# Author: Nic Wolfe # URL: http://code.google.com/p/sickbeard/ # # This file is part of SickGear. # # SickGear is free software: you can redistribute it and/or modify # it under the terms of the GNU General Public License as published by # the Free Software Foundation, either version 3 of the License, or # (at your option) any later version. # # SickGear is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with SickGear. If not, see . from __future__ import with_statement import os.path import datetime import threading import re import glob import stat import traceback import shutil import sickbeard import xml.etree.cElementTree as etree from name_parser.parser import NameParser, InvalidNameException, InvalidShowException from lib import subliminal import fnmatch from imdb._exceptions import IMDbError try: from lib.send2trash import send2trash except ImportError: pass from lib.imdb import imdb from sickbeard import db from sickbeard import helpers, exceptions, logger, name_cache, indexermapper from sickbeard.exceptions import ex from sickbeard import image_cache from sickbeard import notifiers from sickbeard import postProcessor from sickbeard import subtitles from sickbeard import history from sickbeard import network_timezones from sickbeard.sbdatetime import sbdatetime from sickbeard.blackandwhitelist import BlackAndWhiteList from sickbeard.indexermapper import del_mapping, save_mapping, MapStatus from sickbeard.generic_queue import QueuePriorities from sickbeard import encodingKludge as ek from common import Quality, Overview, statusStrings from common import SNATCHED, SNATCHED_PROPER, SNATCHED_BEST, SNATCHED_ANY, DOWNLOADED, ARCHIVED, \ IGNORED, UNAIRED, WANTED, SKIPPED, UNKNOWN, FAILED, SUBTITLED from common import NAMING_DUPLICATE, NAMING_EXTEND, NAMING_LIMITED_EXTEND, NAMING_SEPARATED_REPEAT, \ NAMING_LIMITED_EXTEND_E_PREFIXED concurrent_show_not_found_days = 7 show_not_found_retry_days = 7 def dirty_setter(attr_name, types=None): def wrapper(self, val): if getattr(self, attr_name) != val: if None is types or isinstance(val, types): setattr(self, attr_name, val) self.dirty = True else: logger.log('Didn\'t change property "%s" because expected: %s, but got: %s with value: %s' % (attr_name, types, type(val), val), logger.WARNING) return wrapper def dict_prevent_None(d, key, default): v = getattr(d, key, default) return (v, default)[None is v] class TVShow(object): def __init__(self, indexer, indexerid, lang=''): self._indexerid = int(indexerid) self._indexer = int(indexer) self._name = '' self._imdbid = '' self._network = '' self._genre = '' self._classification = '' self._runtime = 0 self._imdb_info = {} self._quality = int(sickbeard.QUALITY_DEFAULT) self._flatten_folders = int(sickbeard.FLATTEN_FOLDERS_DEFAULT) self._status = "" self._airs = "" self._startyear = 0 self._paused = 0 self._air_by_date = 0 self._subtitles = int(sickbeard.SUBTITLES_DEFAULT if sickbeard.SUBTITLES_DEFAULT else 0) self._dvdorder = 0 self._upgrade_once = 0 self._lang = lang self._last_update_indexer = 1 self._sports = 0 self._anime = 0 self._scene = 0 self._rls_ignore_words = '' self._rls_require_words = '' self._overview = '' self._tag = '' self._mapped_ids = {} self._not_found_count = None self._last_found_on_indexer = -1 self.dirty = True self._location = '' self.lock = threading.Lock() self.isDirGood = False self.episodes = {} self.nextaired = '' self.release_groups = None otherShow = helpers.findCertainShow(sickbeard.showList, self.indexerid) if otherShow != None: raise exceptions.MultipleShowObjectsException('Can\'t create a show if it already exists') self.loadFromDB() name = property(lambda self: self._name, dirty_setter('_name')) indexerid = property(lambda self: self._indexerid, dirty_setter('_indexerid')) indexer = property(lambda self: self._indexer, dirty_setter('_indexer')) # location = property(lambda self: self._location, dirty_setter('_location')) imdbid = property(lambda self: self._imdbid, dirty_setter('_imdbid')) network = property(lambda self: self._network, dirty_setter('_network')) genre = property(lambda self: self._genre, dirty_setter('_genre')) classification = property(lambda self: self._classification, dirty_setter('_classification')) runtime = property(lambda self: self._runtime, dirty_setter('_runtime')) imdb_info = property(lambda self: self._imdb_info, dirty_setter('_imdb_info')) quality = property(lambda self: self._quality, dirty_setter('_quality')) flatten_folders = property(lambda self: self._flatten_folders, dirty_setter('_flatten_folders')) status = property(lambda self: self._status, dirty_setter('_status')) airs = property(lambda self: self._airs, dirty_setter('_airs')) startyear = property(lambda self: self._startyear, dirty_setter('_startyear')) air_by_date = property(lambda self: self._air_by_date, dirty_setter('_air_by_date')) subtitles = property(lambda self: self._subtitles, dirty_setter('_subtitles')) dvdorder = property(lambda self: self._dvdorder, dirty_setter('_dvdorder')) upgrade_once = property(lambda self: self._upgrade_once, dirty_setter('_upgrade_once')) lang = property(lambda self: self._lang, dirty_setter('_lang')) last_update_indexer = property(lambda self: self._last_update_indexer, dirty_setter('_last_update_indexer')) sports = property(lambda self: self._sports, dirty_setter('_sports')) anime = property(lambda self: self._anime, dirty_setter('_anime')) scene = property(lambda self: self._scene, dirty_setter('_scene')) rls_ignore_words = property(lambda self: self._rls_ignore_words, dirty_setter('_rls_ignore_words')) rls_require_words = property(lambda self: self._rls_require_words, dirty_setter('_rls_require_words')) overview = property(lambda self: self._overview, dirty_setter('_overview')) tag = property(lambda self: self._tag, dirty_setter('_tag')) def _helper_load_failed_db(self): if None is self._not_found_count or self._last_found_on_indexer == -1: myDB = db.DBConnection() results = myDB.select('SELECT fail_count, last_success FROM tv_shows_not_found WHERE indexer = ? AND indexer_id = ?', [self.indexer, self.indexerid]) if results: self._not_found_count = helpers.tryInt(results[0]['fail_count']) self._last_found_on_indexer = helpers.tryInt(results[0]['last_success']) else: self._not_found_count = 0 self._last_found_on_indexer = 0 @property def not_found_count(self): self._helper_load_failed_db() return self._not_found_count @not_found_count.setter def not_found_count(self, v): if isinstance(v, (int, long)) and v != self._not_found_count: self._last_found_on_indexer = self.last_found_on_indexer myDB = db.DBConnection() # noinspection PyUnresolvedReferences last_check = sbdatetime.now().totimestamp(default=0) # in case of flag change (+/-) don't change last_check date if abs(v) == abs(self._not_found_count): results = myDB.select('SELECT last_check FROM tv_shows_not_found WHERE indexer = ? AND indexer_id = ?', [self.indexer, self.indexerid]) if results: last_check = helpers.tryInt(results[0]['last_check']) myDB.upsert('tv_shows_not_found', {'fail_count': v, 'last_check': last_check, 'last_success': self._last_found_on_indexer}, {'indexer': self.indexer, 'indexer_id': self.indexerid}) self._not_found_count = v @property def last_found_on_indexer(self): self._helper_load_failed_db() return (self._last_found_on_indexer, self.last_update_indexer)[self._last_found_on_indexer <= 0] def inc_not_found_count(self): myDB = db.DBConnection() results = myDB.select('SELECT last_check FROM tv_shows_not_found WHERE indexer = ? AND indexer_id = ?', [self.indexer, self.indexerid]) days = (show_not_found_retry_days - 1, 0)[abs(self.not_found_count) <= concurrent_show_not_found_days] if not results or datetime.datetime.fromtimestamp(helpers.tryInt(results[0]['last_check'])) + \ datetime.timedelta(days=days, hours=18) < datetime.datetime.now(): self.not_found_count += (-1, 1)[0 <= self.not_found_count] def reset_not_found_count(self): if 0 != self.not_found_count: self._not_found_count = 0 self._last_found_on_indexer = 0 myDB = db.DBConnection() myDB.action('DELETE FROM tv_shows_not_found WHERE indexer = ? AND indexer_id = ?', [self.indexer, self.indexerid]) @property def paused(self): return self._paused @paused.setter def paused(self, value): if value != self._paused: if isinstance(value, bool) or (isinstance(value, (int, long)) and value in [0, 1]): self._paused = int(value) self.dirty = True else: logger.log('tried to set paused property to invalid value: %s of type: %s' % (value, type(value)), logger.ERROR) @property def ids(self): if not self._mapped_ids: acquired_lock = self.lock.acquire(False) if acquired_lock: try: indexermapper.map_indexers_to_show(self) finally: self.lock.release() return self._mapped_ids @ids.setter def ids(self, value): if isinstance(value, dict): for k, v in value.iteritems(): if k not in sickbeard.indexermapper.indexer_list or not isinstance(v, dict) or \ not isinstance(v.get('id'), (int, long)) or not isinstance(v.get('status'), (int, long)) or \ v.get('status') not in indexermapper.MapStatus.allstatus or \ not isinstance(v.get('date'), datetime.date): return self._mapped_ids = value @property def is_anime(self): if int(self.anime) > 0: return True else: return False @property def is_sports(self): if int(self.sports) > 0: return True else: return False @property def is_scene(self): if int(self.scene) > 0: return True else: return False def _getLocation(self): # no dir check needed if missing show dirs are created during post-processing if sickbeard.CREATE_MISSING_SHOW_DIRS: return self._location if ek.ek(os.path.isdir, self._location): return self._location else: raise exceptions.ShowDirNotFoundException('Show folder doesn\'t exist, you shouldn\'t be using it') def _setLocation(self, newLocation): logger.log('Setter sets location to %s' % newLocation, logger.DEBUG) # Don't validate dir if user wants to add shows without creating a dir if sickbeard.ADD_SHOWS_WO_DIR or ek.ek(os.path.isdir, newLocation): dirty_setter('_location')(self, newLocation) self._isDirGood = True else: raise exceptions.NoNFOException('Invalid folder for the show!') location = property(_getLocation, _setLocation) # delete references to anything that's not in the internal lists def flushEpisodes(self): for curSeason in self.episodes: for curEp in self.episodes[curSeason]: myEp = self.episodes[curSeason][curEp] self.episodes[curSeason][curEp] = None del myEp def getAllEpisodes(self, season=None, has_location=False, check_related_eps=True): sql_selection = 'SELECT season, episode' if check_related_eps: # subselection to detect multi-episodes early, share_location > 0 sql_selection += ' , (SELECT COUNT (*) FROM tv_episodes WHERE showid = tve.showid AND season = ' \ 'tve.season