#include "generator.h"

#include <hkutil/progress.h>
#include <hkutil/string.h>

#include <algorithm>
#include <filesystem>
#include <fstream>
#include <list>
#include <regex>
#include <set>
#include <stdexcept>
#include <string>
#include <unordered_map>
#include <unordered_set>
#include <vector>

constexpr int MIN_FREQUENCY = 2000000;

namespace {

std::list<std::string> readFile(std::string path, bool uniq = false) {
  std::ifstream file(path);
  if (!file) {
    throw std::invalid_argument("Could not find file " + path);
  }

  std::list<std::string> lines;
  std::string line;
  while (std::getline(file, line)) {
    if (line.back() == '\r') {
      line.pop_back();
    }

    lines.push_back(line);
  }

  if (uniq) {
    std::vector<std::string> uniq(std::begin(lines), std::end(lines));
    lines.clear();

    std::sort(std::begin(uniq), std::end(uniq));
    std::unique_copy(std::begin(uniq), std::end(uniq),
                     std::back_inserter(lines));
  }

  return lines;
}

}  // namespace

generator::generator(std::string agidPath, std::string wordNetPath,
                     std::string cmudictPath, std::string wordfreqPath,
                     std::string datadirPath, std::string outputPath)
    : agidPath_(agidPath),
      wordNetPath_(wordNetPath),
      cmudictPath_(cmudictPath),
      wordfreqPath_(wordfreqPath),
      datadirPath_(datadirPath),
      outputPath_(outputPath) {
  // Ensure AGID infl.txt exists
  if (!std::ifstream(agidPath_)) {
    throw std::invalid_argument("AGID infl.txt file not found");
  }

  // Add directory separator to WordNet path
  if ((wordNetPath_.back() != '/') && (wordNetPath_.back() != '\\')) {
    wordNetPath_ += '/';
  }

  // Ensure WordNet tables exist
  for (std::string table : {"s", "sk", "ant", "at", "cls", "hyp", "ins", "mm",
                            "mp", "ms", "per", "sa", "sim", "syntax"}) {
    if (!std::ifstream(wordNetPath_ + "wn_" + table + ".pl")) {
      throw std::invalid_argument("WordNet " + table + " table not found");
    }
  }

  // Ensure CMUDICT file exists
  if (!std::ifstream(cmudictPath_)) {
    throw std::invalid_argument("CMUDICT file not found");
  }
}

void generator::run() {
  std::unordered_map<std::string, int> word_frequencies;
  {
    std::list<std::string> lines(readFile(wordfreqPath_));

    hatkirby::progress ppgs("Reading word frequencies...", lines.size());

    for (std::string line : lines) {
      ppgs.update();

      std::regex freqline("([a-z]+),([0-9]+)");
      std::smatch freqline_data;
      if (std::regex_search(line, freqline_data, freqline)) {
        std::string text = freqline_data[1];
        std::string freqnumstr = freqline_data[2];
        long long freqnumnum = std::atoll(freqnumstr.c_str());
        word_frequencies[text] = freqnumnum > std::numeric_limits<int>::max()
                                     ? std::numeric_limits<int>::max()
                                     : freqnumnum;
      }
    }
  }

  std::unordered_set<std::string> profane;
  {
    std::list<std::string> lines(readFile(datadirPath_ / "profane.txt"));
    for (const std::string& line : lines) {
      profane.insert(line);
    }
  }

  {
    std::list<std::string> lines(readFile(wordNetPath_ + "wn_s.pl"));
    hatkirby::progress ppgs("Reading synsets from WordNet...", lines.size());

    for (std::string line : lines) {
      ppgs.update();

      std::regex relation(
          "^s\\(([1234]\\d{8}),(\\d+),'(.+)',\\w,\\d+,(\\d+)\\)\\.$");

      std::smatch relation_data;
      if (!std::regex_search(line, relation_data, relation)) {
        continue;
      }

      int synset_id = std::stoi(relation_data[1]);
      int wnum = std::stoi(relation_data[2]);
      std::string text = relation_data[3];
      int tag_count = std::stoi(relation_data[4]);
      size_t word_it;
      while ((word_it = text.find("''")) != std::string::npos) {
        text.erase(word_it, 1);
      }

