever p(x,y) and p(y,z), then also p(x,z)) equality check.
// O(n^2)
// If you need O(n*log(n)), sort and then use group_by
template <typename BinaryPredicate, typename ContainerIn,
        typename ContainerOut = typename std::vector<ContainerIn>>
ContainerOut group_globally_by(BinaryPredicate p, const ContainerIn& xs)
{
    internal::check_binary_predicate_for_container<BinaryPredicate, ContainerIn>();
    static_assert(std::is_same<ContainerIn,
        typename ContainerOut::value_type>::value,
        "Containers do not match.");
    typedef typename ContainerOut::value_type InnerContainerOut;
    ContainerOut result;
    for (const auto& x : xs)
    {
        bool found = false;
        for (auto& ys : result)
        {
            if (internal::invoke(p, x, ys.back()))
            {
                *internal::get_back_inserter(ys) = x;
                found = true;
                break;
            }
        }
        if (!found)
        {
            *internal::get_back_inserter(result) = InnerContainerOut(1, x);
        }
    }
    return result;
}

// API search type: group_globally_on : ((a -> b), [a]) -> [[a]]
// fwd bind count: 1
// Arrange elements equal after applying a transformer into groups.
// group_globally_on((mod 10), [12,34,22]) == [[12,22],[34]]
// O(n^2)
// If you need O(n*log(n)), sort and then use group_on
template <typename F, typename ContainerIn>
auto group_globally_on(F f, const ContainerIn& xs)
{
    return group_globally_by(is_equal_by(f), xs);
}

// API search type: group_globally_on_labeled : ((a -> b), [a]) -> [(b, [a])]
// fwd bind count: 1
// Arrange elements equal after applying a transformer into groups,
// adding the transformation result as a label to the group.
// group_globally_on_labeled((mod 10), [12,34,22]) == [(2,[12,22]),(4, [34])]
// O(n^2)
// If you need O(n*log(n)), sort and then use group_on_labeled
template <typename F, typename ContainerIn>
auto group_globally_on_labeled(F f, const ContainerIn& xs)
{
    const auto group = [](auto f1, const auto& xs1)
    {
        return group_globally_by(f1, xs1);
    };

    return internal::group_on_labeled_impl(group, f, xs);
}

// API search type: group_globally : [a] -> [[a]]
// fwd bind count: 0
// Arrange equal elements into groups.
// group_globally([1,2,2,2,3,2,2,4,5,5]) == [[1],[2,2,2,2,2],[3],[4],[5,5]]
// O(n^2)
// If you need O(n*log(n)), sort and then use group
template <typename ContainerIn,
        typename ContainerOut = typename std::vector<ContainerIn>>
ContainerOut group_globally(const ContainerIn& xs)
{
    static_assert(std::is_same<ContainerIn,
        typename ContainerOut::value_type>::value,
        "Containers do not match.");
    typedef typename ContainerIn::value_type T;
    auto pred = [](const T& x, const T& y) { return x == y; };
    return group_globally_by(pred, xs);
}

// API search type: cluster_by : (((a, a) -> Bool), [a]) -> [[a]]
// fwd bind count: 1
// Groups connected components, stable regarding initial order.
// cluster_by(\x y -> abs (y - x) <= 3), [2,3,6,4,12,11,20,23,8,4])
// == [[2,3,6,4,12,11,8,4],[20,23]]
// BinaryPredicate p is a connectivity check, being
//  a) commutative (p(x,y) = p(y,x))
//  b) reflexive (p(x,x) = true)
//  c) not neccessarily transitive, but can be
// O(n^2), memory complexity also O(n^2)
template <typename BinaryPredicate, typename ContainerIn,
        typename ContainerOut = typename std::vector<ContainerIn>>
ContainerOut cluster_by(BinaryPredicate p, const ContainerIn& xs)
{
    internal::check_binary_predicate_for_container<BinaryPredicate, ContainerIn>();
    static_assert(std::is_same<ContainerIn,
        typename ContainerOut::value_type>::value,
        "Containers do not match.");

    typedef std::vector<unsigned char> bools;
    bools zero_filled_row(size_of_cont(xs), 0);

