// generator.hpp
// Copyright (c) 2005-2018 Ben Hanson (http://www.benhanson.net/)
//
// Distributed under the Boost Software License, Version 1.0. (See accompanying
// file licence_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
#ifndef LEXERTL_GENERATOR_HPP
#define LEXERTL_GENERATOR_HPP

#include <algorithm>
#include "partition/charset.hpp"
#include "char_traits.hpp"
#include "partition/equivset.hpp"
#include <list>
#include <memory>
#include "parser/parser.hpp"
#include "rules.hpp"
#include "state_machine.hpp"
#include <type_traits>

namespace lexertl
{
template<typename rules, typename sm, typename char_traits = basic_char_traits
    <typename sm::traits::input_char_type> >
class basic_generator
{
public:
    using id_type = typename rules::id_type;
    using rules_char_type = typename rules::rules_char_type;
    using sm_traits = typename sm::traits;
    using parser = detail::basic_parser<rules_char_type, sm_traits>;
    using charset_map = typename parser::charset_map;
    using node = typename parser::node;
    using node_ptr_vector = typename parser::node_ptr_vector;

    static void build(const rules &rules_, sm &sm_)
    {
        const std::size_t size_ = rules_.statemap().size();
        // Strong exception guarantee
        // http://www.boost.org/community/exception_safety.html
        internals internals_;
        sm temp_sm_;
        node_ptr_vector node_ptr_vector_;

        internals_._eoi = rules_.eoi();
        internals_.add_states(size_);

        for (id_type index_ = 0; index_ < size_; ++index_)
        {
            if (rules_.regexes()[index_].empty())
            {
                std::ostringstream ss_;

                ss_ << "Lexer states with no rules are not allowed "
                    "(lexer state " << index_ << ".)";
                throw runtime_error(ss_.str());
            }
            else
            {
                // Note that the following variables are per DFA.
                // Map of regex charset tokens (strings) to index
                charset_map charset_map_;
                // Used to fix up $ and \n clashes.
                id_type nl_id_ = sm_traits::npos();
                // Regex syntax tree
                observer_ptr<node> root_ = build_tree(rules_, index_,
                    node_ptr_vector_, charset_map_, nl_id_);

                build_dfa(charset_map_, root_, internals_, temp_sm_, index_,
                    nl_id_);

                if (internals_._dfa[index_].size() /
                    internals_._dfa_alphabet[index_] >= sm_traits::npos())
                {
                    // Overflow
                    throw runtime_error("The data type you have chosen "
                        "cannot hold this many DFA rows.");
                }
            }
        }

        // If you get a compile error here the id_type from rules and
        // state machine do no match.
        create(internals_, temp_sm_, rules_.features(), lookup());
        sm_.swap(temp_sm_);
    }

    static observer_ptr<node> build_tree(const rules &rules_,
        const std::size_t dfa_, node_ptr_vector &node_ptr_vector_,
        charset_map &charset_map_, id_type &nl_id_)
    {
        parser parser_(rules_.locale(), node_ptr_vector_, charset_map_,
            rules_.eoi());
        const auto &regexes_ = rules_.regexes();
        auto regex_iter_ = regexes_[dfa_].cbegin();
        auto regex_iter_end_ = regexes_[dfa_].cend();
        const auto &ids_ = rules_.ids();
        const auto &user_ids_ = rules_.user_ids();
        auto id_iter_ = ids_[dfa_].cbegin();
        auto user_id_iter_ = user_ids_[dfa_].cbegin();
        const auto &next_dfas_ = rules_.next_dfas();
        const auto &pushes_ = rules_.pushes();
        const auto &pops_ = rules_.pops();
        auto next_dfa_iter_ = next_dfas_[dfa_].cbegin();
        auto push_dfa_iter_ = pushes_[dfa_].cbegin();
        auto pop_dfa_iter_ = pops_[dfa_].cbegin();
        const bool seen_bol_ = (rules_.features()[dfa_] & bol_bit) != 0;
        observer_ptr<node> root_ = nullptr;

        root_ = parser_.parse(*regex_iter_, *id_iter_, *user_id_iter_,
            *next_dfa_iter_, *push_dfa_iter_, *pop_dfa_iter_,
            rules_.flags(), nl_id_, seen_bol_);
        ++regex_iter_;
        ++id_iter_;
        ++user_id_iter_;
        ++next_dfa_iter_;
        ++push_dfa_iter_;
        ++pop_dfa_iter_;