      // The word must be common enough.
      if (word_frequencies[text] < MIN_FREQUENCY) {
        continue;
      }

      // We are looking for single words.
      if (std::count(std::begin(text), std::end(text), ' ') > 0) {
        continue;
      }

      // This should filter our proper nouns.
      if (std::any_of(std::begin(text), std::end(text), ::isupper)) {
        continue;
      }

      // Ignore any profane words.
      if (profane.count(text)) {
        continue;
      }

      // The WordNet data does contain duplicates, so we need to check that we
      // haven't already created this word.
      std::pair<int, int> lookup(synset_id, wnum);
      if (word_by_wnid_and_wnum_.count(lookup)) {
        continue;
      }

      size_t word_id = LookupOrCreateWord(text);
      word_by_wnid_and_wnum_[lookup] = word_id;
      AddWordToSynset(word_id, synset_id);
    }
  }

  {
    std::list<std::string> lines(readFile(agidPath_));
    hatkirby::progress ppgs("Reading inflections from AGID...", lines.size());

    for (std::string line : lines) {
      ppgs.update();

      int divider = line.find_first_of(" ");
      std::string infinitive = line.substr(0, divider);
      line = line.substr(divider + 1);
      char type = line[0];

      if (line[1] == '?') {
        line.erase(0, 4);
      } else {
        line.erase(0, 3);
      }

      if (!words_by_base_.count(infinitive)) {
        continue;
      }

      auto inflWordList = hatkirby::split<std::list<std::string>>(line, " | ");

      std::vector<std::list<std::string>> agidForms;
      for (std::string inflForms : inflWordList) {
        auto inflFormList =
            hatkirby::split<std::list<std::string>>(std::move(inflForms), ", ");

        std::list<std::string> forms;
        for (std::string inflForm : inflFormList) {
          int sympos = inflForm.find_first_of("~<!? ");
          if (sympos != std::string::npos) {
            inflForm = inflForm.substr(0, sympos);
          }

          forms.push_back(std::move(inflForm));
        }

        agidForms.push_back(std::move(forms));
      }

      std::vector<std::list<std::string>> inflections;
      switch (type) {
        case 'V': {
          if (agidForms.size() == 4) {
            inflections.push_back(agidForms[0]);
            inflections.push_back(agidForms[1]);
            inflections.push_back(agidForms[2]);
            inflections.push_back(agidForms[3]);
          } else if (agidForms.size() == 3) {
            inflections.push_back(agidForms[0]);
            inflections.push_back(agidForms[1]);
            inflections.push_back(agidForms[2]);
          } else if (agidForms.size() == 8) {
            // As of AGID 2014.08.11, this is only "to be"
            inflections.push_back(agidForms[0]);
            inflections.push_back(agidForms[2]);
            inflections.push_back(agidForms[3]);
            inflections.push_back(agidForms[4]);
          } else {
            // Words that don't fit the cases above as of AGID 2014.08.11:
            // - may and shall do not conjugate the way we want them to
            // - methinks only has a past tense and is an outlier
            // - wit has five forms, and is archaic/obscure enough that we can
            // ignore it for now
            std::cout << " Ignoring verb \"" << infinitive
                      << "\" due to non-standard number of forms." << std::endl;
          }

          break;
        }

        case 'A': {
          if (agidForms.size() == 2) {
            inflections.push_back(agidForms[0]);
            inflections.push_back(agidForms[1]);
          } else {
            // As of AGID 2014.08.11, this is only "only", which has only the
            // form "onliest"
            std::cout << " Ignoring adjective/adverb \"" << infinitive
                      << "\" due to non-standard number of forms." << std::endl;
          }

          break;
        }

        case 'N': {
          if (agidForms.size() == 1) {
            inflections.push_back(agidForms[0]);
          } else {
            // As of AGID 2014.08.11, this is non-existent.
            std::cout << " Ignoring noun \"" << infinitive
                      << "\" due to non-standard number of forms." << std::endl;
          }

          break;
        }
      }

      // Compile the forms we have mapped.
      for (size_t word_id : words_by_base_.at(infinitive)) {
        for (const std::list<std::string>& infl_list : inflections) {
          for (const std::string& infl : infl_list) {
            if (!profane.count(infl)) {
              size_t form_id = LookupOrCreateForm(infl);
              AddFormToWord(form_id, word_id);
            }
          }
        }
      }
    }
  }

  word_frequencies.clear();  // Not needed anymore.