    // adjecency matrix
    typedef std::vector<bools> boolss;
    boolss adj_mat(size_of_cont(xs), zero_filled_row);

    for (const auto& idx_and_val_y : enumerate(xs))
    {
        auto idx_y = idx_and_val_y.first;
        auto val_y = idx_and_val_y.second;
        for (const auto& idx_and_val_x : enumerate(xs))
        {
            auto idx_x = idx_and_val_x.first;
            auto val_x = idx_and_val_x.second;
            if (internal::invoke(p, val_y, val_x))
            {
                adj_mat[idx_y][idx_x] = 1;
            }
        }
    }

    bools already_used = zero_filled_row;
    auto is_already_used = [&](std::size_t i) -> bool
    {
        return already_used[i] != 0;
    };

    typedef std::vector<std::size_t> idxs;
    typedef std::vector<idxs> idxss;

    auto bools_to_idxs = [](const bools& activations) -> idxs
    {
        auto unsigned_char_to_bool = [](unsigned char x)
        {
            return x != 0;
        };
        return find_all_idxs_by(unsigned_char_to_bool, activations);
    };

    idxss idx_clusters;
    std::function<void(std::size_t)> process_idx = [&](std::size_t idx) -> void
    {
        auto connected_idxs = bools_to_idxs(adj_mat[idx]);
        auto new_connected_idxs = drop_if(is_already_used, connected_idxs);
        if (is_empty(new_connected_idxs))
        {
            return;
        }
        idx_clusters.back() = append(idx_clusters.back(), new_connected_idxs);
        for (const auto& new_idx : new_connected_idxs)
        {
            already_used[new_idx] = 1;
        }
        for (const auto& new_idx : new_connected_idxs)
        {
            process_idx(new_idx);
        }
    };

    typedef typename ContainerOut::value_type InnerContainerOut;

    for (const auto& idx : all_idxs(xs))
    {
        if (is_already_used(idx))
        {
            continue;
        }
        *internal::get_back_inserter(idx_clusters) = idxs();
        *internal::get_back_inserter(idx_clusters.back()) = idx;
        already_used[idx] = 1;
        process_idx(idx);
    }

    typedef typename ContainerIn::value_type T;

    auto idx_to_val = [&](std::size_t idx) -> T
    {
        return elem_at_idx(idx, xs);
    };

    auto idxs_to_vals = [&](const idxs& val_idxs) -> InnerContainerOut
    {
        return transform_convert<InnerContainerOut>(idx_to_val, sort(val_idxs));
    };

    return transform_convert<ContainerOut>(idxs_to_vals, idx_clusters);
}

// API search type: split_by : ((a -> Bool), Bool, [a]) -> [[a]]
// fwd bind count: 2
// Split a sequence at every element fulfilling a predicate.
// The splitting elements are discarded.
// split_by(is_even, true, [1,3,2,2,5,5,3,6,7,9]) == [[1,3],[],[5,5,3],[7,9]]
// also known as split_when
// O(n)
template <typename UnaryPredicate, typename ContainerIn,
        typename ContainerOut = typename std::vector<ContainerIn>>
ContainerOut split_by
        (UnaryPredicate pred, bool allow_empty, const ContainerIn& xs)
{
    internal::check_unary_predicate_for_container<UnaryPredicate, ContainerIn>();
    static_assert(std::is_same<ContainerIn,
        typename ContainerOut::value_type>::value,
        "Containers do not match.");

    if (allow_empty && is_empty(xs))
    {
        return {{}};
    }

    ContainerOut result;
    auto itOut = internal::get_back_inserter(result);
    auto start = std::begin(xs);

    while (start != std::end(xs))
    {
        const auto stop = std::find_if(start, std::end(xs), pred);
        if (start != stop || allow_empty)
        {
            *itOut = { start, stop };
        }
        if (stop == std::end(xs))
        {
            break;
        }
        start = internal::add_to_iterator(stop);
        if (allow_empty && start == std::end(xs))
        {
            *itOut = {};
        }
    }
    return result;
}