AND location != "" AND location = tve.location AND episode != tve.episode) ' \ 'AS share_location ' sql_selection += ' FROM tv_episodes tve WHERE indexer = ? AND showid = ?' sql_parameter = [self.indexer, self.indexerid] if season is not None: sql_selection += ' AND season = ?' sql_parameter += [season] if has_location: sql_selection += ' AND location != "" ' # need ORDER episode ASC to rename multi-episodes in order S01E01-02 sql_selection += ' ORDER BY season ASC, episode ASC' myDB = db.DBConnection() results = myDB.select(sql_selection, sql_parameter) ep_list = [] for cur_result in results: cur_ep = self.getEpisode(int(cur_result['season']), int(cur_result['episode'])) if cur_ep: cur_ep.relatedEps = [] if check_related_eps and cur_ep.location: # if there is a location, check if it's a multi-episode (share_location > 0) and put them in relatedEps if cur_result['share_location'] > 0: related_eps_result = myDB.select( 'SELECT * FROM tv_episodes WHERE showid = ? AND season = ? AND location = ? AND ' 'episode != ? ORDER BY episode ASC', [self.indexerid, cur_ep.season, cur_ep.location, cur_ep.episode]) for cur_related_ep in related_eps_result: related_ep = self.getEpisode(int(cur_related_ep["season"]), int(cur_related_ep["episode"])) if related_ep not in cur_ep.relatedEps: cur_ep.relatedEps.append(related_ep) ep_list.append(cur_ep) return ep_list def getEpisode(self, season=None, episode=None, file=None, noCreate=False, absolute_number=None, forceUpdate=False, ep_sql=None): # if we get an anime get the real season and episode if self.is_anime and absolute_number and not season and not episode: myDB = db.DBConnection() sql = 'SELECT * FROM tv_episodes WHERE showid = ? and absolute_number = ? and season != 0' sqlResults = myDB.select(sql, [self.indexerid, absolute_number]) if len(sqlResults) == 1: episode = int(sqlResults[0]['episode']) season = int(sqlResults[0]['season']) logger.log( 'Found episode by absolute_number: %s which is %sx%s' % (absolute_number, season, episode), logger.DEBUG) elif len(sqlResults) > 1: logger.log('Multiple entries for absolute number: %s in show: %s found.' % (absolute_number, self.name), logger.ERROR) return None else: logger.log( 'No entries for absolute number: %s in show: %s found.' % (absolute_number, self.name), logger.DEBUG) return None if not season in self.episodes: self.episodes[season] = {} if not episode in self.episodes[season] or self.episodes[season][episode] is None: if noCreate: return None logger.log('%s: An object for episode %sx%s didn\'t exist in the cache, trying to create it' % (self.indexerid, season, episode), logger.DEBUG) if file: ep = TVEpisode(self, season, episode, file, show_sql=ep_sql) else: ep = TVEpisode(self, season, episode, show_sql=ep_sql) if ep != None: self.episodes[season][episode] = ep return self.episodes[season][episode] def should_update(self, update_date=datetime.date.today()): cur_indexerid = self.indexerid # In some situations self.status = None.. need to figure out where that is! if not self.status: self.status = '' logger.log('Status missing for showid: [%s] with status: [%s]' % (cur_indexerid, self.status), logger.DEBUG) last_update_indexer = datetime.date.fromordinal(self.last_update_indexer) # if show was not found for 1 week, only retry to update once a week if (concurrent_show_not_found_days < abs(self.not_found_count)) \ and (update_date - last_update_indexer) < datetime.timedelta(days=show_not_found_retry_days): return False myDB = db.DBConnection() sql_result = myDB.mass_action( [['SELECT airdate FROM [tv_episodes] WHERE showid = ? AND season > "0" ORDER BY season DESC, episode DESC LIMIT 1', [cur_indexerid]], ['SELECT airdate FROM [tv_episodes] WHERE showid = ? AND season > "0" AND airdate > "1" ORDER BY airdate DESC LIMIT 1', [cur_indexerid]]]) last_airdate_unknown = int(sql_result[0][0]['airdate']) <= 1 if sql_result and sql_result[0] else True last_airdate = datetime.date.fromordinal(sql_result[1][0]['airdate']) if sql_result and sql_result[1] else datetime.date.fromordinal(1) # if show is not 'Ended' and last episode aired less then 460 days ago or don't have an airdate for the last episode always update (status 'Continuing' or '') update_days_limit = 2013 ended_limit = datetime.timedelta(days=update_days_limit) if 'Ended' not in self.status and (last_airdate == datetime.date.fromordinal(1) or last_airdate_unknown or (update_date - last_airdate) <= ended_limit or (update_date - last_update_indexer) > ended_limit): return True # in the first 460 days (last airdate), update regularly airdate_diff = update_date - last_airdate last_update_diff = update_date - last_update_indexer update_step_list = [[60, 1], [120, 3], [180, 7], [1281, 15], [update_days_limit, 30]] for date_diff, interval in update_step_list: if airdate_diff <= datetime.timedelta(days=date_diff) and last_update_diff >= datetime.timedelta(days=interval): return True # update shows without an airdate for the last episode for update_days_limit days every 7 days if last_airdate_unknown and airdate_diff <= ended_limit and last_update_diff >= datetime.timedelta(days=7): return True else: return False def writeShowNFO(self): result = False if not ek.ek(os.path.isdir, self._location): logger.log('%s: Show directory doesn\'t exist, skipping NFO generation' % self.indexerid) return False logger.log('%s: Writing NFOs for show' % self.indexerid) for cur_provider in sickbeard.metadata_provider_dict.values(): result = cur_provider.create_show_metadata(self) or result return result def writeMetadata(self, show_only=False): if not ek.ek(os.path.isdir, self._location): logger.log('%s: Show directory doesn\'t exist, skipping NFO generation' % self.indexerid) return self.getImages() self.writeShowNFO() if not show_only: self.writeEpisodeNFOs() def writeEpisodeNFOs(self): if not ek.ek(os.path.isdir, self._location): logger.log('%s: Show directory doesn\'t exist, skipping NFO generation' % self.indexerid) return logger.log('%s: Writing NFOs for all episodes' % self.indexerid) myDB = db.DBConnection() sqlResults = myDB.select("SELECT * FROM tv_episodes WHERE showid = ? AND location != ''", [self.indexerid]) for epResult in sqlResults: logger.log('%s: Retrieving/creating episode %sx%s' % (self.indexerid, epResult["season"], epResult["episode"]), logger.DEBUG) curEp = self.getEpisode(epResult["season"], epResult["episode"]) curEp.createMetaFiles() def updateMetadata(self): if not ek.ek(os.path.isdir, self._location): logger.log('%s: Show directory doesn\'t exist, skipping NFO generation' % self.indexerid) return self.updateShowNFO() def updateShowNFO(self): result = False if not ek.ek(os.path.isdir, self._location): logger.log('%s: Show directory doesn\'t exist, skipping NFO generation' % self.indexerid) return False logger.log('%s: Updating NFOs for show with new indexer info' % self.indexerid) for cur_provider in sickbeard.metadata_provider_dict.values(): result = cur_provider.update_show_indexer_metadata(self) or result return result # find all media files in the show folder and create episodes for as many as possible def loadEpisodesFromDir(self): if not ek.ek(os.path.isdir, self._location): logger.log('%s: Show directory doesn\'t exist, not loading episodes from disk' % self.indexerid) return logger.log('%s: Loading all episodes from the show directory %s' % (self.indexerid, self._location)) # get file list mediaFiles = helpers.listMediaFiles(self._location) # create TVEpisodes from each media file (if possible) sql_l = [] for mediaFile in mediaFiles: parse_result = None curEpisode = None logger.log('%s: Creating episode from %s' % (self.indexerid, mediaFile), logger.DEBUG) try: curEpisode = self.makeEpFromFile(ek.ek(os.path.join, self._location, mediaFile)) except (exceptions.ShowNotFoundException, exceptions.EpisodeNotFoundException) as e: logger.log('Episode %s returned an exception: %s' % (mediaFile, ex(e)), logger.ERROR) continue except exceptions.EpisodeDeletedException: logger.log('The episode deleted itself when I tried making an object for it', logger.DEBUG) if curEpisode is None: continue # see if we should save the release name in the db ep_file_name = ek.ek(os.path.basename, curEpisode.location) ep_file_name = ek.ek(os.path.splitext, ep_file_name)[0] try: parse_result = None np = NameParser(False, showObj=self) parse_result = np.parse(ep_file_name) except (InvalidNameException, InvalidShowException): pass if ep_file_name and parse_result and None is not parse_result.release_group: logger.log( 'Name %s gave release group of %s, seems valid' % (ep_file_name ,parse_result.release_group), logger.DEBUG) curEpisode.release_name = ep_file_name # store the reference in the show if curEpisode != None: if self.subtitles: try: curEpisode.refreshSubtitles() except: logger.log('%s: Could not refresh subtitles' % self.indexerid, logger.ERROR) logger.log(traceback.format_exc(), logger.ERROR) result = curEpisode.get_sql() if None is not result: sql_l.append(result) if 0 < len(sql_l): myDB = db.DBConnection() myDB.mass_action(sql_l) def loadEpisodesFromDB(self, update=False): logger.log('Loading all episodes for [%s] from the DB' % self.name) myDB = db.DBConnection() sql = 'SELECT * FROM tv_episodes WHERE showid = ? AND indexer = ?' sqlResults = myDB.select(sql, [self.indexerid, self.indexer]) scannedEps = {} lINDEXER_API_PARMS = sickbeard.indexerApi(self.indexer).api_params.copy() if self.lang: lINDEXER_API_PARMS['language'] = self.lang if self.dvdorder != 0: lINDEXER_API_PARMS['dvdorder'] = True t = sickbeard.indexerApi(self.indexer).indexer(**lINDEXER_API_PARMS) cachedShow = t[self.indexerid] if None is cachedShow: return scannedEps cachedSeasons = {} for curResult in sqlResults: deleteEp = False curSeason = int(curResult["season"]) curEpisode = int(curResult["episode"]) if curSeason not in cachedSeasons: try: cachedSeasons[curSeason] = cachedShow[curSeason] except sickbeard.indexer_seasonnotfound as e: logger.log('Error when trying to load the episode for [%s] from %s: %s' % (self.name, sickbeard.indexerApi(self.indexer).name, e.message), logger.WARNING) deleteEp = True if not curSeason in scannedEps: scannedEps[curSeason] = {} logger.log('Loading episode %sx%s for [%s] from the DB' % (curSeason, curEpisode, self.name), logger.DEBUG) try: curEp = self.getEpisode(curSeason, curEpisode) # if we found out that the ep is no longer on TVDB then delete it from our database too if deleteEp and helpers.should_delete_episode(curEp.status): curEp.deleteEpisode() curEp.loadFromDB(curSeason, curEpisode) curEp.loadFromIndexer(tvapi=t, cachedSeason=cachedSeasons[curSeason], update=update) scannedEps[curSeason][curEpisode] = True except exceptions.EpisodeDeletedException: logger.log('Tried