  {
    std::list<std::string> lines(readFile(cmudictPath_));

    hatkirby::progress ppgs("Reading pronunciations from CMUDICT...",
                            lines.size());

    for (std::string line : lines) {
      ppgs.update();

      std::regex phoneme("([A-Z][^ \\(]*)(?:\\(\\d+\\))?  ([A-Z 0-9]+)");
      std::smatch phoneme_data;
      if (std::regex_search(line, phoneme_data, phoneme)) {
        std::string canonical = hatkirby::lowercase(phoneme_data[1]);

        if (!form_by_text_.count(canonical)) {
          continue;
        }

        std::string phonemes = phoneme_data[2];
        size_t pronunciation_id = LookupOrCreatePronunciation(phonemes);
        AddPronunciationToForm(pronunciation_id, form_by_text_[canonical]);
      }
    }
  }

  std::cout << "Words: " << words_.size() << std::endl;
  std::cout << "Forms: " << forms_.size() << std::endl;
  std::cout << "Pronunciations: " << pronunciations_.size() << std::endl;

  // White Top
  {
    hatkirby::progress ppgs("Generating white top puzzles...", forms_.size());

    for (Form& form : forms_) {
      ppgs.update();

      for (size_t p_id : form.pronunciation_ids) {
        const Pronunciation& pronunciation = pronunciations_.at(p_id);
        for (size_t other_form_id : pronunciation.form_ids) {
          if (other_form_id != form.id) {
            form.puzzles[kWhiteTop].insert(other_form_id);
          }
        }
      }
    }
  }

  // White Bottom
  {
    hatkirby::progress ppgs("Generating white bottom puzzles...",
                            words_.size());

    for (const Word& word : words_) {
      ppgs.update();

      Form& form = forms_.at(word.base_form_id);
      for (size_t synset_id : word.synsets) {
        for (size_t other_word_id : synsets_.at(synset_id)) {
          if (other_word_id != word.id) {
            const Word& other_word = words_.at(other_word_id);
            form.puzzles[kWhiteBottom].insert(other_word.base_form_id);
          }
        }
      }
    }
  }

  // Yellow Top
  {
    hatkirby::progress ppgs("Generating yellow top puzzles...",
                            anaphone_sets_.size());

    for (const std::vector<size_t>& anaphone_set : anaphone_sets_) {
      ppgs.update();

      std::set<size_t> all_forms;
      for (size_t p_id : anaphone_set) {
        const Pronunciation& pronunciation = pronunciations_.at(p_id);
        for (size_t form_id : pronunciation.form_ids) {
          all_forms.insert(form_id);
        }
      }

      for (size_t f_id1 : all_forms) {
        for (size_t f_id2 : all_forms) {
          if (f_id1 != f_id2) {
            Form& form = forms_.at(f_id1);
            form.puzzles[kYellowTop].insert(f_id2);
          }
        }
      }
    }
  }

  // Yellow Middle
  {
    hatkirby::progress ppgs("Generating yellow middle puzzles...",
                            anagram_sets_.size());

    for (const std::vector<size_t>& anagram_set : anagram_sets_) {
      ppgs.update();

      for (size_t f_id1 : anagram_set) {
        for (size_t f_id2 : anagram_set) {
          if (f_id1 != f_id2) {
            Form& form = forms_.at(f_id1);
            form.puzzles[kYellowMiddle].insert(f_id2);
          }
        }
      }
    }
  }