// API search type: split_by_keep_separators : ((a -> Bool), [a]) -> [[a]]
// fwd bind count: 1
// Split a sequence at every element fulfilling a predicate.
// The splitting elements are kept.
// split_by_keep_separators(is_even, true, [1,3,2,2,5,5,3,6,7,9])
// == [[1,3],[2],[2,5,5,3],[6,7,9]]
// O(n)
template <typename UnaryPredicate, typename ContainerIn,
        typename ContainerOut = typename std::vector<ContainerIn>>
ContainerOut split_by_keep_separators
        (UnaryPredicate pred, const ContainerIn& xs)
{
    internal::check_unary_predicate_for_container<UnaryPredicate, ContainerIn>();
    static_assert(std::is_same<ContainerIn,
        typename ContainerOut::value_type>::value,
        "Containers do not match.");
    ContainerOut result;
    if (is_empty(xs))
        return result;
    auto itOut = internal::get_back_inserter(result);
    auto start = std::begin(xs);
    while (start != std::end(xs))
    {
        const auto stop = std::find_if(
            internal::add_to_iterator(start), std::end(xs), pred);
        *itOut = { start, stop };
        if (stop == std::end(xs))
        {
            break;
        }
        start = stop;
    }
    return result;
}

// API search type: split : (a, Bool, [a]) -> [[a]]
// fwd bind count: 2
// Split a sequence at every element equal to x.
// The splitting elements are discarded.
// split(0, true, [1,3,2,0,0,6,0,7,5]) == [[1,3,2],[],[6],[7,5]]
// O(n)
template <typename ContainerIn,
        typename T = typename ContainerIn::value_type>
auto split(const T& x, bool allow_empty, const ContainerIn& xs)
{
    return split_by(is_equal_to(x), allow_empty, xs);
}

// API search type: split_one_of : ([a], Bool, [a]) -> [[a]]
// fwd bind count: 2
// Split a sequence at every element present in delimiters.
// The splitting elements are discarded.
// Also known as split_words_by_many.
// split_one_of([0,3], true [1,3,2,0,0,6,0,7,5]) == [[1],[2],[],[6],[7,5]]
// split_one_of(" o", false, "How are u?") == ["H","w","are","u?"]
// O(n)
template <typename ContainerIn,
        typename ContainerDelims>
auto split_one_of(
    const ContainerDelims delimiters, bool allow_empty, const ContainerIn& xs)
{
    const auto pred = [&](const typename ContainerIn::value_type& x) -> bool
    {
        return is_elem_of(x, delimiters);
    };
    return split_by(pred, allow_empty, xs);
}

// API search type: split_keep_separators : ((a -> Bool), [a]) -> [[a]]
// fwd bind count: 1
// Split a sequence at every element equal to x.
// The splitting elements are kept.
// split_keep_separators(2, true, [1,3,2,2,5,5,3,2,7,9])
// == [[1,3],[2],[2,5,5,3],[6,7,9]]
// O(n)
template <typename ContainerIn,
        typename T = typename ContainerIn::value_type>
auto split_keep_separators(const T& x, const ContainerIn& xs)
{
    return split_by_keep_separators(is_equal_to(x), xs);
}

// API search type: split_at_idx : (Int, [a]) -> ([a], [a])
// fwd bind count: 1
// Split a sequence at a specific position.
// split_at_idx(2, [0,1,2,3,4]) == ([0,1],[2,3,4])
template <typename Container>
std::pair<Container, Container> split_at_idx
        (std::size_t idx, const Container& xs)
{
    assert(idx <= size_of_cont(xs));
    return make_pair(get_segment(0, idx, xs),
        get_segment(idx, size_of_cont(xs), xs));
}

// API search type: insert_at_idx : (Int, a, [a]) -> [a]
// fwd bind count: 2
// Insert an element into a sequence at a specific position.
// insert_at_idx(2, 0, [1,2,3,4]) == [1,2,0,3,4].
template <typename Container,
        typename T = typename Container::value_type>
Container insert_at_idx(std::size_t idx, const T& x, const Container& xs)
{
    const auto splitted = split_at_idx(idx, xs);
    return concat(std::vector<Container>(
        {
            splitted.first,
            singleton_seq<T, Container>(x),
            splitted.second
        }));
}