loading an episode from [%s] from the DB that should have been deleted, skipping it' % self.name, logger.DEBUG) continue return scannedEps def loadEpisodesFromIndexer(self, cache=True, update=False): lINDEXER_API_PARMS = sickbeard.indexerApi(self.indexer).api_params.copy() if not cache: lINDEXER_API_PARMS['cache'] = False if self.lang: lINDEXER_API_PARMS['language'] = self.lang if self.dvdorder != 0: lINDEXER_API_PARMS['dvdorder'] = True logger.log('%s: Loading all episodes for [%s] from %s..' % (self.indexerid, self.name, sickbeard.indexerApi(self.indexer).name)) try: t = sickbeard.indexerApi(self.indexer).indexer(**lINDEXER_API_PARMS) showObj = t[self.indexerid] except sickbeard.indexer_error: logger.log('%s timed out, unable to update episodes for [%s] from %s' % (sickbeard.indexerApi(self.indexer).name, self.name, sickbeard.indexerApi(self.indexer).name), logger.ERROR) return None scannedEps = {} sql_l = [] for season in showObj: scannedEps[season] = {} for episode in showObj[season]: # need some examples of wtf episode 0 means to decide if we want it or not if episode == 0: continue try: ep = self.getEpisode(season, episode) except exceptions.EpisodeNotFoundException: logger.log('%s: %s object for %sx%s from [%s] is incomplete, skipping this episode' % (self.indexerid, sickbeard.indexerApi(self.indexer).name, season, episode, self.name)) continue else: try: ep.loadFromIndexer(tvapi=t, update=update) except exceptions.EpisodeDeletedException: logger.log('The episode from [%s] was deleted, skipping the rest of the load' % self.name) continue with ep.lock: logger.log('%s: Loading info from %s for episode %sx%s from [%s]' % (self.indexerid, sickbeard.indexerApi(self.indexer).name, season, episode, self.name), logger.DEBUG) ep.loadFromIndexer(season, episode, tvapi=t, update=update) result = ep.get_sql() if None is not result: sql_l.append(result) scannedEps[season][episode] = True if 0 < len(sql_l): myDB = db.DBConnection() myDB.mass_action(sql_l) # Done updating save last update date self.last_update_indexer = datetime.date.today().toordinal() self.saveToDB() return scannedEps def getImages(self, fanart=None, poster=None): fanart_result = poster_result = banner_result = False season_posters_result = season_banners_result = season_all_poster_result = season_all_banner_result = False for cur_provider in sickbeard.metadata_provider_dict.values(): # FIXME: Needs to not show this message if the option is not enabled? logger.log('Running metadata routines for %s' % cur_provider.name, logger.DEBUG) fanart_result = cur_provider.create_fanart(self) or fanart_result poster_result = cur_provider.create_poster(self) or poster_result banner_result = cur_provider.create_banner(self) or banner_result season_posters_result = cur_provider.create_season_posters(self) or season_posters_result season_banners_result = cur_provider.create_season_banners(self) or season_banners_result season_all_poster_result = cur_provider.create_season_all_poster(self) or season_all_poster_result season_all_banner_result = cur_provider.create_season_all_banner(self) or season_all_banner_result return fanart_result or poster_result or banner_result or season_posters_result or season_banners_result or season_all_poster_result or season_all_banner_result # make a TVEpisode object from a media file def makeEpFromFile(self, file): if not ek.ek(os.path.isfile, file): logger.log('%s: Not a real file... %s' % (self.indexerid, file)) return None logger.log('%s: Creating episode object from %s' % (self.indexerid, file), logger.DEBUG) try: my_parser = NameParser(showObj=self) parse_result = my_parser.parse(file) except InvalidNameException: logger.log('Unable to parse the filename %s into a valid episode' % file, logger.DEBUG) return None except InvalidShowException: logger.log('Unable to parse the filename %s into a valid show' % file, logger.DEBUG) return None if not len(parse_result.episode_numbers): logger.log('parse_result: %s' % parse_result) logger.log('No episode number found in %s, ignoring it' % file, logger.ERROR) return None # for now lets assume that any episode in the show dir belongs to that show season = parse_result.season_number if None is not parse_result.season_number else 1 episodes = parse_result.episode_numbers root_ep = None sql_l = [] for cur_ep_num in episodes: episode = int(cur_ep_num) logger.log('%s: %s parsed to %s %sx%s' % (self.indexerid, file, self.name, season, episode), logger.DEBUG) check_quality_again = False same_file = False cur_ep = self.getEpisode(season, episode) if None is cur_ep: try: cur_ep = self.getEpisode(season, episode, file) except exceptions.EpisodeNotFoundException: logger.log('%s: Unable to figure out what this file is, skipping' % self.indexerid, logger.ERROR) continue else: # if there is a new file associated with this ep then re-check the quality status, quality = sickbeard.common.Quality.splitCompositeStatus(cur_ep.status) if IGNORED == status: continue if (cur_ep.location and ek.ek(os.path.normpath, cur_ep.location) != ek.ek(os.path.normpath, file)) or \ (not cur_ep.location and file) or \ (SKIPPED == status): logger.log('The old episode had a different file associated with it, re-checking the quality ' + 'based on the new filename %s' % file, logger.DEBUG) check_quality_again = True with cur_ep.lock: old_size = cur_ep.file_size if cur_ep.location and status != SKIPPED else 0 cur_ep.location = file # if the sizes are the same then it's probably the same file if old_size and cur_ep.file_size == old_size: same_file = True else: same_file = False cur_ep.checkForMetaFiles() if None is root_ep: root_ep = cur_ep else: if cur_ep not in root_ep.relatedEps: root_ep.relatedEps.append(cur_ep) # if it's a new file then if not same_file: cur_ep.release_name = '' # if they replace a file on me I'll make some attempt at re-checking the quality unless I know it's the same file if check_quality_again and not same_file: new_quality = Quality.nameQuality(file, self.is_anime) if Quality.UNKNOWN == new_quality: new_quality = Quality.fileQuality(file) logger.log('Since this file was renamed, file %s was checked and quality "%s" found' % (file, Quality.qualityStrings[new_quality]), logger.DEBUG) status, quality = sickbeard.common.Quality.splitCompositeStatus(cur_ep.status) if Quality.UNKNOWN != new_quality or SKIPPED == status: cur_ep.status = Quality.compositeStatus(DOWNLOADED, new_quality) # check for status/quality changes as long as it's a new file elif not same_file and sickbeard.helpers.has_media_ext(file)\ and cur_ep.status not in Quality.DOWNLOADED + Quality.ARCHIVED + [IGNORED]: old_status, old_quality = Quality.splitCompositeStatus(cur_ep.status) new_quality = Quality.nameQuality(file, self.is_anime) if Quality.UNKNOWN == new_quality: new_quality = Quality.fileQuality(file) if Quality.UNKNOWN == new_quality: new_quality = Quality.assumeQuality(file) new_status = None # if it was snatched and now exists then set the status correctly if SNATCHED == old_status and old_quality <= new_quality: logger.log('STATUS: this episode used to be snatched with quality %s but a file exists with quality %s so setting the status to DOWNLOADED' % (Quality.qualityStrings[old_quality], Quality.qualityStrings[new_quality]), logger.DEBUG) new_status = DOWNLOADED # if it was snatched proper and we found a higher quality one then allow the status change elif SNATCHED_PROPER == old_status and old_quality < new_quality: logger.log('STATUS: this episode used to be snatched proper with quality %s but a file exists with quality %s so setting the status to DOWNLOADED' % (Quality.qualityStrings[old_quality], Quality.qualityStrings[new_quality]), logger.DEBUG) new_status = DOWNLOADED elif old_status not in SNATCHED_ANY: new_status = DOWNLOADED if None is not new_status: with cur_ep.lock: logger.log('STATUS: we have an associated file, so setting the status from %s to DOWNLOADED/%s' % (cur_ep.status, Quality.compositeStatus(new_status, new_quality)), logger.DEBUG) cur_ep.status = Quality.compositeStatus(new_status, new_quality) with cur_ep.lock: result = cur_ep.get_sql() if None is not result: sql_l.append(result) if 0 < len(sql_l): my_db = db.DBConnection() my_db.mass_action(sql_l) # creating metafiles on the root should be good enough if sickbeard.USE_FAILED_DOWNLOADS and root_ep is not None: with root_ep.lock: root_ep.createMetaFiles() return root_ep def loadFromDB(self, skipNFO=False): myDB = db.DBConnection() sqlResults = myDB.select('SELECT * FROM tv_shows WHERE indexer_id = ?', [self.indexerid]) if len(sqlResults) > 1: logger.log('%s: Loading show info [%s] from database' % (self.indexerid, self.name)) raise exceptions.MultipleDBShowsException() elif len(sqlResults) == 0: logger.log('%s: Unable to find the show [%s] in the database' % (self.indexerid, self.name)) return else: if not self.indexer: self.indexer = int(sqlResults[0]['indexer']) if not self.name: self.name = sqlResults[0]['show_name'] if not self.network: self.network = sqlResults[0]['network'] if not self.genre: self.genre = sqlResults[0]['genre'] if self.classification is None: self.classification = sqlResults[0]['classification'] self.runtime = sqlResults[0]['runtime'] self.status = sqlResults[0]['status'] if not self.status: self.status = '' self.airs = sqlResults[0]['airs'] if not self.airs: self.airs = '' self.startyear = sqlResults[0]['startyear'] if not self.startyear: self.startyear = 0 self.air_by_date = sqlResults[0]['air_by_date'] if not self.air_by_date: self.air_by_date = 0 self.anime = sqlResults[0]['anime'] if None is self.anime: self.anime = 0 self.sports = sqlResults[0]['sports'] if not self.sports: self.sports = 0 self.scene = sqlResults[0]['scene'] if not self.scene: self.scene = 0 self.subtitles = sqlResults[0]['subtitles'] if self.subtitles: self.subtitles = 1 else: self.subtitles = 0 self.dvdorder = sqlResults[0]['dvdorder'] if not self.dvdorder: self.dvdorder = 0 self.upgrade_once = sqlResults[0]['archive_firstmatch'] if not self.upgrade_once: self.upgrade_once = 0 self.quality = int(sqlResults[0]['quality']) self.flatten_folders = int(sqlResults[0]['flatten_folders']) self.paused = int(sqlResults[0]['paused']) try: self.location = sqlResults[0]['location'] except Exception: dirty_setter('_location')(self, sqlResults[0]['location']) self._isDirGood = False if not self.lang: self.lang = sqlResults[0]['lang'] self.last_update_indexer = sqlResults[0]['last_update_indexer'] self.rls_ignore_words = sqlResults[0]['rls_ignore_words'] self.rls_require_words = sqlResults[0]['rls_require_words'] if not self.imdbid: imdbid = sqlResults[0]['imdb_id'] or '' self.imdbid = ('', imdbid)[2 < len(imdbid)] if self.is_anime: self.release_groups = BlackAndWhiteList(self.indexerid) if