  // Black Top
  {
    hatkirby::progress ppgs("Generating black top puzzles...",
                            pronunciations_.size());

    for (const Pronunciation& pronunciation : pronunciations_) {
      ppgs.update();

      auto reversed_list = hatkirby::split<std::vector<std::string>>(
          pronunciation.stressless_phonemes, " ");
      std::reverse(reversed_list.begin(), reversed_list.end());
      std::string reversed_phonemes =
          hatkirby::implode(reversed_list.begin(), reversed_list.end(), " ");
      if (pronunciations_by_blank_phonemes_.count(reversed_phonemes)) {
        std::set<size_t> all_forms;

        for (size_t p_id :
             pronunciations_by_blank_phonemes_.at(reversed_phonemes)) {
          const Pronunciation& other_pronunciation = pronunciations_.at(p_id);
          for (size_t form_id : other_pronunciation.form_ids) {
            all_forms.insert(form_id);
          }
        }

        for (size_t f_id1 : pronunciation.form_ids) {
          for (size_t f_id2 : all_forms) {
            Form& form = forms_.at(f_id1);
            form.puzzles[kBlackTop].insert(f_id2);
          }
        }
      }
    }
  }

  // Black Middle
  {
    hatkirby::progress ppgs("Generating black middle puzzles...",
                            forms_.size());

    for (Form& form : forms_) {
      ppgs.update();

      std::string reversed_text = form.text;
      std::reverse(reversed_text.begin(), reversed_text.end());

      if (form_by_text_.count(reversed_text)) {
        form.puzzles[kBlackMiddle].insert(form_by_text_.at(reversed_text));
      }
    }
  }

  // Black Bottom
  std::unordered_map<size_t, std::set<size_t>> antonyms;
  {
    std::list<std::string> lines(readFile(wordNetPath_ + "wn_ant.pl", true));

    hatkirby::progress ppgs("Generating black bottom puzzles...", lines.size());
    for (const std::string& line : lines) {
      ppgs.update();

      std::regex relation(
          "^ant\\(([134]\\d{8}),(\\d+),([134]\\d{8}),(\\d+)\\)\\.");

      std::smatch relation_data;
      if (!std::regex_search(line, relation_data, relation)) {
        continue;
      }

      std::pair<int, int> lookup1(std::stoi(relation_data[1]),
                                  std::stoi(relation_data[2]));

      std::pair<int, int> lookup2(std::stoi(relation_data[3]),
                                  std::stoi(relation_data[4]));

      if (word_by_wnid_and_wnum_.count(lookup1) &&
          word_by_wnid_and_wnum_.count(lookup2)) {
        const Word& word1 = words_.at(word_by_wnid_and_wnum_.at(lookup1));
        const Word& word2 = words_.at(word_by_wnid_and_wnum_.at(lookup2));

        Form& form1 = forms_.at(word1.base_form_id);
        form1.puzzles[kBlackBottom].insert(word2.base_form_id);

        antonyms[word1.id].insert(word2.id);
      }
    }
  }

  // Black Double Bottom
  {
    hatkirby::progress ppgs("Generating black double bottom puzzles...",
                            antonyms.size());
    for (const auto& [word1, ant_words] : antonyms) {
      ppgs.update();

      for (size_t word2 : ant_words) {
        const Word& word2_obj = words_.at(word2);
        const Form& form2 = forms_.at(word2_obj.base_form_id);

        for (size_t word25 : form2.word_ids) {
          if (word25 == word2) {
            continue;
          }

          const auto& double_ant_words = antonyms[word25];

          for (size_t word3 : double_ant_words) {
            const Word& word1_obj = words_.at(word1);
            const Word& word3_obj = words_.at(word3);

            bool synset_overlap = false;
            for (size_t synset1 : word1_obj.synsets) {
              for (size_t synset3 : word3_obj.synsets) {
                if (synset1 == synset3) {
                  synset_overlap = true;
                  break;
                }
              }
              if (synset_overlap) {
                break;
              }
            }
            if (!synset_overlap) {
              Form& form1 = forms_.at(word1_obj.base_form_id);
              form1.puzzles[kDoubleBlackBottom].insert(word3_obj.base_form_id);
            }
          }
        }
      }
    }
  }