// API search type: partition : ((a -> Bool), [a]) -> ([a], [a])
// fwd bind count: 1
// Split a sequence into two groups.
// The first group contains all elements fulfilling the predicate.
// The second group contains the remaining elements.
// partition(is_even, [0,1,1,3,7,2,3,4]) == ([0,2,4],[1,1,3,7,3])
template <typename UnaryPredicate, typename Container>
std::pair<Container, Container> partition
        (UnaryPredicate pred, const Container& xs)
{
    internal::check_unary_predicate_for_container<UnaryPredicate, Container>();
    Container matching;
    Container notMatching;
    auto itOutMatching = internal::get_back_inserter(matching);
    auto itOutNotMatching = internal::get_back_inserter(notMatching);
    for (const auto& x : xs)
    {
        if (internal::invoke(pred, x))
            *itOutMatching = x;
        else
            *itOutNotMatching = x;
    }
    return make_pair(matching, notMatching);
}

// API search type: split_at_idxs : ([Int], [a]) -> [[a]]
// fwd bind count: 1
// Split a sequence at specific indices.
// split_at_idxs([2,5], [0,1,2,3,4,5,6,7]) == [[0,1],[2,3,4],[5,6,7]]
// split_at_idxs([2,5,5], [0,1,2,3,4,5,6,7]) == [[0,1],[2,3,4],[],[5,6,7]]
template <typename ContainerIdxs, typename ContainerIn,
        typename ContainerOut = std::vector<ContainerIn>>
ContainerOut split_at_idxs(const ContainerIdxs& idxsIn, const ContainerIn& xs)
{
    static_assert(std::is_same<typename ContainerIdxs::value_type, std::size_t>::value,
        "Indices must be std::size_t");
    static_assert(std::is_same<ContainerIn,
        typename ContainerOut::value_type>::value,
        "Containers do not match.");
    ContainerIdxs idxStartC = {0};
    ContainerIdxs idxEndC = {size_of_cont(xs)};
    std::vector<ContainerIdxs> containerIdxss = {idxStartC, idxsIn, idxEndC};
    auto idxs = concat(containerIdxss);
    auto idxsClean = sort(idxs);
    ContainerOut result;
    internal::prepare_container(result, size_of_cont(idxsClean) - 1);
    auto itOut = internal::get_back_inserter(result);
    auto idxPairs = overlapping_pairs(idxsClean);
    for (const auto& idxPair : idxPairs)
    {
        *itOut = get_segment(idxPair.first, idxPair.second, xs);
    }
    return result;
}

// API search type: split_every : (Int, [a]) -> [[a]]
// fwd bind count: 1
// Split a sequence every n elements.
// split_every(3, [0,1,2,3,4,5,6,7]) == [[0,1,2],[3,4,5],[6,7]]
// Also known as chunk or chunks.
template <typename ContainerIn,
        typename ContainerOut = std::vector<ContainerIn>>
ContainerOut split_every(std::size_t n, const ContainerIn& xs)
{
    return split_at_idxs<
        std::vector<std::size_t>,
        ContainerIn,
        ContainerOut>(
            numbers_step<std::size_t>(
                n, size_of_cont(xs), n),
            xs);
}

// API search type: split_by_token : ([a], Bool, [a]) -> [[a]]
// fwd bind count: 2
// Split a sequence at every segment matching a token.
// split_by_token(", ", true, "foo, bar, baz") == ["foo", "bar", "baz"]
template <typename ContainerIn,
        typename ContainerOut = typename std::vector<ContainerIn>>
ContainerOut split_by_token(const ContainerIn& token,
        bool allow_empty, const ContainerIn& xs)
{
    static_assert(std::is_same<ContainerIn,
        typename ContainerOut::value_type>::value,
        "Containers do not match.");
    const auto token_begins =
        find_all_instances_of_token_non_overlapping(token, xs);
    const auto token_ends =
        transform(add_to<std::size_t>(size_of_cont(token)), token_begins);
    assert(is_sorted(interweave(token_begins, token_ends)));

    typedef std::vector<std::size_t> idx_vec;
    const auto segments = zip(
        fplus::append(idx_vec(1, 0), token_ends),
        fplus::append(token_begins, idx_vec(1, size_of_cont(xs))));

    ContainerOut result;
    auto itOut = internal::get_back_inserter(result);
    for (const auto& segment : segments)
    {
        if (segment.first != segment.second || allow_empty)
        *itOut = get_segment(segment.first, segment.second, xs);
    }
    return result;
}