not self.overview: self.overview = sqlResults[0]['overview'] self.tag = sqlResults[0]['tag'] if not self.tag: self.tag = 'Show List' logger.log('Loaded.. {: <9} {: <8} {}'.format( sickbeard.indexerApi(self.indexer).config.get('name') + ',', str(self.indexerid) + ',', self.name)) # Get IMDb_info from database myDB = db.DBConnection() sqlResults = myDB.select('SELECT * FROM imdb_info WHERE indexer_id = ?', [self.indexerid]) if 0 < len(sqlResults): self.imdb_info = dict(zip(sqlResults[0].keys(), sqlResults[0])) elif sickbeard.USE_IMDB_INFO: logger.log('%s: The next show update will attempt to find IMDb info for [%s]' % (self.indexerid, self.name), logger.DEBUG) return self.dirty = False return True def loadFromIndexer(self, cache=True, tvapi=None, cachedSeason=None): logger.log('%s: Loading show info [%s] from %s' % (self.indexerid, self.name, sickbeard.indexerApi(self.indexer).name)) # There's gotta be a better way of doing this but we don't wanna # change the cache value elsewhere if tvapi is None: lINDEXER_API_PARMS = sickbeard.indexerApi(self.indexer).api_params.copy() if not cache: lINDEXER_API_PARMS['cache'] = False if self.lang: lINDEXER_API_PARMS['language'] = self.lang if self.dvdorder != 0: lINDEXER_API_PARMS['dvdorder'] = True t = sickbeard.indexerApi(self.indexer).indexer(**lINDEXER_API_PARMS) else: t = tvapi myEp = t[self.indexerid, False] if None is myEp: if hasattr(t, 'show_not_found') and t.show_not_found: self.inc_not_found_count() logger.log('Show [%s] not found (maybe even removed?)' % self.name, logger.WARNING) else: logger.log('Show data [%s] not found' % self.name, logger.WARNING) return False self.reset_not_found_count() try: self.name = myEp['seriesname'].strip() except AttributeError: raise sickbeard.indexer_attributenotfound( "Found %s, but attribute 'seriesname' was empty." % (self.indexerid)) self.classification = dict_prevent_None(myEp, 'classification', 'Scripted') self.genre = dict_prevent_None(myEp, 'genre', '') self.network = dict_prevent_None(myEp, 'network', '') self.runtime = dict_prevent_None(myEp, 'runtime', '') self.imdbid = dict_prevent_None(myEp, 'imdb_id', '') if getattr(myEp, 'airs_dayofweek', None) is not None and getattr(myEp, 'airs_time', None) is not None: self.airs = ('%s %s' % (myEp['airs_dayofweek'], myEp['airs_time'])).strip() if getattr(myEp, 'firstaired', None) is not None: self.startyear = int(str(myEp["firstaired"]).split('-')[0]) self.status = dict_prevent_None(myEp, 'status', '') self.overview = dict_prevent_None(myEp, 'overview', '') def load_imdb_info(self): if not sickbeard.USE_IMDB_INFO: return from lib.imdb import _exceptions as imdb_exceptions logger.log('Retrieving show info [%s] from IMDb' % self.name, logger.DEBUG) try: self._get_imdb_info() except imdb_exceptions.IMDbDataAccessError as e: logger.log('Timeout waiting for IMDb api: %s' % ex(e), logger.WARNING) except imdb_exceptions.IMDbError as e: logger.log('Something is wrong with IMDb api: %s' % ex(e), logger.WARNING) except Exception as e: logger.log('Error loading IMDb info: %s' % ex(e), logger.ERROR) logger.log('%s' % traceback.format_exc(), logger.ERROR) def _get_imdb_info(self): if not self.imdbid and self.ids.get(indexermapper.INDEXER_IMDB, {'id': 0}).get('id', 0) <= 0: return imdb_info = {'imdb_id': self.imdbid or 'tt%07d' % self.ids[indexermapper.INDEXER_IMDB]['id'], 'title': '', 'year': '', 'akas': [], 'runtimes': '', 'genres': [], 'countries': '', 'country_codes': [], 'certificates': [], 'rating': '', 'votes': '', 'last_update': ''} try: i = imdb.IMDb() imdbTv = i.get_movie( str(re.sub('[^0-9]', '', self.imdbid or '%07d' % self.ids[indexermapper.INDEXER_IMDB]['id']))) except IMDbError: return for key in filter(lambda x: x.replace('_', ' ') in imdbTv.keys(), imdb_info.keys()): # Store only the first value for string type if type(imdb_info[key]) == type('') and type(imdbTv.get(key)) == type([]): imdb_info[key] = imdbTv.get(key.replace('_', ' '))[0] else: imdb_info[key] = imdbTv.get(key.replace('_', ' ')) # Filter only the value if imdb_info['runtimes']: imdb_info['runtimes'] = re.search('\d+', imdb_info['runtimes']).group(0) else: imdb_info['runtimes'] = self.runtime if imdb_info['akas']: imdb_info['akas'] = '|'.join(imdb_info['akas']) else: imdb_info['akas'] = '' # Join all genres in a string if imdb_info['genres']: imdb_info['genres'] = '|'.join(imdb_info['genres']) else: imdb_info['genres'] = '' # Get only the production country certificate if any if imdb_info['certificates'] and imdb_info['countries']: dct = {} try: for item in imdb_info['certificates']: dct[item.split(':')[0]] = item.split(':')[1] imdb_info['certificates'] = dct[imdb_info['countries']] except: imdb_info['certificates'] = '' else: imdb_info['certificates'] = '' if imdb_info['country_codes']: imdb_info['country_codes'] = '|'.join(imdb_info['country_codes']) else: imdb_info['country_codes'] = '' imdb_info['last_update'] = datetime.date.today().toordinal() # Rename dict keys without spaces for DB upsert self.imdb_info = dict( (k.replace(' ', '_'), k(v) if hasattr(v, 'keys') else v) for k, v in imdb_info.items()) logger.log('%s: Obtained info from IMDb -> %s' % (self.indexerid, self.imdb_info), logger.DEBUG) logger.log('%s: Parsed latest IMDb show info for [%s]' % (self.indexerid, self.name)) def nextEpisode(self): logger.log('%s: Finding the episode which airs next for: %s' % (self.indexerid, self.name), logger.DEBUG) curDate = datetime.date.today().toordinal() if not self.nextaired or self.nextaired and curDate > self.nextaired: myDB = db.DBConnection() sqlResults = myDB.select( 'SELECT airdate, season, episode FROM tv_episodes WHERE showid = ? AND airdate >= ? AND status in (?,?,?) ORDER BY airdate ASC LIMIT 1', [self.indexerid, datetime.date.today().toordinal(), UNAIRED, WANTED, FAILED]) if sqlResults == None or len(sqlResults) == 0: logger.log('%s: No episode found... need to implement a show status' % self.indexerid, logger.DEBUG) self.nextaired = '' else: logger.log('%s: Found episode %sx%s' % (self.indexerid, sqlResults[0]['season'], sqlResults[0]['episode']), logger.DEBUG) self.nextaired = sqlResults[0]['airdate'] return self.nextaired def deleteShow(self, full=False): sql_l = [["DELETE FROM tv_episodes WHERE showid = ? AND indexer = ?", [self.indexerid, self.indexer]], ["DELETE FROM tv_shows WHERE indexer_id = ? AND indexer = ?", [self.indexerid, self.indexer]], ["DELETE FROM imdb_info WHERE indexer_id = ?", [self.indexerid]], ["DELETE FROM xem_refresh WHERE indexer_id = ? AND indexer = ?", [self.indexerid, self.indexer]], ["DELETE FROM scene_numbering WHERE indexer_id = ? AND indexer = ?", [self.indexerid, self.indexer]], ["DELETE FROM whitelist WHERE show_id = ?", [self.indexerid]], ["DELETE FROM blacklist WHERE show_id = ?", [self.indexerid]], ["DELETE FROM indexer_mapping WHERE indexer_id = ? AND indexer = ?", [self.indexerid, self.indexer]], ["DELETE FROM tv_shows_not_found WHERE indexer = ? AND indexer_id = ?", [self.indexer, self.indexerid]]] myDB = db.DBConnection() myDB.mass_action(sql_l) name_cache.remove_from_namecache(self.indexerid) action = ('delete', 'trash')[sickbeard.TRASH_REMOVE_SHOW] # remove self from show list sickbeard.showList = [x for x in sickbeard.showList if int(x.indexerid) != self.indexerid] # clear the cache ic = image_cache.ImageCache() for cache_obj in ek.ek(glob.glob, ic.poster_path(self.indexerid).replace('poster.jpg', '*')) \ + ek.ek(glob.glob, ic.poster_thumb_path(self.indexerid).replace('poster.jpg', '*')) \ + ek.ek(glob.glob, ic.fanart_path(self.indexerid).replace('%s.fanart.jpg' % self.indexerid, '')): cache_dir = ek.ek(os.path.isdir, cache_obj) logger.log('Attempt to %s cache %s %s' % (action, cache_dir and 'dir' or 'file', cache_obj)) try: if sickbeard.TRASH_REMOVE_SHOW: ek.ek(send2trash, cache_obj) elif cache_dir: ek.ek(shutil.rmtree, cache_obj) else: ek.ek(os.remove, cache_obj) except OSError as e: logger.log('Unable to %s %s: %s / %s' % (action, cache_obj, repr(e), str(e)), logger.WARNING) show_id = '%s' % self.indexerid if show_id in sickbeard.FANART_RATINGS: del sickbeard.FANART_RATINGS[show_id] # remove entire show folder if full: try: logger.log('Attempt to %s show folder %s' % (action, self._location)) # check first the read-only attribute file_attribute = ek.ek(os.stat, self.location)[0] if not file_attribute & stat.S_IWRITE: # File is read-only, so make it writeable logger.log('Attempting to make writeable the read only folder %s' % self._location, logger.DEBUG) try: ek.ek(os.chmod, self.location, stat.S_IWRITE) except: logger.log('Unable to change permissions of %s' % self._location, logger.WARNING) if sickbeard.TRASH_REMOVE_SHOW: ek.ek(send2trash, self.location) else: ek.ek(shutil.rmtree, self.location) logger.log('%s show folder %s' % (('Deleted', 'Trashed')[sickbeard.TRASH_REMOVE_SHOW], self._location)) except exceptions.ShowDirNotFoundException: logger.log('Show folder does not exist, no need to %s %s' % (action, self._location), logger.WARNING) except OSError as e: logger.log('Unable to %s %s: %s / %s' % (action, self._location, repr(e), str(e)), logger.WARNING) def populateCache(self, force=False): cache_inst = image_cache.ImageCache() logger.log('Checking & filling cache for show %s' % self.name) cache_inst.fill_cache(self, force) def refreshDir(self): # make sure the show dir is where we think it is unless dirs are created on the fly if not ek.ek(os.path.isdir, self._location) and not sickbeard.CREATE_MISSING_SHOW_DIRS: return False # load from dir self.loadEpisodesFromDir() # run through all locations from DB, check that they exist logger.log('%s: Loading all episodes for [%s] with a location from the database' % (self.indexerid, self.name)) myDB = db.DBConnection() sqlResults = myDB.select("SELECT * FROM tv_episodes WHERE showid = ? AND location != ''", [self.indexerid]) sql_l = [] for ep in sqlResults: curLoc = ek.ek(os.path.normpath, ep['location']) season = int(ep['season']) episode = int(ep['episode']) try: curEp = self.getEpisode(season, episode) except exceptions.EpisodeDeletedException: logger.log('The episode from [%s] was deleted while we were refreshing it, moving on to the next one' % self.name, logger.DEBUG) continue # if the path doesn't exist or if it's not in our show dir if not ek.ek(os.path.isfile, curLoc) or not ek.ek(os.path.normpath, curLoc).startswith( ek.ek(os.path.normpath, self.location)): # check if downloaded files still exist, update our data if this has changed if 1 != sickbeard.SKIP_REMOVED_FILES: with curEp.lock: # if it used to have a file associated with it and it doesn't anymore then set it to IGNORED if curEp.location and curEp.status in Quality.DOWNLOADED: if ARCHIVED == sickbeard.SKIP_REMOVED_FILES: curEp.status = Quality.compositeStatus( ARCHIVED, Quality.qualityDownloaded(curEp.status)) else: curEp.status = (sickbeard.SKIP_REMOVED_FILES, IGNORED)[not sickbeard.SKIP_REMOVED_FILES] logger.log('%s: File no longer at location for s%02de%02d, episode removed and status changed to %s' % (str(self.indexerid), season, episode, statusStrings[curEp.status]), logger.DEBUG) curEp.subtitles = list() curEp.subtitles_searchcount = 0 curEp.subtitles_lastsearch = str(datetime.datetime.min) curEp.location = '' curEp.hasnfo = False curEp.hastbn = False curEp.release_name = '' result = curEp.get_sql() if None is not result: sql_l.append(result) else: # the file exists, set its modify file stamp if sickbeard.AIRDATE_EPISODES: curEp.airdateModifyStamp() if 0 < len(sql_l): myDB = db.DBConnection() myDB.mass_action(sql_l) def downloadSubtitles(self, force=False): # TODO: Add support for force option if not ek.ek(os.path.isdir, self._location): logger.log('%s: Show directory doesn\'t exist, can\'t download subtitles' % self.indexerid, logger.DEBUG) return logger.log('%s: Downloading subtitles' % self.indexerid, logger.DEBUG) try: myDB = db.DBConnection() episodes = myDB.select( "SELECT location FROM tv_episodes WHERE showid = ? AND location NOT LIKE '' ORDER BY season DESC, episode DESC", [self.indexerid]) for episodeLoc in episodes: episode = self.makeEpFromFile(episodeLoc['location']) subtitles = episode.downloadSubtitles(force=force) except Exception as e: logger.log('Error occurred when downloading subtitles: %s' % traceback.format_exc(), logger.ERROR) return def switchIndexer(self, old_indexer, old_indexerid, pausestatus_after=None): myDB = db.DBConnection() myDB.mass_action([['UPDATE tv_shows SET indexer = ?, indexer_id = ? WHERE indexer = ? AND indexer_id = ?', [self.indexer, self.indexerid, old_indexer, old_indexerid]], ['UPDATE tv_episodes SET showid = ?, indexer = ?, indexerid = 0 WHERE indexer = ? AND showid = ?', [self.indexerid, self.indexer, old_indexer, old_indexerid]], ['UPDATE blacklist SET show_id = ? WHERE show_id = ?', [self.indexerid, old_indexerid]], ['UPDATE history SET showid = ? WHERE showid = ?', [self.indexerid, old_indexerid]], ['UPDATE imdb_info SET indexer_id = ? WHERE indexer_id = ?', [self.indexerid, old_indexerid]], ['UPDATE scene_exceptions SET indexer_id = ? WHERE indexer_id = ?', [self.indexerid, old_indexerid]], ['UPDATE scene_numbering SET indexer = ?, indexer_id = ? WHERE indexer = ? AND indexer_id = ?', [self.indexer, self.indexerid, old_indexer, old_indexerid]], ['UPDATE whitelist SET show_id = ? WHERE show_id = ?', [self.indexerid, old_indexerid]], ['UPDATE xem_refresh SET indexer = ?, indexer_id = ? WHERE indexer = ? AND indexer_id = ?', [self.indexer, self.indexerid, old_indexer, old_indexerid]], ['DELETE FROM tv_shows_not_found WHERE indexer = ? AND indexer_id = ?', [old_indexer, old_indexerid]]]) myFailedDB = db.DBConnection('failed.db') myFailedDB.action('UPDATE history SET showid = ? WHERE showid = ?', [self.indexerid, old_indexerid]) del_mapping(old_indexer, old_indexerid) self.ids[old_indexer]['status'] = MapStatus.NONE self.ids[self.indexer]['status'] = MapStatus.SOURCE self.ids[self.indexer]['id'] = self.indexerid save_mapping(self) name_cache.remove_from_namecache(old_indexerid) image_cache_dir = ek.ek(os.path.join, sickbeard.CACHE_DIR, 'images') for path, dirs, files in ek.ek(os.walk, image_cache_dir): for filename in ek.ek(fnmatch.filter, files, '%s.*' % old_indexerid): cache_file = ek.ek(os.path.join, path, filename) new_cachefile = ek.ek(os.path.join, path, filename.replace(str(old_indexerid), str(self.indexerid))) try: helpers.moveFile(cache_file, new_cachefile) except Exception as e: logger.log('Unable to rename %s to %s: %s / %s' % (cache_file, new_cachefile, repr(e), str(e)), logger.WARNING) ic = image_cache.ImageCache() cache_dir = ic.fanart_path(old_indexerid).replace('%s.fanart.jpg' % old_indexerid, '').rstrip('\/') new_cache_dir = ic.fanart_path(self.indexerid).replace('%s.fanart.jpg' % self.indexerid, '').rstrip('\/') try: helpers.moveFile(cache_dir, new_cache_dir) except Exception as e: logger.log('Unable to rename %s to %s: %s / %s' % (cache_dir, new_cache_dir, repr(e), str(e)), logger.WARNING) rating = sickbeard.FANART_RATINGS.get('%s' % old_indexerid) if rating: del sickbeard.FANART_RATINGS['%s' % old_indexerid] sickbeard.FANART_RATINGS['%s' % self.indexerid] = rating sickbeard.save_config() name_cache.buildNameCache(self) self.reset_not_found_count() # force the update try: sickbeard.showQueueScheduler.action.updateShow( self, force=True, web=True, priority=QueuePriorities.VERYHIGH, pausestatus_after=pausestatus_after) except exceptions.CantUpdateException as e: logger.log('Unable to update this show. %s' % ex(e), logger.ERROR) def saveToDB(self, forceSave=False): if not self.dirty and not forceSave: logger.log('%s: Not saving show to db - record is not dirty' % self.indexerid, logger.DEBUG) return logger.log('%s: Saving show info to database' % self.indexerid, logger.DEBUG) controlValueDict = {'indexer_id': self.indexerid} newValueDict = {'indexer': self.indexer, 'show_name': self.name, 'location': self._location, 'network': self.network, 'genre': self.genre, 'classification': self.classification, 'runtime': self.runtime, 'quality': self.quality, 'airs': self.airs, 'status': self.status, 'flatten_folders': self.flatten_folders, 'paused': self.paused, 'air_by_date': self.air_by_date, 'anime': self.anime, 'scene': self.scene, 'sports': self.sports, 'subtitles': self.subtitles, 'dvdorder': self.dvdorder, 'archive_firstmatch': self.upgrade_once, 'startyear': self.startyear, 'lang': self.lang, 'imdb_id': self.imdbid, 'last_update_indexer': self.last_update_indexer, 'rls_ignore_words': self.rls_ignore_words, 'rls_require_words': self.rls_require_words, 'overview': self.overview, 'tag': self.tag, } myDB = db.DBConnection() myDB.upsert('tv_shows', newValueDict, controlValueDict) self.dirty = False if sickbeard.USE_IMDB_INFO and len(self.imdb_info): controlValueDict = {'indexer_id': self.indexerid} newValueDict = self.imdb_info myDB = db.DBConnection() myDB.upsert('imdb_info', newValueDict, controlValueDict) def __str__(self): return 'indexerid: %s\n' % self.indexerid \ + 'indexer: %s\n' % self.indexerid \ + 'name: %s\n' % self.name \ + 'location: %s\n' % self._location \ + ('', 'network: %s\n' % self.network)[self.network] \ + ('', 'airs: %s\n' % self.airs)[self.airs] \ + ('', 'status: %s\n' % self.status)[self.status] \ + 'startyear: %s\n' % self.startyear \ + ('', 'genre: %s\n' % self.genre)[self.genre] \ + 'classification: %s\n' % self.classification \ + 'runtime: %s\n' % self.runtime \ + 'quality: %s\n' % self.quality \ + 'scene: %s\n' % self.is_scene \ + 'sports: %s\n' % self.is_sports \ + 'anime: %s\n' % self.is_anime def wantEpisode(self, season, episode, quality, manualSearch=False, multi_ep=False): logger.log('Checking if found %sepisode %sx%s is wanted at quality %s' % (('', 'multi-part ')[multi_ep], season, episode, Quality.qualityStrings[quality]), logger.DEBUG) if not multi_ep: try: wq = getattr(self.episodes.get(season, {}).get(episode, {}), 'wantedQuality', None) if None is not wq: if quality in wq: curStatus, curQuality = Quality.splitCompositeStatus(self.episodes[season][episode].status) if curStatus in (WANTED, UNAIRED, SKIPPED, FAILED): logger.log('Existing episode status is wanted/unaired/skipped/failed, getting found episode', logger.DEBUG) return True elif manualSearch: logger.log('Usually ignoring found episode, but forced search allows the quality, getting found' ' episode', logger.DEBUG) return True elif quality > curQuality: logger.log( 'Episode already exists but the found episode has better quality, getting found episode', logger.DEBUG) return True logger.log('None of the conditions were met, ignoring found episode', logger.DEBUG) return False except (StandardError, Exception): pass # if the quality isn't one we want under any circumstances then just say no initialQualities, archiveQualities = Quality.splitQuality(self.quality) allQualities = list(set(initialQualities + archiveQualities)) initial = '= (%s)' % ','.join([Quality.qualityStrings[qual] for qual in initialQualities]) if 0 < len(archiveQualities): initial = '+ upgrade to %s + (%s)'\ % (initial, ','.join([Quality.qualityStrings[qual] for qual in archiveQualities])) logger.log('Want initial %s and found %s' % (initial, Quality.qualityStrings[quality]), logger.DEBUG) if quality not in allQualities: logger.log('Don\'t want this quality, ignoring found episode', logger.DEBUG) return False myDB = db.DBConnection() sqlResults = myDB.select('SELECT status FROM tv_episodes WHERE showid = ? AND season = ? AND episode = ?', [self.indexerid, season, episode]) if not sqlResults or not len(sqlResults): logger.log('Unable to find a matching episode in database, ignoring found episode', logger.DEBUG) return False curStatus, curQuality = Quality.splitCompositeStatus(int(sqlResults[0]['status'])) epStatus_text = statusStrings[curStatus] logger.log('Existing episode status: %s (%s)' % (statusStrings[curStatus], epStatus_text), logger.DEBUG) # if we know we don't want it then just say no if curStatus in [IGNORED, ARCHIVED] + ([SKIPPED], [])[multi_ep] and not manualSearch: logger.log('Existing episode status is %signored/archived, ignoring found episode' % ('skipped/', '')[multi_ep], logger.DEBUG) return False # if it's one of these then we want it as long as it's in our allowed initial qualities if quality in allQualities: if curStatus in [WANTED, UNAIRED, SKIPPED, FAILED] + ([], SNATCHED_ANY)[multi_ep]: logger.log('Existing episode status is wanted/unaired/skipped/failed, getting found episode', logger.DEBUG) return True elif manualSearch: logger.log( 'Usually ignoring found episode, but forced search allows the quality, getting found episode', logger.DEBUG) return True else: logger.log('Quality is on wanted list, need to check if it\'s better than existing quality', logger.DEBUG) downloadedStatusList = SNATCHED_ANY + [DOWNLOADED] # special case: already downloaded quality is not in any of the wanted Qualities if curStatus in downloadedStatusList and curQuality not in allQualities: wantedQualities = allQualities else: wantedQualities = archiveQualities # if we are re-downloading then we only want it if it's in our archiveQualities list and better than what we have if curStatus in downloadedStatusList and quality in wantedQualities and quality > curQuality: logger.log('Episode already exists but the found episode has better quality, getting found episode', logger.DEBUG) return True else: logger.log('Episode already exists and the found episode has same/lower quality, ignoring found episode', logger.DEBUG) logger.log('None of the conditions were met, ignoring found episode', logger.DEBUG) return False def getOverview(self, epStatus): status, quality = Quality.splitCompositeStatus(epStatus) if ARCHIVED == status: return Overview.GOOD if WANTED == status: return Overview.WANTED if status in (SKIPPED, IGNORED): return Overview.SKIPPED if status in (UNAIRED, UNKNOWN): return Overview.UNAIRED if status in [SUBTITLED] + Quality.SNATCHED_ANY + Quality.DOWNLOADED + Quality.FAILED: if FAILED == status: return Overview.WANTED if status in SNATCHED_ANY: return Overview.SNATCHED void, best_qualities = Quality.splitQuality(self.quality) # if re-downloads aren't wanted then mark it "good" if there is anything if not len(best_qualities): return Overview.GOOD min_best, max_best = min(best_qualities), max(best_qualities) if quality >= max_best \ or (self.upgrade_once and (quality in best_qualities or (None is not min_best and quality > min_best))): return Overview.GOOD return Overview.QUAL def __getstate__(self): d = dict(self.