        // Build syntax trees
        while (regex_iter_ != regex_iter_end_)
        {
            observer_ptr<node> rhs_ = parser_.parse(*regex_iter_, *id_iter_,
                *user_id_iter_, *next_dfa_iter_, *push_dfa_iter_,
                *pop_dfa_iter_, rules_.flags(), nl_id_,
                (rules_.features()[dfa_] & bol_bit) != 0);

            node_ptr_vector_.emplace_back
                (std::make_unique<selection_node>(root_, rhs_));
            root_ = node_ptr_vector_.back().get();

            ++regex_iter_;
            ++id_iter_;
            ++user_id_iter_;
            ++next_dfa_iter_;
            ++push_dfa_iter_;
            ++pop_dfa_iter_;
        }

        return root_;
    }

protected:
    using compressed = std::integral_constant<bool, sm_traits::compressed>;
    using equivset = detail::basic_equivset<id_type>;
    using equivset_list = std::list<std::unique_ptr<equivset>>;
    using equivset_ptr = std::unique_ptr<equivset>;
    using sm_char_type = typename sm_traits::char_type;
    using charset = detail::basic_charset<sm_char_type, id_type>;
    using charset_ptr = std::unique_ptr<charset>;
    using charset_list = std::list<std::unique_ptr<charset>>;
    using internals = detail::basic_internals<id_type>;
    using id_type_set = typename std::set<id_type>;
    using id_type_vector = typename internals::id_type_vector;
    using index_set = typename charset::index_set;
    using index_set_vector = std::vector<index_set>;
    using is_dfa = std::integral_constant<bool, sm_traits::is_dfa>;
    using lookup = std::integral_constant<bool, sm_traits::lookup>;
    using node_set = std::set<observer_ptr<const node>>;
    using node_set_vector = std::vector<std::unique_ptr<node_set>>;
    using node_vector = typename node::node_vector;
    using node_vector_vector = std::vector<std::unique_ptr<node_vector>>;
    using selection_node = typename parser::selection_node;
    using size_t_vector = typename std::vector<std::size_t>;
    using string_token = typename parser::string_token;

    static void build_dfa(const charset_map &charset_map_,
        const observer_ptr<node> root_, internals &internals_, sm &sm_,
        const id_type dfa_index_, id_type &nl_id_)
    {
        // partitioned charset list
        charset_list charset_list_;
        // vector mapping token indexes to partitioned token index sets
        index_set_vector set_mapping_;
        auto &dfa_ = internals_._dfa[dfa_index_];
        std::size_t dfa_alphabet_ = 0;
        const node_vector &followpos_ = root_->firstpos();
        node_set_vector seen_sets_;
        node_vector_vector seen_vectors_;
        size_t_vector hash_vector_;
        id_type zero_id_ = sm_traits::npos();
        id_type_set eol_set_;

        set_mapping_.resize(charset_map_.size());
        partition_charsets(charset_map_, charset_list_, is_dfa());
        build_set_mapping(charset_list_, internals_, dfa_index_,
            set_mapping_);

        if (nl_id_ != sm_traits::npos())
        {
            nl_id_ = *set_mapping_[nl_id_].begin();
            zero_id_ = sm_traits::compressed ?
                *set_mapping_[charset_map_.find(string_token(0, 0))->
                second].begin() : sm_traits::npos();
        }

        dfa_alphabet_ = charset_list_.size() + transitions_index +
            (nl_id_ == sm_traits::npos() ? 0 : 1);

        if (dfa_alphabet_ > sm_traits::npos())
        {
            // Overflow
            throw runtime_error("The data type you have chosen cannot hold "
                "the dfa alphabet.");
        }

        internals_._dfa_alphabet[dfa_index_] =
            static_cast<id_type>(dfa_alphabet_);
        // 'jam' state
        dfa_.resize(dfa_alphabet_, 0);
        closure(followpos_, seen_sets_, seen_vectors_, hash_vector_,
            static_cast<id_type>(dfa_alphabet_), dfa_);