// API search type: run_length_encode_by : (((a, a) -> Bool), [a]) -> [(Int, a)]
// fwd bind count: 1
// RLE using a specific binary predicate as equality check.
// run_length_encode_by((==),[1,2,2,2,2,3,3,2)) == [(1,1),(4,2),(2,3),(1,2)]
template <typename BinaryPredicate,
        typename ContainerIn,
        typename T = typename ContainerIn::value_type,
        typename ContainerOut =
            typename std::vector<std::pair<std::size_t, T>>>
ContainerOut run_length_encode_by(BinaryPredicate pred, const ContainerIn& xs)
{
    internal::check_binary_predicate_for_container<BinaryPredicate, ContainerIn>();
    ContainerOut result;
    auto groups = group_by(pred, xs);
    auto group_to_pair = [](const ContainerIn& group) -> std::pair<std::size_t, T>
    {
        return std::make_pair(size_of_cont(group), group.front());
    };
    return transform(group_to_pair, groups);
}

// API search type: run_length_encode : [a] -> [(Int, a)]
// fwd bind count: 0
// RLE.
// run_length_encode([1,2,2,2,2,3,3,2)) == [(1,1),(4,2),(2,3),(1,2)]
template <typename ContainerIn,
        typename T = typename ContainerIn::value_type>
auto run_length_encode(const ContainerIn& xs)
{
    return run_length_encode_by(is_equal<T>, xs);
}

// API search type: run_length_decode : [(Int, a)] -> [a]
// fwd bind count: 0
// Inverse operation to run_length_encode.
// run_length_decode([(1,1),(4,2),(2,3),(1,2)]) == [1,2,2,2,2,3,3,2)
template <typename ContainerIn,
        typename Pair = typename ContainerIn::value_type,
        typename Cnt = typename Pair::first_type>
auto run_length_decode(const ContainerIn& pairs)
{
    static_assert(std::is_convertible<Cnt, std::size_t>::value,
        "Count type must be convertible to std::size_t.");
    const auto pair_to_vec =
        [](const Pair& p)
    {
        return replicate(p.first, p.second);
    };
    return concat(transform(pair_to_vec, pairs));
}

// API search type: span : ((a -> Bool), [a]) -> ([a], [a])
// fwd bind count: 1
// span, applied to a predicate p and a list xs,
// returns a tuple where first element is longest prefix (possibly empty)
// of xs of elements that satisfy p
// and second element is the remainder of the list.
// span(is_even, [0,2,4,5,6,7,8]) == ([0,2,4], [5,6,7,8])
template <typename Container, typename UnaryPredicate>
std::pair<Container, Container> span(UnaryPredicate pred, const Container& xs)
{
    auto maybeIdx = find_first_idx_by(logical_not(pred), xs);
    return {
        take(just_with_default<std::size_t>(size_of_cont(xs), maybeIdx), xs),
        drop(just_with_default<std::size_t>(size_of_cont(xs), maybeIdx), xs)
    };
}

// API search type: divvy : (Int, Int, [a]) -> [[a]]
// fwd bind count: 2
// Generates subsequences overlapping with a specific step.
// divvy(5, 2, [0,1,2,3,4,5,6,7,8,9]) == [[0,1,2,3,4],[2,3,4,5,6],[4,5,6,7,8]]
// divvy(length, 1, xs) is also known as aperture
// divvy(1, step, xs) is also known as stride
//     (but withouts the nested lists in the result)
template <typename ContainerIn,
        typename ContainerOut = std::vector<ContainerIn>>
ContainerOut divvy(std::size_t length, std::size_t step, const ContainerIn& xs)
{
    assert(length > 0);
    assert(step > 0);
    const auto start_idxs =
        numbers_step<std::size_t>(
            0, size_of_cont(xs) - (length - 1), step);

    ContainerOut result;
    internal::prepare_container(result, size_of_cont(start_idxs));
    auto itOut = internal::get_back_inserter(result);

    for (const auto start_idx : start_idxs)
    {
        *itOut = get_segment(start_idx, start_idx + length, xs);
    }
    return result;
}

// API search type: aperture : (Int, [a]) -> [[a]]
// fwd bind count: 1
// Generates overlapping subsequences.
// aperture(5, [0,1,2,3,4,5,6]) == [[0,1,2,3,4],[1,2,3,4,5],[2,3,4,5,6]]
template <typename ContainerIn,
        typename ContainerOut = std::vector<ContainerIn>>
ContainerOut aperture(std::size_t length, const ContainerIn& xs)
{
    assert(length > 0);
    const auto start_idxs =
        numbers<std::size_t>(
            0, size_of_cont(xs) - (length - 1));