__dict__) del d['lock'] return d def __setstate__(self, d): d['lock'] = threading.Lock() self.__dict__.update(d) class TVEpisode(object): def __init__(self, show, season, episode, file='', show_sql=None): self._name = '' self._season = season self._episode = episode self._absolute_number = 0 self._description = '' self._subtitles = list() self._subtitles_searchcount = 0 self._subtitles_lastsearch = str(datetime.datetime.min) self._airdate = datetime.date.fromordinal(1) self._hasnfo = False self._hastbn = False self._status = UNKNOWN self._indexerid = 0 self._file_size = 0 self._release_name = '' self._is_proper = False self._version = 0 self._release_group = '' # setting any of the above sets the dirty flag self.dirty = True self.show = show self.scene_season = 0 self.scene_episode = 0 self.scene_absolute_number = 0 self._location = file self._indexer = int(self.show.indexer) self.lock = threading.Lock() self.specifyEpisode(self.season, self.episode, show_sql) self.relatedEps = [] self.checkForMetaFiles() self.wantedQuality = [] name = property(lambda self: self._name, dirty_setter('_name', basestring)) season = property(lambda self: self._season, dirty_setter('_season')) episode = property(lambda self: self._episode, dirty_setter('_episode')) absolute_number = property(lambda self: self._absolute_number, dirty_setter('_absolute_number')) description = property(lambda self: self._description, dirty_setter('_description')) subtitles = property(lambda self: self._subtitles, dirty_setter('_subtitles')) subtitles_searchcount = property(lambda self: self._subtitles_searchcount, dirty_setter('_subtitles_searchcount')) subtitles_lastsearch = property(lambda self: self._subtitles_lastsearch, dirty_setter('_subtitles_lastsearch')) airdate = property(lambda self: self._airdate, dirty_setter('_airdate')) hasnfo = property(lambda self: self._hasnfo, dirty_setter('_hasnfo')) hastbn = property(lambda self: self._hastbn, dirty_setter('_hastbn')) status = property(lambda self: self._status, dirty_setter('_status')) indexer = property(lambda self: self._indexer, dirty_setter('_indexer')) indexerid = property(lambda self: self._indexerid, dirty_setter('_indexerid')) # location = property(lambda self: self._location, dirty_setter('_location')) file_size = property(lambda self: self._file_size, dirty_setter('_file_size')) release_name = property(lambda self: self._release_name, dirty_setter('_release_name')) is_proper = property(lambda self: self._is_proper, dirty_setter('_is_proper')) version = property(lambda self: self._version, dirty_setter('_version')) release_group = property(lambda self: self._release_group, dirty_setter('_release_group')) def _set_location(self, new_location): log_vals = (('clears', ''), ('sets', ' to ' + new_location))[any(new_location)] logger.log(u'Setter %s location%s' % log_vals, logger.DEBUG) # self._location = newLocation dirty_setter('_location')(self, new_location) if new_location and ek.ek(os.path.isfile, new_location): self.file_size = ek.ek(os.path.getsize, new_location) else: self.file_size = 0 location = property(lambda self: self._location, _set_location) def refreshSubtitles(self): """Look for subtitles files and refresh the subtitles property""" self.subtitles = subtitles.subtitlesLanguages(self.location) def downloadSubtitles(self, force=False): # TODO: Add support for force option if not ek.ek(os.path.isfile, self.location): logger.log('%s: Episode file doesn\'t exist, can\'t download subtitles for episode %sx%s' % (self.show.indexerid, self.season, self.episode), logger.DEBUG) return logger.log('%s: Downloading subtitles for episode %sx%s' % (self.show.indexerid, self.season, self.episode), logger.DEBUG) previous_subtitles = self.subtitles try: need_languages = set(sickbeard.SUBTITLES_LANGUAGES) - set(self.subtitles) subtitles = subliminal.download_subtitles([self.location], languages=need_languages, services=sickbeard.subtitles.getEnabledServiceList(), force=force, multi=True, cache_dir=sickbeard.CACHE_DIR) if sickbeard.SUBTITLES_DIR: for video in subtitles: subs_new_path = ek.ek(os.path.join, ek.ek(os.path.dirname, video.path), sickbeard.SUBTITLES_DIR) dir_exists = helpers.makeDir(subs_new_path) if not dir_exists: logger.log('Unable to create subtitles folder %s' % subs_new_path, logger.ERROR) else: helpers.chmodAsParent(subs_new_path) for subtitle in subtitles.get(video): new_file_path = ek.ek(os.path.join, subs_new_path, ek.ek(os.path.basename, subtitle.path)) helpers.moveFile(subtitle.path, new_file_path) helpers.chmodAsParent(new_file_path) else: for video in subtitles: for subtitle in subtitles.get(video): helpers.chmodAsParent(subtitle.path) except Exception as e: logger.log('Error occurred when downloading subtitles: %s' % traceback.format_exc(), logger.ERROR) return self.refreshSubtitles() self.subtitles_searchcount = self.subtitles_searchcount + 1 if self.subtitles_searchcount else 1 # added the if because sometime it raise an error self.subtitles_lastsearch = datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S') self.saveToDB() newsubtitles = set(self.subtitles).difference(set(previous_subtitles)) if newsubtitles: try: subtitleList = ", ".join(subliminal.language.Language(x).name for x in newsubtitles) except(StandardError, Exception): logger.log('Could not parse a language to use to fetch subtitles for episode %sx%s' % (self.season, self.episode), logger.DEBUG) return logger.log('%s: Downloaded %s subtitles for episode %sx%s' % (self.show.indexerid, subtitleList, self.season, self.episode), logger.DEBUG) notifiers.notify_subtitle_download(self.prettyName(), subtitleList) else: logger.log('%s: No subtitles downloaded for episode %sx%s' % (self.show.indexerid, self.season, self.episode), logger.DEBUG) if sickbeard.SUBTITLES_HISTORY: for video in subtitles: for subtitle in subtitles.get(video): history.log_subtitle(self.show.indexerid, self.season, self.episode, self.status, subtitle) return subtitles def checkForMetaFiles(self): oldhasnfo = self.hasnfo oldhastbn = self.hastbn cur_nfo = False cur_tbn = False # check for nfo and tbn if ek.ek(os.path.isfile, self.location): for cur_provider in sickbeard.metadata_provider_dict.values(): if cur_provider.episode_metadata: new_result = cur_provider._has_episode_metadata(self) else: new_result = False cur_nfo = new_result or cur_nfo if cur_provider.episode_thumbnails: new_result = cur_provider._has_episode_thumb(self) else: new_result = False cur_tbn = new_result or cur_tbn self.hasnfo = cur_nfo self.hastbn = cur_tbn # if either setting has changed return true, if not return false return oldhasnfo != self.hasnfo or oldhastbn != self.hastbn def specifyEpisode(self, season, episode, show_sql=None): sqlResult = self.loadFromDB(season, episode, show_sql) if not sqlResult: # only load from NFO if we didn't load from DB if ek.ek(os.path.isfile, self.location): try: self.loadFromNFO(self.location) except exceptions.NoNFOException: logger.log('%s: There was an error loading the NFO for episode %sx%s' % (self.show.indexerid, season, episode), logger.ERROR) pass # if we tried loading it from NFO and didn't find the NFO, try the Indexers if not self.hasnfo: try: result = self.loadFromIndexer(season, episode) except exceptions.EpisodeDeletedException: result = False # if we failed SQL *and* NFO, Indexers then fail if not result: raise exceptions.EpisodeNotFoundException( 'Couldn\'t find episode %sx%s' % (season, episode)) def loadFromDB(self, season, episode, show_sql=None): logger.log('%s: Loading episode details from DB for episode %sx%s' % (self.show.indexerid, season, episode), logger.DEBUG) sql_results = None if show_sql: sql_results = [s for s in show_sql if episode == s['episode'] and season == s['season']] if not sql_results: myDB = db.DBConnection() sql_results = myDB.select('SELECT * FROM tv_episodes WHERE showid = ? AND season = ? AND episode = ?', [self.show.indexerid, season, episode]) if len(sql_results) > 1: raise exceptions.MultipleDBEpisodesException('Your DB has two records for the same show somehow.') elif len(sql_results) == 0: logger.log('%s: Episode %sx%s not found in the database' % (self.show.indexerid, self.season, self.episode), logger.DEBUG) return False else: # NAMEIT logger.log(u'AAAAA from' + str(self.season)+'x'+str(self.episode) + ' -' + self.name + ' to ' + str(sql_results[0]['name'])) if sql_results[0]['name']: self.name = sql_results[0]['name'] self.season = season self.episode = episode self.absolute_number = sql_results[0]['absolute_number'] self.description = sql_results[0]['description'] if not self.description: self.description = '' if sql_results[0]['subtitles'] and sql_results[0]['subtitles']: self.subtitles = sql_results[0]['subtitles'].split(',') self.subtitles_searchcount = sql_results[0]['subtitles_searchcount'] self.subtitles_lastsearch = sql_results[0]['subtitles_lastsearch'] self.airdate = datetime.date.fromordinal(int(sql_results[0]['airdate'])) # logger.log(u'1 Status changes from ' + str(self.status) + ' to ' + str(sql_results[0]['status']), logger.DEBUG) if sql_results[0]['status'] is not None: self.status = int(sql_results[0]['status']) # don't overwrite my location if sql_results[0]['location'] and sql_results[0]['location']: self.location = ek.ek(os.path.normpath, sql_results[0]['location']) if sql_results[0]['file_size']: self.file_size = int(sql_results[0]['file_size']) else: self.file_size = 0 self.indexerid = int(sql_results[0]['indexerid']) self.indexer = int(sql_results[0]['indexer']) sickbeard.scene_numbering.xem_refresh(self.show.indexerid, self.show.indexer) try: self.scene_season = int(sql_results[0]['scene_season']) except: self.scene_season = 0 try: self.scene_episode = int(sql_results[0]['scene_episode']) except: self.scene_episode = 0 try: self.scene_absolute_number = int(sql_results[0]['scene_absolute_number']) except: self.scene_absolute_number = 0 if self.scene_absolute_number == 0: self.scene_absolute_number = sickbeard.scene_numbering.get_scene_absolute_numbering( self.show.indexerid, self.show.indexer, self.absolute_number ) if self.scene_season == 0 or self.scene_episode == 0: self.scene_season, self.scene_episode = sickbeard.scene_numbering.get_scene_numbering( self.show.indexerid, self.show.indexer, self.season, self.episode ) if sql_results[0]['release_name'] is not None: self.release_name = sql_results[0]['release_name'] if sql_results[0]['is_proper']: self.is_proper = int(sql_results[0]['is_proper']) if sql_results[0]['version']: self.version = int(sql_results[0]['version']) if sql_results[0]['release_group'] is not None: self.release_group = sql_results[0]['release_group'] self.dirty = False return True def loadFromIndexer(self, season=None, episode=None, cache=True, tvapi=None, cachedSeason=None, update=False): if None is season: season = self.season if None is episode: episode = self.episode logger.log('%s: Loading episode details from %s for episode %sx%s' % (self.show.indexerid, sickbeard.indexerApi(self.show.indexer).name, season, episode), logger.DEBUG) indexer_lang = self.show.lang try: if None is cachedSeason: if None is tvapi: lINDEXER_API_PARMS = sickbeard.indexerApi(self.indexer).api_params.copy() if not cache: lINDEXER_API_PARMS['cache'] = False if indexer_lang: lINDEXER_API_PARMS['language'] = indexer_lang if 0 != self.show.dvdorder: lINDEXER_API_PARMS['dvdorder'] = True t = sickbeard.indexerApi(self.indexer).indexer(**lINDEXER_API_PARMS) else: t = tvapi myEp = t[self.show.indexerid][season][episode] else: myEp = cachedSeason[episode] except (sickbeard.indexer_error, IOError) as e: logger.log('%s threw up an error: %s' % (sickbeard.indexerApi(self.indexer).name, ex(e)), logger.DEBUG) # if the episode is already valid just log it, if not throw it up if self.name: logger.log('%s timed out but we have enough info from other sources, allowing the error' % sickbeard.indexerApi(self.indexer).name, logger.DEBUG) return else: logger.log('%s timed out, unable to create the episode' % sickbeard.indexerApi(self.indexer).name, logger.ERROR) return False except (sickbeard.indexer_episodenotfound, sickbeard.indexer_seasonnotfound): logger.log('Unable to find the episode on %s... has it been removed? Should I delete from db?' % sickbeard.indexerApi(self.indexer).name, logger.DEBUG) # if I'm no longer on the Indexers but I once was then delete myself from the DB if -1 != self.indexerid and helpers.should_delete_episode(self.status): self.deleteEpisode() return if getattr(myEp, 'absolute_number', None) in (None, ''): logger.log('This episode (%s - %sx%s) has no absolute number on %s' % (self.show.name, season, episode, sickbeard.indexerApi(self.indexer).name), logger.DEBUG) else: logger.log('%s: The absolute_number for %sx%s is : %s' % (self.show.indexerid, season, episode, myEp['absolute_number']), logger.DEBUG) self.absolute_number = int(myEp['absolute_number']) self.name = dict_prevent_None(myEp, 'episodename', '') self.season = season self.episode = episode sickbeard.scene_numbering.xem_refresh(self.show.indexerid, self.show.indexer) self.scene_absolute_number = sickbeard.scene_numbering.get_scene_absolute_numbering( self.show.indexerid, self.show.indexer, self.absolute_number ) self.scene_season, self.scene_episode = sickbeard.scene_numbering.get_scene_numbering( self.show.indexerid, self.show.indexer, self.season, self.episode ) self.description = dict_prevent_None(myEp, 'overview', '') firstaired = getattr(myEp, 'firstaired', None) if None is firstaired or firstaired in '0000-00-00': firstaired = str(datetime.date.fromordinal(1)) rawAirdate = [int(x) for x in firstaired.split('-')] old_airdate_future = self.airdate == datetime.date.fromordinal(1) or self.airdate >= datetime.date.today() try: self.airdate = datetime.date(rawAirdate[0], rawAirdate[1], rawAirdate[2]) except (ValueError, IndexError): logger.log('Malformed air date retrieved from %s (%s - %sx%s)' % (sickbeard.indexerApi(self.indexer).name, self.show.name, season, episode), logger.ERROR) # if I'm incomplete on TVDB but I once was complete then just delete myself from the DB for now if -1 != self.indexerid and helpers.should_delete_episode(self.status): self.deleteEpisode() return False # early conversion to int so that episode doesn't get marked dirty self.indexerid = getattr(myEp, 'id', None) if None is self.indexerid: logger.log('Failed to retrieve ID from %s' % sickbeard.indexerApi(self.indexer).name, logger.ERROR) if helpers.should_delete_episode(self.status): self.deleteEpisode() return False # don't update show status if show dir is missing, unless it's missing on purpose if not ek.ek(os.path.isdir, self.show._location) and not sickbeard.CREATE_MISSING_SHOW_DIRS and not sickbeard.ADD_SHOWS_WO_DIR: logger.log( 'The show directory is missing, not bothering to change the episode statuses since it\'d probably be invalid') return if self.location: logger.log('%s: Setting status for %sx%s based on status %s and existence of %s' % (self.show.indexerid, season, episode, statusStrings[self.status], self.location), logger.DEBUG) # if we don't have the file if not ek.ek(os.path.isfile, self.location): if self.status in [SKIPPED, UNAIRED, UNKNOWN, WANTED]: today = datetime.date.today() delta = datetime.timedelta(days=1) show_time = network_timezones.parse_date_time(self.airdate.toordinal(), self.show.airs, self.show.network) show_length = datetime.timedelta(minutes=helpers.tryInt(self.show.runtime, 60)) tz_now = datetime.datetime.now(network_timezones.sb_timezone) future_airtime = (self.airdate > (today + delta) or (not self.airdate < (today - delta) and ((show_time + show_length) > tz_now))) # if this episode hasn't aired yet set the status to UNAIRED if future_airtime: msg = 'Episode airs in the future, marking it %s' self.status = UNAIRED # if there's no airdate then set it to unaired (and respect ignored) elif self.airdate == datetime.date.fromordinal(1): if IGNORED == self.status: msg = 'Episode has no air date and marked %s, no change' else: msg = 'Episode has no air date, marking it %s' self.status = UNAIRED # if the airdate is in the past elif UNAIRED == self.status: msg = ('Episode status %s%s, with air date in the past, marking it ' % ( statusStrings[self.status], ','.join([(' is a special', '')[0 < self.season], ('', ' is paused')[self.show.paused]])) + '%s') self.status = (SKIPPED, WANTED)[0 < self.season and not self.show.paused] # if still UNKNOWN or SKIPPED with the deprecated future airdate method elif UNKNOWN == self.status or (SKIPPED == self.status and old_airdate_future): msg = ('Episode status %s%s, with air date in the past, marking it ' % ( statusStrings[self.status], ','.join([ ('', ' has old future date format')[SKIPPED == self.status and old_airdate_future], ('', ' is being updated')[bool(update)], (' is a special', '')[0 < self.season]])) + '%s') self.status = (SKIPPED, WANTED)[update and not self.show.paused and 0 < self.season] else: msg = 'Not touching episode status %s, with air date in the past, because there is no file' else: msg = 'Not touching episode status %s, because there is no file' logger.log(msg % statusStrings[self.status], logger.DEBUG) # if we have a media file then it's downloaded elif sickbeard.helpers.has_media_ext(self.location): # leave propers alone, you have to either post-process them or manually change them back if self.status not in Quality.SNATCHED_ANY + Quality.DOWNLOADED + Quality.ARCHIVED: msg = '(1) Status changes from %s to ' % statusStrings[self.status] self.status = Quality.statusFromNameOrFile(self.location, anime=self.show.is_anime) logger.log('%s%s' % (msg, statusStrings[self.status]), logger.DEBUG) # shouldn't get here probably else: msg = '(2) Status changes from %s to ' % statusStrings[self.status] self.status = UNKNOWN logger.log('%s%s' % (msg, statusStrings[self.status]), logger.DEBUG) def loadFromNFO(self, location): if not ek.ek(os.path.isdir, self.show._location): logger.log('%s: The show directory is missing, not bothering to try loading the episode NFO' % self.show.indexerid) return logger.log('%s: Loading episode details from the NFO file associated with %s' % (self.show.indexerid, location), logger.DEBUG) self.location = location if self.location != "": if UNKNOWN == self.status and sickbeard.helpers.has_media_ext(self.location): status_quality = Quality.statusFromNameOrFile(self.location, anime=self.show.is_anime) logger.log('(3) Status changes from %s to %s' % (self.status, status_quality), logger.DEBUG) self.status = status_quality nfoFile = sickbeard.helpers.replaceExtension(self.location, 'nfo') logger.log('%s: Using NFO name %s' % (self.show.indexerid, nfoFile), logger.DEBUG) if ek.ek(os.path.isfile, nfoFile): try: showXML = etree.ElementTree(file=nfoFile) except (SyntaxError, ValueError) as e: logger.log('Error loading the NFO, backing up the NFO and skipping for now: %s' % ex(e), logger.ERROR) # TODO: figure out what's wrong and fix it try: ek.ek(os.rename, nfoFile, '%s.old' % nfoFile) except Exception as e: logger.log( 'Failed to rename your episode\'s NFO file - you need to delete it or fix it: %s' % ex(e), logger.ERROR) raise exceptions.NoNFOException('Error in NFO format') for epDetails in showXML.getiterator('episodedetails'): if epDetails.findtext('season') is None or int(epDetails.findtext('season')) != self.season or \ epDetails.findtext('episode') is None or int( epDetails.findtext('episode')) != self.episode: logger.log('%s: NFO has an block for a different episode - wanted %sx%s but got %sx%s' % (self.show.indexerid, self.season, self.episode, epDetails.findtext('season'), epDetails.findtext('episode')), logger.DEBUG) continue if epDetails.findtext('title') is None or epDetails.findtext('aired') is None: raise exceptions.NoNFOException('Error in NFO format (missing episode title or airdate)') self.name = epDetails.findtext('title') self.episode = int(epDetails.findtext('episode')) self.season = int(epDetails.findtext('season')) sickbeard.scene_numbering.xem_refresh(self.show.indexerid, self.show.indexer) self.scene_absolute_number = sickbeard.scene_numbering.get_scene_absolute_numbering( self.show.indexerid, self.show.indexer, self.absolute_number ) self.scene_season, self.scene_episode = sickbeard.scene_numbering.get_scene_numbering( self.show.indexerid, self.show.indexer, self.season, self.episode ) self.description = epDetails.findtext('plot') if self.description is