  // Red/Blue Top
  {
    std::map<std::list<std::string>, std::vector<size_t>> tokenized;
    for (const auto& [phonemes, pronunciations] :
         pronunciations_by_blank_phonemes_) {
      tokenized[hatkirby::split<std::list<std::string>>(phonemes, " ")] =
          pronunciations;
    }

    hatkirby::progress ppgs("Generating top red/blue puzzles...",
                            tokenized.size());
    for (const auto& [phonemes, pronunciations] : tokenized) {
      ppgs.update();

      std::set<std::list<std::string>> visited;
      for (int i = 0; i < phonemes.size(); i++) {
        for (int l = 2; l <= phonemes.size() - i; l++) {
          if (i == 0 && l == phonemes.size()) {
            continue;
          }

          std::list<std::string> sublist;
          for (auto j = std::next(phonemes.begin(), i);
               j != std::next(phonemes.begin(), i + l); j++) {
            sublist.push_back(*j);
          }

          if (tokenized.count(sublist) && !visited.count(sublist)) {
            visited.insert(sublist);

            for (size_t holophone_id : pronunciations) {
              for (size_t merophone_id : tokenized[sublist]) {
                const Pronunciation& holophone =
                    pronunciations_.at(holophone_id);
                const Pronunciation& merophone =
                    pronunciations_.at(merophone_id);

                for (size_t holo_form_id : holophone.form_ids) {
                  Form& holo_form = forms_.at(holo_form_id);
                  for (size_t mero_form_id : merophone.form_ids) {
                    Form& mero_form = forms_.at(mero_form_id);

                    holo_form.puzzles[kBlueTop].insert(mero_form_id);
                    mero_form.puzzles[kRedTop].insert(holo_form_id);
                  }
                }
              }
            }
          }
        }
      }
    }
  }

  // Red/Blue Middle
  std::unordered_map<size_t, std::set<size_t>> left_shorter_by_longer;
  std::unordered_map<size_t, std::set<size_t>> left_longer_by_shorter;
  std::unordered_map<size_t, std::set<size_t>> right_shorter_by_longer;
  std::unordered_map<size_t, std::set<size_t>> right_longer_by_shorter;
  {
    hatkirby::progress ppgs("Generating red/blue middle puzzles...",
                            form_by_text_.size());
    for (const auto& [text, form_id] : form_by_text_) {
      ppgs.update();

      Form& holograph = forms_.at(form_id);
      std::unordered_set<std::string> visited;
      for (int i = 0; i < text.size(); i++) {
        for (int l = 3; l <= text.size() - i; l++) {
          if (i == 0 && l == text.size()) {
            continue;
          }

          std::string substr = text.substr(i, l);
          if (form_by_text_.count(substr) && !visited.count(substr)) {
            visited.insert(substr);

            Form& merograph = forms_.at(form_by_text_.at(substr));

            holograph.puzzles[kBlueMiddle].insert(merograph.id);
            merograph.puzzles[kRedMiddle].insert(form_id);

            if (i == 0) {
              left_shorter_by_longer[form_id].insert(merograph.id);
              left_longer_by_shorter[merograph.id].insert(form_id);
            } else if (i + l == text.size()) {
              right_shorter_by_longer[form_id].insert(merograph.id);
              right_longer_by_shorter[merograph.id].insert(form_id);
            }
          }
        }
      }
    }
  }

  // Purple Middle
  {
    hatkirby::progress ppgs(
        "Generating purple middle puzzles...",
        left_shorter_by_longer.size() + right_shorter_by_longer.size());

    for (const auto& [holograph_id, merograph_ids] : left_shorter_by_longer) {
      ppgs.update();

      Form& holograph = forms_.at(holograph_id);
      for (size_t merograph_id : merograph_ids) {
        for (size_t other_id : left_longer_by_shorter[merograph_id]) {
          if (other_id != holograph_id) {
            holograph.puzzles[kPurpleMiddle].insert(other_id);
          }
        }
      }
    }

    for (const auto& [holograph_id, merograph_ids] : right_shorter_by_longer) {
      ppgs.update();

      Form& holograph = forms_.at(holograph_id);
      for (size_t merograph_id : merograph_ids) {
        for (size_t other_id : right_longer_by_shorter[merograph_id]) {
          if (other_id != holograph_id) {
            holograph.puzzles[kPurpleMiddle].insert(other_id);
          }
        }
      }
    }
  }

