discourse/app/services/search_indexer.rb

208 lines
5.9 KiB
Ruby
Raw Normal View History

# frozen_string_literal: true
require_dependency 'search'
class SearchIndexer
def self.disable
@disabled = true
end
def self.enable
@disabled = false
end
2013-02-06 03:16:51 +08:00
def self.scrub_html_for_search(html)
HtmlScrubber.scrub(html)
end
def self.inject_extra_terms(raw)
# 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
raw.gsub(/[^[:space:]]*[\.]+[^[:space:]]*/) do |with_dot|
split = with_dot.split(".")
if split.length > 1
with_dot + ((+" ") << split[1..-1].join(" "))
else
with_dot
end
end
end
def self.update_index(table: , id: , raw_data:)
search_data = raw_data.map do |data|
inject_extra_terms(Search.prepare_data(data || "", :index))
end
table_name = "#{table}_search_data"
foreign_key = "#{table}_id"
# for user login and name use "simple" lowercase stemmer
stemmer = table == "user" ? "simple" : Search.ts_config
ranked_index = <<~SQL
setweight(to_tsvector('#{stemmer}', coalesce(:a,'')), 'A') ||
setweight(to_tsvector('#{stemmer}', coalesce(:b,'')), 'B') ||
setweight(to_tsvector('#{stemmer}', coalesce(:c,'')), 'C') ||
setweight(to_tsvector('#{stemmer}', coalesce(:d,'')), 'D')
SQL
indexed_data = search_data.select { |d| d.length > 0 }.join(' ')
params = {
a: search_data[0],
b: search_data[1],
c: search_data[2],
d: search_data[3],
raw_data: indexed_data,
id: id,
locale: SiteSetting.default_locale,
version: Search::INDEX_VERSION
}
# 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(<<~SQL, params)
UPDATE #{table_name}
SET
raw_data = :raw_data,
locale = :locale,
search_data = #{ranked_index},
version = :version
WHERE #{foreign_key} = :id
SQL
if rows == 0
Post.exec_sql(<<~SQL, params)
INSERT INTO #{table_name}
(#{foreign_key}, search_data, locale, raw_data, version)
VALUES (:id, #{ranked_index}, :locale, :raw_data, :version)
SQL
2013-02-06 03:16:51 +08:00
end
rescue
# TODO is there any way we can safely avoid this?
# best way is probably pushing search indexer into a dedicated process so it no longer happens on save
# instead in the post processor
2013-02-06 03:16:51 +08:00
end
def self.update_topics_index(topic_id, title, cooked)
scrubbed_cooked = scrub_html_for_search(cooked)[0...Topic::MAX_SIMILAR_BODY_LENGTH]
# a bit inconsitent that we use title as A and body as B when in
# the post index body is C
update_index(table: 'topic', id: topic_id, raw_data: [title, scrubbed_cooked])
end
def self.update_posts_index(post_id, title, category, tags, cooked)
update_index(table: 'post', id: post_id, raw_data: [title, category, tags, scrub_html_for_search(cooked)])
2013-02-06 03:16:51 +08:00
end
def self.update_users_index(user_id, username, name)
update_index(table: 'user', id: user_id, raw_data: [username, name])
2013-02-06 03:16:51 +08:00
end
2013-02-07 23:45:24 +08:00
2013-02-06 03:16:51 +08:00
def self.update_categories_index(category_id, name)
update_index(table: 'category', id: category_id, raw_data: [name])
2013-02-06 03:16:51 +08:00
end
2017-08-25 23:52:18 +08:00
def self.update_tags_index(tag_id, name)
update_index(table: 'tag', id: tag_id, raw_data: [name])
end
def self.queue_post_reindex(topic_id)
return if @disabled
ActiveRecord::Base.exec_sql(<<~SQL, topic_id: topic_id)
UPDATE post_search_data
SET version = 0
WHERE post_id IN (SELECT id FROM posts WHERE topic_id = :topic_id)
SQL
2017-08-25 23:52:18 +08:00
end
def self.index(obj, force: false)
return if @disabled
category_name, tag_names = nil
topic = nil
if Topic === obj
topic = obj
elsif Post === obj
topic = obj.topic
end
category_name = topic.category&.name if topic
tag_names = topic.tags.pluck(:name).join(' ') if topic
if Post === obj && (obj.saved_change_to_cooked? || force)
if topic
SearchIndexer.update_posts_index(obj.id, topic.title, category_name, tag_names, obj.cooked)
SearchIndexer.update_topics_index(topic.id, topic.title, obj.cooked) if obj.is_first_post?
2014-05-07 10:35:26 +08:00
else
Rails.logger.warn("Orphan post skipped in search_indexer, topic_id: #{obj.topic_id} post_id: #{obj.id} raw: #{obj.raw}")
2014-05-07 10:35:26 +08:00
end
2013-02-06 03:16:51 +08:00
end
if User === obj && (obj.saved_change_to_username? || obj.saved_change_to_name? || force)
SearchIndexer.update_users_index(obj.id, obj.username_lower || '', obj.name ? obj.name.downcase : '')
2013-02-06 03:16:51 +08:00
end
if Topic === obj && (obj.saved_change_to_title? || force)
2013-02-06 03:16:51 +08:00
if obj.posts
post = obj.posts.find_by(post_number: 1)
2013-02-06 03:16:51 +08:00
if post
SearchIndexer.update_posts_index(post.id, obj.title, category_name, tag_names, post.cooked)
SearchIndexer.update_topics_index(obj.id, obj.title, post.cooked)
2013-02-06 03:16:51 +08:00
end
end
end
if Category === obj && (obj.saved_change_to_name? || force)
SearchIndexer.update_categories_index(obj.id, obj.name)
2013-02-06 03:16:51 +08:00
end
2017-08-25 23:52:18 +08:00
if Tag === obj && (obj.saved_change_to_name? || force)
2017-08-25 23:52:18 +08:00
SearchIndexer.update_tags_index(obj.id, obj.name)
end
2013-02-06 03:16:51 +08:00
end
class HtmlScrubber < Nokogiri::XML::SAX::Document
attr_reader :scrubbed
def initialize
@scrubbed = +""
2013-02-06 03:16:51 +08:00
end
def self.scrub(html)
me = new
2013-02-06 03:16:51 +08:00
parser = Nokogiri::HTML::SAX::Parser.new(me)
2013-02-07 23:45:24 +08:00
begin
copy = +"<div>"
2013-02-06 03:16:51 +08:00
copy << html unless html.nil?
copy << "</div>"
parser.parse(html) unless html.nil?
end
me.scrubbed
end
2017-07-28 09:20:09 +08:00
def start_element(name, attributes = [])
2013-02-06 03:16:51 +08:00
attributes = Hash[*attributes.flatten]
if attributes["alt"]
scrubbed << " "
scrubbed << attributes["alt"]
scrubbed << " "
end
if attributes["title"]
scrubbed << " "
2013-02-07 23:45:24 +08:00
scrubbed << attributes["title"]
2013-02-06 03:16:51 +08:00
scrubbed << " "
end
end
def characters(string)
scrubbed << " "
scrubbed << string
2013-02-06 03:16:51 +08:00
scrubbed << " "
end
end
end