        // Loop over states
        for (id_type index_ = 0; index_ < static_cast<id_type>
            (seen_vectors_.size()); ++index_)
        {
            equivset_list equiv_list_;

            // Intersect charsets
            build_equiv_list(*seen_vectors_[index_].get(), set_mapping_,
                equiv_list_, is_dfa());

            for (auto &equivset_ : equiv_list_)
            {
                const id_type transition_ = closure
                    (equivset_->_followpos, seen_sets_, seen_vectors_,
                    hash_vector_, static_cast<id_type>(dfa_alphabet_), dfa_);

                if (transition_ != sm_traits::npos())
                {
                    observer_ptr<id_type> ptr_ = &dfa_.front() +
                        ((index_ + 1) * dfa_alphabet_);

                    // Prune abstemious transitions from end states.
                    if (*ptr_ && !equivset_->_greedy) continue;

                    set_transitions(transition_, equivset_.get(), dfa_, ptr_,
                        index_, eol_set_);
                }
            }
        }

        fix_clashes(eol_set_, nl_id_, zero_id_, dfa_, dfa_alphabet_,
            compressed());
        append_dfa(charset_list_, internals_, sm_, dfa_index_, lookup());
    }

    static void set_transitions(const id_type transition_, equivset *equivset_,
        typename internals::id_type_vector &dfa_, id_type *ptr_,
        const id_type index_, id_type_set &eol_set_)
    {
        for (typename equivset::index_vector::const_iterator
            equiv_iter_ = equivset_->_index_vector.begin(),
            equiv_end_ = equivset_->_index_vector.end();
            equiv_iter_ != equiv_end_; ++equiv_iter_)
        {
            const id_type i_ = *equiv_iter_;

            if (i_ == parser::bol_token())
            {
                dfa_.front() = transition_;
            }
            else if (i_ == parser::eol_token())
            {
                ptr_[eol_index] = transition_;
                eol_set_.insert(index_ + 1);
            }
            else
            {
                ptr_[i_ + transitions_index] = transition_;
            }
        }
    }

    // Uncompressed
    static void fix_clashes(const id_type_set &eol_set_,
        const id_type nl_id_, const id_type /*zero_id_*/,
        typename internals::id_type_vector &dfa_,
        const std::size_t dfa_alphabet_, const std::false_type &)
    {
        for (const auto &eol_ : eol_set_)
        {
            observer_ptr<id_type> ptr_ = &dfa_.front() + eol_ * dfa_alphabet_;
            const id_type eol_state_ = ptr_[eol_index];
            const id_type nl_state_ = ptr_[nl_id_ + transitions_index];

            if (nl_state_)
            {
                ptr_[transitions_index + nl_id_] = 0;
                ptr_ = &dfa_.front() + eol_state_ * dfa_alphabet_;

                if (ptr_[transitions_index + nl_id_] == 0)
                {
                    ptr_[transitions_index + nl_id_] = nl_state_;
                }
            }
        }
    }

    // Compressed
    static void fix_clashes(const id_type_set &eol_set_,
        const id_type nl_id_, const id_type zero_id_,
        typename internals::id_type_vector &dfa_,
        const std::size_t dfa_alphabet_, const std::true_type &)
    {
        std::size_t i_ = 0;

        for (const auto &eol_ : eol_set_)
        {
            observer_ptr<id_type> ptr_ = &dfa_.front() + eol_ * dfa_alphabet_;
            const id_type eol_state_ = ptr_[eol_index];
            id_type nl_state_ = 0;

            for (; i_ < (sm_traits::char_24_bit ? 2 : 1); ++i_)
            {
                ptr_ = &dfa_.front() + ptr_[transitions_index + zero_id_] *
                    dfa_alphabet_;
            }

            nl_state_ = ptr_[transitions_index + nl_id_];