  // Red/Blue Bottom
  std::unordered_map<size_t, std::set<size_t>> meronyms_by_holonym;
  {
    std::list<std::string> lines(readFile(wordNetPath_ + "wn_mm.pl"));
    hatkirby::progress ppgs("Reading member meronymy...", lines.size());
    for (auto line : lines) {
      ppgs.update();

      std::regex relation("^mm\\((1\\d{8}),(1\\d{8})\\)\\.");
      std::smatch relation_data;
      if (!std::regex_search(line, relation_data, relation)) {
        continue;
      }

      int lookup1 = std::stoi(relation_data[1]);
      int lookup2 = std::stoi(relation_data[2]);

      if (synset_by_wnid_.count(lookup1) && synset_by_wnid_.count(lookup2)) {
        for (size_t word_id1 : synsets_.at(synset_by_wnid_.at(lookup1))) {
          for (size_t word_id2 : synsets_.at(synset_by_wnid_.at(lookup2))) {
            meronyms_by_holonym[word_id1].insert(word_id2);
          }
        }
      }
    }
  }

  {
    std::list<std::string> lines(readFile(wordNetPath_ + "wn_mp.pl"));
    hatkirby::progress ppgs("Reading part meronymy...", lines.size());
    for (auto line : lines) {
      ppgs.update();

      std::regex relation("^mp\\((1\\d{8}),(1\\d{8})\\)\\.");
      std::smatch relation_data;
      if (!std::regex_search(line, relation_data, relation)) {
        continue;
      }

      int lookup1 = std::stoi(relation_data[1]);
      int lookup2 = std::stoi(relation_data[2]);

      if (synset_by_wnid_.count(lookup1) && synset_by_wnid_.count(lookup2)) {
        for (size_t word_id1 : synsets_.at(synset_by_wnid_.at(lookup1))) {
          for (size_t word_id2 : synsets_.at(synset_by_wnid_.at(lookup2))) {
            meronyms_by_holonym[word_id1].insert(word_id2);
          }
        }
      }
    }
  }

  {
    std::list<std::string> lines(readFile(wordNetPath_ + "wn_ms.pl"));
    hatkirby::progress ppgs("Reading substance meronymy...", lines.size());
    for (auto line : lines) {
      ppgs.update();

      std::regex relation("^ms\\((1\\d{8}),(1\\d{8})\\)\\.");
      std::smatch relation_data;
      if (!std::regex_search(line, relation_data, relation)) {
        continue;
      }

      int lookup1 = std::stoi(relation_data[1]);
      int lookup2 = std::stoi(relation_data[2]);

      if (synset_by_wnid_.count(lookup1) && synset_by_wnid_.count(lookup2)) {
        for (size_t word_id1 : synsets_.at(synset_by_wnid_.at(lookup1))) {
          for (size_t word_id2 : synsets_.at(synset_by_wnid_.at(lookup2))) {
            meronyms_by_holonym[word_id1].insert(word_id2);
          }
        }
      }
    }
  }

  {
    hatkirby::progress ppgs("Generating red/blue bottom puzzles...",
                            meronyms_by_holonym.size());

    for (const auto& [holonym_id, meronym_ids] : meronyms_by_holonym) {
      ppgs.update();

      for (size_t meronym_id : meronym_ids) {
        const Word& holonym_word = words_.at(holonym_id);
        const Word& meronym_word = words_.at(meronym_id);

        Form& holonym_form = forms_.at(holonym_word.base_form_id);
        Form& meronym_form = forms_.at(meronym_word.base_form_id);

        holonym_form.puzzles[kBlueBottom].insert(meronym_form.id);
        meronym_form.puzzles[kRedBottom].insert(holonym_form.id);
      }
    }
  }

  // Purple Top
  {
    hatkirby::progress ppgs("Generating purple top puzzles...",
                            pronunciations_by_rhyme_.size());

    for (const auto& [rhyme, pronunciation_ids] : pronunciations_by_rhyme_) {
      ppgs.update();

      for (size_t p_id1 : pronunciation_ids) {
        const Pronunciation& p1 = pronunciations_.at(p_id1);
        if (p1.prerhyme.empty()) {
          continue;
        }

        for (size_t p_id2 : pronunciation_ids) {
          const Pronunciation& p2 = pronunciations_.at(p_id2);
          if (p2.prerhyme.empty()) {
            continue;
          }

          if (p1.prerhyme != p2.prerhyme) {
            for (size_t f_id1 : p1.form_ids) {
              for (size_t f_id2 : p2.form_ids) {
                if (f_id1 != f_id2) {
                  Form& form1 = forms_.at(f_id1);
                  form1.puzzles[kPurpleTop].insert(f_id2);
                }
              }
            }
          }
        }
      }
    }
  }

