discourse/app/services/search_indexer.rb

require_dependency 'search'

class SearchIndexer

  def self.disable
    @disabled = true
  end

  def self.enable
    @disabled = false
  end

  def self.scrub_html_for_search(html)
    HtmlScrubber.scrub(html)
  end

  def self.update_index(table, id, raw_data)
    raw_data = Search.prepare_data(raw_data, :index)

    table_name = "#{table}_search_data"
    foreign_key = "#{table}_id"

    # insert some extra words for I.am.a.word so "word" is tokenized
    # I.am.a.word becomes I.am.a.word am a word
    # uses \p{L} which matchs a single code point in category letter
    # uses \p{N} which matchs a single code point in category number
    search_data = raw_data.gsub(/(\p{L}|\p{N}|_|-|\.)*\.(\p{L}|\p{N}|_|-|\.)*/) do |with_dot|
      split = with_dot.split(".")
      if split.length > 1
        with_dot + (" " << split[1..-1].join(" "))
      else
        with_dot
      end
    end

    # for user login and name use "simple" lowercase stemmer
    stemmer = table == "user" ? "simple" : Search.ts_config

    # Would be nice to use AR here but not sure how to execut Postgres functions
    # when inserting data like this.
    rows = Post.exec_sql_row_count("UPDATE #{table_name}
                                   SET
                                      raw_data = :raw_data,
                                      locale = :locale,
                                      search_data = TO_TSVECTOR('#{stemmer}', :search_data),
                                      version = :version
                                   WHERE #{foreign_key} = :id",
                                    raw_data: raw_data,
                                    search_data: search_data,
                                    id: id,
                                    locale: SiteSetting.default_locale,
                                    version: Search::INDEX_VERSION)
    if rows == 0
      Post.exec_sql("INSERT INTO #{table_name}
                    (#{foreign_key}, search_data, locale, raw_data, version)
                    VALUES (:id, TO_TSVECTOR('#{stemmer}', :search_data), :locale, :raw_data, :version)",
                                    raw_data: raw_data,
                                    search_data: search_data,
                                    id: id,
                                    locale: SiteSetting.default_locale,
                                    version: Search::INDEX_VERSION)
    end
  rescue
    # don't allow concurrency to mess up saving a post
  end

  def self.update_topics_index(topic_id, title, cooked)
    search_data = title.dup << " " << scrub_html_for_search(cooked)[0...Topic::MAX_SIMILAR_BODY_LENGTH]
    update_index('topic', topic_id, search_data)
  end

  def self.update_posts_index(post_id, cooked, title, category)
    search_data = scrub_html_for_search(cooked) << " " << title.dup.force_encoding('UTF-8')
    search_data << " " << category if category
    update_index('post', post_id, search_data)
  end

  def self.update_users_index(user_id, username, name)
    search_data = username.dup << " " << (name || "")
    update_index('user', user_id, search_data)
  end

  def self.update_categories_index(category_id, name)
    update_index('category', category_id, name)
  end

  def self.update_tags_index(tag_id, name)
    update_index('tag', tag_id, name)
  end

  def self.index(obj, force: false)
    return if @disabled

    if obj.class == Post && (obj.cooked_changed? || force)
      if obj.topic
        category_name = obj.topic.category.name if obj.topic.category
        SearchIndexer.update_posts_index(obj.id, obj.cooked, obj.topic.title, category_name)
        SearchIndexer.update_topics_index(obj.topic_id, obj.topic.title, obj.cooked) if obj.is_first_post?
      else
        Rails.logger.warn("Orphan post skipped in search_indexer, topic_id: #{obj.topic_id} post_id: #{obj.id} raw: #{obj.raw}")
      end
    end

    if obj.class == User && (obj.username_changed? || obj.name_changed? || force)
      SearchIndexer.update_users_index(obj.id, obj.username_lower || '', obj.name ? obj.name.downcase : '')
    end

    if obj.class == Topic && (obj.title_changed? || force)
      if obj.posts
        post = obj.posts.find_by(post_number: 1)
        if post
          category_name = obj.category.name if obj.category
          SearchIndexer.update_posts_index(post.id, post.cooked, obj.title, category_name)
          SearchIndexer.update_topics_index(obj.id, obj.title, post.cooked)
        end
      end
    end

    if obj.class == Category && (obj.name_changed? || force)
      SearchIndexer.update_categories_index(obj.id, obj.name)
    end

    if obj.class == Tag && (obj.name_changed? || force)
      SearchIndexer.update_tags_index(obj.id, obj.name)
    end
  end

  class HtmlScrubber < Nokogiri::XML::SAX::Document
    attr_reader :scrubbed

    def initialize
      @scrubbed = ""
    end

    def self.scrub(html)
      me = new
      parser = Nokogiri::HTML::SAX::Parser.new(me)
      begin
        copy = "<div>"
        copy << html unless html.nil?
        copy << "</div>"
        parser.parse(html) unless html.nil?
      end
      me.scrubbed
    end

    def start_element(name, attributes = [])
      attributes = Hash[*attributes.flatten]
      if attributes["alt"]
        scrubbed << " "
        scrubbed << attributes["alt"]
        scrubbed << " "
      end
      if attributes["title"]
        scrubbed << " "
        scrubbed << attributes["title"]
        scrubbed << " "
      end
    end

    def characters(string)
      scrubbed << " "
      scrubbed << string
      scrubbed << " "
    end
  end
end