            if (nl_state_)
            {
                ptr_ = &dfa_.front() + eol_state_ * dfa_alphabet_;

                if (ptr_[transitions_index + zero_id_] != 0) continue;

                ptr_[transitions_index + zero_id_] =
                    static_cast<id_type>(dfa_.size() / dfa_alphabet_);
                dfa_.resize(dfa_.size() + dfa_alphabet_, 0);

                for (i_ = 0; i_ < (sm_traits::char_24_bit ? 1 : 0); ++i_)
                {
                    ptr_ = &dfa_.front() + dfa_.size() - dfa_alphabet_;
                    ptr_[transitions_index + zero_id_] =
                        static_cast<id_type>(dfa_.size() / dfa_alphabet_);
                    dfa_.resize(dfa_.size() + dfa_alphabet_, 0);
                }

                ptr_ = &dfa_.front() + dfa_.size() - dfa_alphabet_;
                ptr_[transitions_index + nl_id_] = nl_state_;
            }
        }
    }

    // char_state_machine version
    static void append_dfa(const charset_list &charset_list_,
        const internals &internals_, sm &sm_, const id_type dfa_index_,
        const std::false_type &)
    {
        std::size_t size_ = charset_list_.size();
        typename sm::string_token_vector token_vector_;

        token_vector_.reserve(size_);

        for (const auto &charset_ : charset_list_)
        {
            token_vector_.push_back(charset_->_token);
        }

        sm_.append(token_vector_, internals_, dfa_index_);
    }

    // state_machine version
    static void append_dfa(const charset_list &, const internals &, sm &,
        const id_type, const std::true_type &)
    {
        // Nothing to do - will use create() instead
    }

    // char_state_machine version
    static void create(internals &, sm &, const id_type_vector &,
        const std::false_type &)
    {
        // Nothing to do - will use append_dfa() instead
    }

    // state_machine version
    static void create(internals &internals_, sm &sm_,
        const id_type_vector &features_, const std::true_type &)
    {
        for (std::size_t i_ = 0, size_ = internals_._dfa.size();
            i_ < size_; ++i_)
        {
            internals_._features |= features_[i_];
        }

        if (internals_._dfa.size() > 1)
        {
            internals_._features |= multi_state_bit;
        }

        sm_.data().swap(internals_);
    }

    // NFA version
    static void partition_charsets(const charset_map &map_,
        charset_list &lhs_, const std::false_type &)
    {
        fill_rhs_list(map_, lhs_);
    }

    // DFA version
    static void partition_charsets(const charset_map &map_,
        charset_list &lhs_, const std::true_type &)
    {
        charset_list rhs_;

        fill_rhs_list(map_, rhs_);

        if (!rhs_.empty())
        {
            typename charset_list::iterator iter_;
            typename charset_list::iterator end_;
            charset_ptr overlap_ = std::make_unique<charset>();

            lhs_.emplace_back(std::move(rhs_.front()));
            rhs_.pop_front();

            while (!rhs_.empty())
            {
                charset_ptr r_(rhs_.front().release());

                rhs_.pop_front();
                iter_ = lhs_.begin();
                end_ = lhs_.end();

                while (!r_->empty() && iter_ != end_)
                {
                    auto l_iter_ = iter_;

                    (*l_iter_)->intersect(*r_.get(), *overlap_.get());

                    if (overlap_->empty())
                    {
                        ++iter_;
                    }
                    else if ((*l_iter_)->empty())
                    {
                        l_iter_->reset(overlap_.release());
                        overlap_ = std::make_unique<charset>();
                        ++iter_;
                    }
                    else if (r_->empty())
                    {
                        r_.reset(overlap_.release());
                        overlap_ = std::make_unique<charset>();
                        break;
                    }
                    else
                    {
                        iter_ = lhs_.insert(++iter_, charset_ptr());
                        iter_->reset(overlap_.release());
                        overlap_ = std::make_unique<charset>();
                        ++iter_;
                        end_ = lhs_.end();
                    }
                }

                if (!r_->empty())
                {
                    lhs_.emplace_back(std::move(r_));
                }
            }
        }
    }