  // Count up all of the generated puzzles.
  int total_puzzles = 0;
  int reusable_words = 0;
  std::unordered_map<PuzzleType, int> per_puzzle_type;
  for (const Form& form : forms_) {
    for (const auto& [puzzle_type, puzzles] : form.puzzles) {
      total_puzzles += puzzles.size();
      per_puzzle_type[puzzle_type]++;
    }
    if (form.puzzles.size() > 1) {
      reusable_words++;
    }
  }
  std::cout << "Puzzles: " << total_puzzles << std::endl;
  std::cout << "Reusable words: " << reusable_words << std::endl;
  std::cout << "White tops: " << per_puzzle_type[kWhiteTop] << std::endl;
  std::cout << "White bottom: " << per_puzzle_type[kWhiteBottom] << std::endl;
  std::cout << "Yellow tops: " << per_puzzle_type[kYellowTop] << std::endl;
  std::cout << "Yellow middles: " << per_puzzle_type[kYellowMiddle]
            << std::endl;
  std::cout << "Black tops: " << per_puzzle_type[kBlackTop] << std::endl;
  std::cout << "Black middles: " << per_puzzle_type[kBlackMiddle] << std::endl;
  std::cout << "Black bottoms: " << per_puzzle_type[kBlackBottom] << std::endl;
  std::cout << "Black double bottoms: " << per_puzzle_type[kDoubleBlackBottom]
            << std::endl;
  std::cout << "Red tops: " << per_puzzle_type[kRedTop] << std::endl;
  std::cout << "Red middles: " << per_puzzle_type[kRedMiddle] << std::endl;
  std::cout << "Red bottoms: " << per_puzzle_type[kRedBottom] << std::endl;
  std::cout << "Blue tops: " << per_puzzle_type[kBlueTop] << std::endl;
  std::cout << "Blue middles: " << per_puzzle_type[kBlueMiddle] << std::endl;
  std::cout << "Blue bottoms: " << per_puzzle_type[kBlueBottom] << std::endl;
  std::cout << "Purple tops: " << per_puzzle_type[kPurpleTop] << std::endl;
  std::cout << "Purple middles: " << per_puzzle_type[kPurpleMiddle]
            << std::endl;
}

size_t generator::LookupOrCreatePronunciation(const std::string& phonemes) {
  if (pronunciation_by_phonemes_.count(phonemes)) {
    return pronunciation_by_phonemes_[phonemes];
  } else {
    size_t pronunciation_id = pronunciations_.size();

    auto phonemeList = hatkirby::split<std::list<std::string>>(phonemes, " ");

    std::list<std::string>::iterator rhymeStart =
        std::find_if(std::begin(phonemeList), std::end(phonemeList),
                     [](std::string phoneme) {
                       return phoneme.find("1") != std::string::npos;
                     });

    // Rhyme detection
    std::string prerhyme = "";
    std::string rhyme = "";
    if (rhymeStart != std::end(phonemeList)) {
      std::list<std::string> rhymePhonemes;

      std::transform(
          rhymeStart, std::end(phonemeList), std::back_inserter(rhymePhonemes),
          [](std::string phoneme) {
            std::string naked;

            std::remove_copy_if(std::begin(phoneme), std::end(phoneme),
                                std::back_inserter(naked),
                                [](char ch) { return std::isdigit(ch); });

            return naked;
          });

      rhyme = hatkirby::implode(std::begin(rhymePhonemes),
                                std::end(rhymePhonemes), " ");

      if (rhymeStart != std::begin(phonemeList)) {
        prerhyme = *std::prev(rhymeStart);
      }

