module Opener
  module Stanza
    class TokenizerPos

      DESC            = 'Tokenizer / POS by Stanza'
      VERSION         = '1.0'

      BASE_URL        = ENV['STANZA_SERVER']
      LANGUAGES_CACHE = Opener::ChainedDaemon::LanguagesCache.new

      RTL_LANGUAGES   = [ "ar", "ara", "arc", "ae", "ave", "egy", "he", "heb", "nqo", "pal", "phn", "sam",
                          "syc", "syr", "fa", "per", "fas", "ku", "kur", "ur", "urd" ]

      POS             = {
        'DET'   => 'D',
        'ADJ'   => 'G',
        'NOUN'  => 'N',
        'VERB'  => 'V',
        'AUX'   => 'V',
        'ADV'   => 'A',
        'CCONJ' => 'J',
        'PUNCT' => '.',
        'ADP'   => 'P',
        'PRON'  => 'Q',
        'PROPN' => 'R',
        'PART'  => 'P',
        'NUM'   => 'O',
        'X'     => 'O',
        'SYM'   => 'O',
        'SCONJ' => 'P',
        'INTJ'  => 'O',
      }

      POS_OPEN        = %w[N R G V A O]

      def run input, params
        raise 'missing Stanza server' if ENV['STANZA_SERVER'].blank?

        kaf      = KAF::Document.from_xml input

        prod     = params[:cache_keys][:environment] != 'staging'
        if prod and !LANGUAGES_CACHE.get.include?(kaf.language)
          raise Core::UnsupportedLanguageError.new kaf.language
        end

        input    = kaf.raw
        input    = input.gsub(/\,[^\ ]/, ', ')
        response = Faraday.post BASE_URL, {lang: kaf.language, input: input}.to_query
        raise Core::UnsupportedLanguageError, kaf.language if response.status == 406
        raise response.body if response.status >= 400
        tokens   = JSON.parse response.body

        w_index = 0

        miscs = {}
        tokens.each_with_index do |t, i|
          miscs[i] ||= {}
          t.each do |word|
            word['id'].is_a?(Array) && word['id'].each do |id|
              puts id
              miscs[i][id] = word['misc']
            end
          end
        end

        tokens.map{ |t| t.reverse! } if RTL_LANGUAGES.include? kaf.language
        tokens.each_with_index do |sentence, s_index|
          sentence.each_with_index do |word|
            w_index += 1
            # save misc for later usase in a MWT case
            next if word['id'].is_a? Array

            misc = word['misc'] || miscs[s_index][word['id']]

            Rollbar.scoped({ input: input, params: params, tokens: tokens, word: word }) do
              raise 'Missing misc'
            end if misc.nil?

            offset = misc.match(/start_char=(\d+)|/)[1].to_i
            length = misc.match(/end_char=(\d+)/)[1].to_i - offset

            u_pos  = word['upos']
            pos    = POS[u_pos]
            raise "Didn't find a map for #{u_pos}" if pos.nil?
            type   = if POS_OPEN.include? pos then 'open' else 'close' end

            params = {
              wid:        w_index,
              sid:        s_index + 1,
              tid:        w_index,
              para:       1,
              offset:     offset,
              length:     length,
              text:       word['text'],
              lemma:      word['lemma'],
              morphofeat: u_pos,
              pos:        pos,
              type:       type,
            }

            kaf.add_word_form params
            kaf.add_term params
          end
        end

        kaf.add_linguistic_processor DESC, "#{VERSION}", 'text', timestamp: true

        kaf.to_xml
      end

    end
  end
end