    static void fill_rhs_list(const charset_map &map_, charset_list &list_)
    {
        for (const auto &pair_ : map_)
        {
            list_.emplace_back(std::make_unique<charset>
                (pair_.first, pair_.second));
        }
    }

    static void build_set_mapping(const charset_list &charset_list_,
        internals &internals_, const id_type dfa_index_,
        index_set_vector &set_mapping_)
    {
        auto iter_ = charset_list_.cbegin();
        auto end_ = charset_list_.cend();

        for (id_type index_ = 0; iter_ != end_; ++iter_, ++index_)
        {
            observer_ptr<const charset> cs_ = iter_->get();

            fill_lookup(cs_->_token, &internals_._lookup[dfa_index_],
                index_, lookup());

            for (const id_type i_ : cs_->_index_set)
            {
                set_mapping_[i_].insert(index_);
            }
        }
    }

    // char_state_machine version
    static void fill_lookup(const string_token &, observer_ptr<id_type_vector> ,
        const id_type, const std::false_type &)
    {
        // Do nothing (lookup not used)
    }

    // state_machine version
    static void fill_lookup(const string_token &charset_,
        observer_ptr<id_type_vector> lookup_, const id_type index_,
        const std::true_type &)
    {
        observer_ptr<id_type> ptr_ = &lookup_->front();

        for (const auto &range_ : charset_._ranges)
        {
            for (typename char_traits::index_type char_ = range_.first;
                char_ < range_.second; ++char_)
            {
                // Note char_ must be unsigned
                ptr_[char_] = index_ + transitions_index;
            }

            // Note range_.second must be unsigned
            ptr_[range_.second] = index_ + transitions_index;
        }
    }

    static id_type closure(const node_vector &followpos_,
        node_set_vector &seen_sets_, node_vector_vector &seen_vectors_,
        size_t_vector &hash_vector_, const id_type size_, id_type_vector &dfa_)
    {
        bool end_state_ = false;
        id_type id_ = 0;
        id_type user_id_ = sm_traits::npos();
        id_type next_dfa_ = 0;
        id_type push_dfa_ = sm_traits::npos();
        bool pop_dfa_ = false;
        std::size_t hash_ = 0;

        if (followpos_.empty()) return sm_traits::npos();

        id_type index_ = 0;
        std::unique_ptr<node_set> set_ptr_ = std::make_unique<node_set>();
        std::unique_ptr<node_vector> vector_ptr_ =
            std::make_unique<node_vector>();

        for (observer_ptr<node> node_ : followpos_)
        {
            closure_ex(node_, end_state_, id_, user_id_, next_dfa_,
                push_dfa_, pop_dfa_, *set_ptr_.get(),
                *vector_ptr_.get(), hash_);
        }

        bool found_ = false;
        auto hash_iter_ = hash_vector_.cbegin();
        auto hash_end_ = hash_vector_.cend();
        auto set_iter_ = seen_sets_.cbegin();

        for (; hash_iter_ != hash_end_; ++hash_iter_, ++set_iter_)
        {
            found_ = *hash_iter_ == hash_ && *(*set_iter_) == *set_ptr_;
            ++index_;

            if (found_) break;
        }

        if (!found_)
        {
            seen_sets_.emplace_back(std::move(set_ptr_));
            seen_vectors_.emplace_back(std::move(vector_ptr_));
            hash_vector_.push_back(hash_);
            // State 0 is the jam state...
            index_ = static_cast<id_type>(seen_sets_.size());

            const std::size_t old_size_ = dfa_.size();

            dfa_.resize(old_size_ + size_, 0);

            if (end_state_)
            {
                dfa_[old_size_] |= end_state_bit;

                if (pop_dfa_)
                {
                    dfa_[old_size_] |= pop_dfa_bit;
                }

                dfa_[old_size_ + id_index] = id_;
                dfa_[old_size_ + user_id_index] = user_id_;
                dfa_[old_size_ + push_dfa_index] = push_dfa_;
                dfa_[old_size_ + next_dfa_index] = next_dfa_;
            }
        }