      pronunciations_by_rhyme_[rhyme].push_back(pronunciation_id);
    }

    std::string stressless;
    for (int i = 0; i < phonemes.size(); i++) {
      if (!std::isdigit(phonemes[i])) {
        stressless.push_back(phonemes[i]);
      }
    }
    auto stresslessList =
        hatkirby::split<std::vector<std::string>>(stressless, " ");
    std::string stresslessPhonemes =
        hatkirby::implode(stresslessList.begin(), stresslessList.end(), " ");
    std::sort(stresslessList.begin(), stresslessList.end());
    std::string sortedPhonemes =
        hatkirby::implode(stresslessList.begin(), stresslessList.end(), " ");

    pronunciations_.push_back({.id = pronunciation_id,
                               .phonemes = phonemes,
                               .prerhyme = prerhyme,
                               .rhyme = rhyme,
                               .stressless_phonemes = stresslessPhonemes});

    AddPronunciationToAnaphoneSet(pronunciation_id, sortedPhonemes);

    pronunciation_by_phonemes_[phonemes] = pronunciation_id;
    pronunciations_by_blank_phonemes_[stresslessPhonemes].push_back(
        pronunciation_id);

    return pronunciation_id;
  }
}

size_t generator::LookupOrCreateForm(const std::string& word) {
  if (form_by_text_.count(word)) {
    return form_by_text_[word];
  } else {
    size_t form_id = forms_.size();
    form_by_text_[word] = form_id;
    forms_.push_back({.id = form_id, .text = word});

    std::string sortedText = word;
    std::sort(sortedText.begin(), sortedText.end());
    AddFormToAnagramSet(form_id, sortedText);

    return form_id;
  }
}

size_t generator::LookupOrCreateWord(const std::string& word) {
  size_t word_id = words_.size();
  words_by_base_[word].push_back(word_id);
  size_t form_id = LookupOrCreateForm(word);
  words_.push_back({.id = word_id, .base_form_id = form_id});
  AddFormToWord(form_id, word_id);
  return word_id;
}

void generator::AddPronunciationToForm(size_t pronunciation_id,
                                       size_t form_id) {
  pronunciations_[pronunciation_id].form_ids.push_back(form_id);
  forms_[form_id].pronunciation_ids.push_back(pronunciation_id);
}

void generator::AddFormToWord(size_t form_id, size_t word_id) {
  words_[word_id].form_ids.push_back(form_id);
  forms_[form_id].word_ids.push_back(word_id);
}

void generator::AddWordToSynset(size_t word_id, int wnid) {
  if (!synset_by_wnid_.count(wnid)) {
    synset_by_wnid_[wnid] = synsets_.size();
    synsets_.push_back({word_id});
    words_[word_id].synsets.push_back(synsets_.size() - 1);
  } else {
    size_t synset_id = synset_by_wnid_[wnid];
    synsets_[synset_id].push_back(word_id);
    words_[word_id].synsets.push_back(synset_id);
  }
}

void generator::AddFormToAnagramSet(size_t form_id,
                                    const std::string& sorted_letters) {
  if (!anagram_set_by_sorted_letters_.count(sorted_letters)) {
    anagram_set_by_sorted_letters_[sorted_letters] = anagram_sets_.size();
    anagram_sets_.push_back({form_id});
    forms_[form_id].anagram_set_id = anagram_sets_.size() - 1;
  } else {
    size_t anagram_set_id = anagram_set_by_sorted_letters_[sorted_letters];
    anagram_sets_[anagram_set_id].push_back(form_id);
    forms_[form_id].anagram_set_id = anagram_set_id;
  }
}

void generator::AddPronunciationToAnaphoneSet(
    size_t pronunciation_id, const std::string& sorted_phonemes) {
  if (!anaphone_set_by_sorted_phonemes_.count(sorted_phonemes)) {
    anaphone_set_by_sorted_phonemes_[sorted_phonemes] = anaphone_sets_.size();
    anaphone_sets_.push_back({pronunciation_id});
    pronunciations_[pronunciation_id].anaphone_set_id =
        anaphone_sets_.size() - 1;
  } else {
    size_t anaphone_set_id = anaphone_set_by_sorted_phonemes_[sorted_phonemes];
    anaphone_sets_[anaphone_set_id].push_back(pronunciation_id);
    pronunciations_[pronunciation_id].anaphone_set_id = anaphone_set_id;
  }
}