    ContainerOut result;
    internal::prepare_container(result, size_of_cont(start_idxs));
    auto itOut = internal::get_back_inserter(result);

    for (const auto start_idx : start_idxs)
    {
        *itOut = get_segment(start_idx, start_idx + length, xs);
    }
    return result;
}

// API search type: stride : (Int, [a]) -> [a]
// fwd bind count: 1
// Keeps every nth element.
// stride(3, [0,1,2,3,4,5,6,7]) == [0,3,6]
template <typename Container>
Container stride(std::size_t step, const Container& xs)
{
    assert(step > 0);
    Container ys;
    auto it = internal::get_back_inserter<Container>(ys);
    auto it_in = std::begin(xs);
    std::size_t i = 0;
    const auto xs_size = size_of_cont(xs);
    while(it_in != std::end(xs))
    {
        *it = *it_in;
        std::size_t increment = std::min(step, xs_size - i);
        internal::advance_iterator(it_in, increment);
        i += increment;
    }
    return ys;
}

// API search type: winsorize : (Float, [Float]) -> [Float]
// fwd bind count: 1
// Winsorizing
// winsorize(0.1, [1,3,4,4,4,4,4,4,6,8]) == [3,3,4,4,4,4,4,4,6,6]
template <typename Container>
Container winsorize(double trim_ratio, const Container& xs)
{
    if (size_of_cont(xs) < 2)
    {
        return xs;
    }
    trim_ratio = std::max(trim_ratio, 0.0);
    const auto xs_sorted = sort(xs);
    std::size_t amount =
        floor<double, std::size_t>(
            trim_ratio * static_cast<double>(size_of_cont(xs_sorted)));
    amount = std::min(size_of_cont(xs_sorted) / 2, amount);
    const auto parts = split_at_idxs(
        std::vector<std::size_t>({amount, size_of_cont(xs_sorted) - amount}),
        xs_sorted);
    assert(size_of_cont(parts) == 3);
    typedef typename Container::value_type T;
    if (is_empty(parts[1]))
    {
        return Container(size_of_cont(xs_sorted), median(xs_sorted));
    }
    else
    {
        const T lower = parts[1].front();
        const T upper = parts[1].back();
        const auto result = concat(std::vector<Container>({
            Container(amount, lower),
            parts[1],
            Container(amount, upper)}));
        assert(size_of_cont(result) == size_of_cont(xs_sorted));
        return result;
    }
}

// API search type: separate_on : ((a -> b), [a]) -> [[a]]
// fwd bind count: 1
// Separate elements equal after applying a transformer into groups.
// separate_on((mod 10), [12,22,34]) == [[12,34],[22]]
template <typename F, typename ContainerIn,
        typename ContainerOut = typename std::vector<ContainerIn>>
ContainerOut separate_on(F f, const ContainerIn& xs)
{
    static_assert(std::is_same<ContainerIn,
        typename ContainerOut::value_type>::value,
        "Containers do not match.");

    ContainerOut result;
    if (is_empty(xs)) {
        return result;
    }

    const auto groups = group_globally_on(f, xs);
    bool found = true;
    auto itOut = internal::get_back_inserter(result);
    std::size_t index = 0;
    while (found) {
        typename ContainerOut::value_type sub_result;
        found = false;
        auto itOutInner = internal::get_back_inserter(sub_result);
        for (auto& group: groups) {
            if (size_of_cont(group) > index)
            {
                *itOutInner = group[index];
                found = true;
            }
        }
        if (found) {
            *itOut = sub_result;
            ++index;
        }
    }
    return result;
}

// API search type: separate : [a] -> [[a]]
// fwd bind count: 0
// Separate equal elements into groups.
// separate([1, 2, 2, 3, 3, 4, 4, 4]) == [[1, 2, 3, 4], [2, 3, 4], [4]]
template <typename ContainerIn,
        typename ContainerOut = typename std::vector<ContainerIn>>
ContainerOut separate(const ContainerIn& xs)
{
    static_assert(std::is_same<ContainerIn,
        typename ContainerOut::value_type>::value,
        "Containers do not match.");
    typedef typename ContainerIn::value_type T;
    return separate_on(identity<T>, xs);
}

} // namespace fplus