        return index_;
    }

    static void closure_ex(observer_ptr<node> node_, bool &end_state_,
        id_type &id_, id_type &user_id_, id_type &next_dfa_,
        id_type &push_dfa_, bool &pop_dfa_, node_set &set_ptr_,
        node_vector &vector_ptr_, std::size_t &hash_)
    {
        const bool temp_end_state_ = node_->end_state();

        if (temp_end_state_)
        {
            if (!end_state_)
            {
                end_state_ = true;
                id_ = node_->id();
                user_id_ = node_->user_id();
                next_dfa_ = node_->next_dfa();
                push_dfa_ = node_->push_dfa();
                pop_dfa_ = node_->pop_dfa();
            }
        }

        if (set_ptr_.insert(node_).second)
        {
            vector_ptr_.push_back(node_);
            hash_ += reinterpret_cast<std::size_t>(node_);
        }
    }

    // NFA version
    static void build_equiv_list(const node_vector &vector_,
        const index_set_vector &set_mapping_, equivset_list &lhs_,
        const std::false_type &)
    {
        fill_rhs_list(vector_, set_mapping_, lhs_);
    }

    // DFA version
    static void build_equiv_list(const node_vector &vector_,
        const index_set_vector &set_mapping_, equivset_list &lhs_,
        const std::true_type &)
    {
        equivset_list rhs_;

        fill_rhs_list(vector_, set_mapping_, rhs_);

        if (!rhs_.empty())
        {
            typename equivset_list::iterator iter_;
            typename equivset_list::iterator end_;
            equivset_ptr overlap_ = std::make_unique<equivset>();

            lhs_.emplace_back(std::move(rhs_.front()));
            rhs_.pop_front();

            while (!rhs_.empty())
            {
                equivset_ptr r_(rhs_.front().release());

                rhs_.pop_front();
                iter_ = lhs_.begin();
                end_ = lhs_.end();

                while (!r_->empty() && iter_ != end_)
                {
                    auto l_iter_ = iter_;

                    (*l_iter_)->intersect(*r_.get(), *overlap_.get());

                    if (overlap_->empty())
                    {
                        ++iter_;
                    }
                    else if ((*l_iter_)->empty())
                    {
                        l_iter_->reset(overlap_.release());
                        overlap_ = std::make_unique<equivset>();
                        ++iter_;
                    }
                    else if (r_->empty())
                    {
                        r_.reset(overlap_.release());
                        overlap_ = std::make_unique<equivset>();
                        break;
                    }
                    else
                    {
                        iter_ = lhs_.insert(++iter_, equivset_ptr());
                        iter_->reset(overlap_.release());
                        overlap_ = std::make_unique<equivset>();
                        ++iter_;
                        end_ = lhs_.end();
                    }
                }

                if (!r_->empty())
                {
                    lhs_.emplace_back(std::move(r_));
                }
            }
        }
    }

    static void fill_rhs_list(const node_vector &vector_,
        const index_set_vector &set_mapping_, equivset_list &list_)
    {
        for (observer_ptr<const node> node_ : vector_)
        {
            if (!node_->end_state())
            {
                const id_type token_ = node_->token();

                if (token_ != node::null_token())
                {
                    if (token_ == parser::bol_token() ||
                        token_ == parser::eol_token())
                    {
                        std::set<id_type> index_set_;

                        index_set_.insert(token_);
                        list_.emplace_back
                            (std::make_unique<equivset>(index_set_,
                                token_, node_->greedy(), node_->followpos()));
                    }
                    else
                    {
                        list_.emplace_back(std::make_unique<equivset>
                            (set_mapping_[token_], token_, node_->greedy(),
                                node_->followpos()));
                    }
                }
            }
        }
    }
};

using generator = basic_generator<rules, state_machine>;
using wgenerator = basic_generator<wrules, wstate_machine>;
using u32generator = basic_generator<u32rules, u32state_machine>;
using char_generator = basic_generator<rules, char_state_machine>;
using wchar_generator = basic_generator<wrules, wchar_state_machine>;
using u32char_generator = basic_generator<u32rules, u32char_state_machine>;
}

#endif