None: self.description = '' if epDetails.findtext('aired'): rawAirdate = [int(x) for x in epDetails.findtext('aired').split("-")] self.airdate = datetime.date(rawAirdate[0], rawAirdate[1], rawAirdate[2]) else: self.airdate = datetime.date.fromordinal(1) self.hasnfo = True else: self.hasnfo = False if ek.ek(os.path.isfile, sickbeard.helpers.replaceExtension(nfoFile, 'tbn')): self.hastbn = True else: self.hastbn = False def __str__(self): return '%s - %sx%s - %s\n' % (self.show.name, self.season, self.episode, self.name) \ + 'location: %s\n' % self.location \ + 'description: %s\n' % self.description \ + 'subtitles: %s\n' % ','.join(self.subtitles) \ + 'subtitles_searchcount: %s\n' % self.subtitles_searchcount \ + 'subtitles_lastsearch: %s\n' % self.subtitles_lastsearch \ + 'airdate: %s (%s)\n' % (self.airdate.toordinal(), self.airdate) \ + 'hasnfo: %s\n' % self.hasnfo \ + 'hastbn: %s\n' % self.hastbn \ + 'status: %s\n' % self.status def createMetaFiles(self): if not ek.ek(os.path.isdir, self.show._location): logger.log('%s: The show directory is missing, not bothering to try to create metadata' % self.show.indexerid) return self.createNFO() self.createThumbnail() if self.checkForMetaFiles(): self.saveToDB() def createNFO(self): result = False for cur_provider in sickbeard.metadata_provider_dict.values(): result = cur_provider.create_episode_metadata(self) or result return result def createThumbnail(self): result = False for cur_provider in sickbeard.metadata_provider_dict.values(): result = cur_provider.create_episode_thumb(self) or result return result def deleteEpisode(self): logger.log('Deleting %s %sx%s from the DB' % (self.show.name, self.season, self.episode), logger.DEBUG) # remove myself from the show dictionary if self.show.getEpisode(self.season, self.episode, noCreate=True) == self: logger.log('Removing myself from my show\'s list', logger.DEBUG) del self.show.episodes[self.season][self.episode] # delete myself from the DB logger.log('Deleting myself from the database', logger.DEBUG) myDB = db.DBConnection() sql = 'DELETE FROM tv_episodes WHERE showid=%s AND indexer=%s AND season=%s AND episode=%s' % \ (self.show.indexerid, self.show.indexer, self.season, self.episode) myDB.action(sql) raise exceptions.EpisodeDeletedException() def get_sql(self, forceSave=False): """ Creates SQL queue for this episode if any of its data has been changed since the last save. forceSave: If True it will create SQL queue even if no data has been changed since the last save (aka if the record is not dirty). """ if not self.dirty and not forceSave: logger.log('%s: Not creating SQL queue - record is not dirty' % self.show.indexerid, logger.DEBUG) return self.dirty = False return [ 'INSERT OR REPLACE INTO tv_episodes (episode_id, indexerid, indexer, name, description, subtitles, ' 'subtitles_searchcount, subtitles_lastsearch, airdate, hasnfo, hastbn, status, location, file_size, ' 'release_name, is_proper, showid, season, episode, absolute_number, version, release_group, ' 'scene_absolute_number, scene_season, scene_episode) VALUES ' '((SELECT episode_id FROM tv_episodes WHERE indexer = ? AND showid = ? AND season = ? AND episode = ?)' ',?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,' '(SELECT scene_absolute_number FROM tv_episodes WHERE indexer = ? AND showid = ? AND season = ? AND episode = ?),' '(SELECT scene_season FROM tv_episodes WHERE indexer = ? AND showid = ? AND season = ? AND episode = ?),' '(SELECT scene_episode FROM tv_episodes WHERE indexer = ? AND showid = ? AND season = ? AND episode = ?));', [self.show.indexer, self.show.indexerid, self.season, self.episode, self.indexerid, self.indexer, self.name, self.description, ','.join([sub for sub in self.subtitles]), self.subtitles_searchcount, self.subtitles_lastsearch, self.airdate.toordinal(), self.hasnfo, self.hastbn, self.status, self.location, self.file_size, self.release_name, self.is_proper, self.show.indexerid, self.season, self.episode, self.absolute_number, self.version, self.release_group, self.show.indexer, self.show.indexerid, self.season, self.episode, self.show.indexer, self.show.indexerid, self.season, self.episode, self.show.indexer, self.show.indexerid, self.season, self.episode]] def saveToDB(self, forceSave=False): """ Saves this episode to the database if any of its data has been changed since the last save. forceSave: If True it will save to the database even if no data has been changed since the last save (aka if the record is not dirty). """ if not self.dirty and not forceSave: logger.log('%s: Not saving episode to db - record is not dirty' % self.show.indexerid, logger.DEBUG) return logger.log('%s: Saving episode details to database' % self.show.indexerid, logger.DEBUG) logger.log('STATUS IS %s' % statusStrings[self.status], logger.DEBUG) newValueDict = {'indexerid': self.indexerid, 'indexer': self.indexer, 'name': self.name, 'description': self.description, 'subtitles': ','.join([sub for sub in self.subtitles]), 'subtitles_searchcount': self.subtitles_searchcount, 'subtitles_lastsearch': self.subtitles_lastsearch, 'airdate': self.airdate.toordinal(), 'hasnfo': self.hasnfo, 'hastbn': self.hastbn, 'status': self.status, 'location': self.location, 'file_size': self.file_size, 'release_name': self.release_name, 'is_proper': self.is_proper, 'absolute_number': self.absolute_number, 'version': self.version, 'release_group': self.release_group } controlValueDict = {'showid': self.show.indexerid, 'season': self.season, 'episode': self.episode} # use a custom update/insert method to get the data into the DB myDB = db.DBConnection() myDB.upsert('tv_episodes', newValueDict, controlValueDict) self.dirty = False def fullPath(self): if self.location == None or self.location == "": return None else: return ek.ek(os.path.join, self.show.location, self.location) def createStrings(self, pattern=None): patterns = [ '%S.N.S%SE%0E', '%S.N.S%0SE%E', '%S.N.S%SE%E', '%S.N.S%0SE%0E', '%SN S%SE%0E', '%SN S%0SE%E', '%SN S%SE%E', '%SN S%0SE%0E' ] strings = [] if not pattern: for p in patterns: strings += [self._format_pattern(p)] return strings return self._format_pattern(pattern) def prettyName(self): """ Returns the name of this episode in a "pretty" human-readable format. Used for logging and notifications and such. Returns: A string representing the episode's name and season/ep numbers """ if self.show.anime and not self.show.scene: return self._format_pattern('%SN - %AB - %EN') elif self.show.air_by_date: return self._format_pattern('%SN - %AD - %EN') return self._format_pattern('%SN - %Sx%0E - %EN') def _ep_name(self): """ Returns the name of the episode to use during renaming. Combines the names of related episodes. Eg. "Ep Name (1)" and "Ep Name (2)" becomes "Ep Name" "Ep Name" and "Other Ep Name" becomes "Ep Name & Other Ep Name" """ multiNameRegex = '(.*) \(\d{1,2}\)' self.relatedEps = sorted(self.relatedEps, key=lambda x: x.episode) if len(self.relatedEps) == 0: goodName = self.name else: goodName = '' singleName = True curGoodName = None for curName in [self.name] + [x.name for x in self.relatedEps]: match = re.match(multiNameRegex, curName) if not match: singleName = False break if curGoodName == None: curGoodName = match.group(1) elif curGoodName != match.group(1): singleName = False break if singleName: goodName = curGoodName else: goodName = self.name for relEp in self.relatedEps: goodName += " & " + relEp.name return goodName def _replace_map(self): """ Generates a replacement map for this episode which maps all possible custom naming patterns to the correct value for this episode. Returns: A dict with patterns as the keys and their replacement values as the values. """ ep_name = self._ep_name() def dot(name): return helpers.sanitizeSceneName(name) def us(name): return re.sub('[ -]', '_', name) def release_name(name, is_anime=False): if name: name = helpers.remove_non_release_groups(name, is_anime) return name def release_group(show, name): if name: name = helpers.remove_non_release_groups(name, show.is_anime) else: return '' try: np = NameParser(name, showObj=show, naming_pattern=True) parse_result = np.parse(name) except (InvalidNameException, InvalidShowException) as e: logger.log('Unable to get parse release_group: %s' % ex(e), logger.DEBUG) return '' if not parse_result.release_group: return '' return parse_result.release_group epStatus, epQual = Quality.splitCompositeStatus(self.status) # @UnusedVariable if sickbeard.NAMING_STRIP_YEAR: show_name = re.sub('\(\d+\)$', '', self.show.name).rstrip() else: show_name = self.show.name return { '%SN': show_name, '%S.N': dot(show_name), '%S_N': us(show_name), '%EN': ep_name, '%E.N': dot(ep_name), '%E_N': us(ep_name), '%QN': Quality.qualityStrings[epQual], '%Q.N': dot(Quality.qualityStrings[epQual]), '%Q_N': us(Quality.qualityStrings[epQual]), '%S': str(self.season), '%0S': '%02d' % self.season, '%E': str(self.episode), '%0E': '%02d' % self.episode, '%XS': str(self.scene_season), '%0XS': '%02d' % self.scene_season, '%XE': str(self.scene_episode), '%0XE': '%02d' % self.scene_episode, '%AB': '%(#)03d' % {'#': self.absolute_number}, '%XAB': '%(#)03d' % {'#': self.scene_absolute_number}, '%RN': release_name(self.release_name, self.show.is_anime), '%RG': release_group(self.show, self.release_name), '%AD': str(self.airdate).replace('-', ' '), '%A.D': str(self.airdate).replace('-', '.'), '%A_D': us(str(self.airdate)), '%A-D': str(self.airdate), '%Y': str(self.airdate.year), '%M': str(self.airdate.month), '%D': str(self.airdate.day), '%0M': '%02d' % self.airdate.month, '%0D': '%02d' % self.airdate.day, '%RT': "PROPER" if self.is_proper else "", '%V': 'v%s' % self.version if self.show.is_anime and self.version > 1 else '', } def _format_string(self, pattern, replace_map): """ Replaces all template strings with the correct value """ result_name = pattern # do the replacements for cur_replacement in sorted(replace_map.keys(), reverse=True): result_name = result_name.replace(cur_replacement, helpers.sanitizeFileName(replace_map[cur_replacement])) result_name = result_name.replace(cur_replacement.lower(), helpers.sanitizeFileName(replace_map[cur_replacement].lower())) return result_name def _format_pattern(self, pattern=None, multi=None, anime_type=None): """ Manipulates an episode naming pattern and then fills the template in """ if pattern == None: pattern = sickbeard.NAMING_PATTERN if multi == None: multi = sickbeard.NAMING_MULTI_EP if anime_type == None: anime_type = sickbeard.NAMING_ANIME replace_map = self._replace_map() result_name = pattern # if there's no release group then replace it with a reasonable facsimile if not replace_map['%RN']: if self.show.air_by_date or self.show.sports: result_name = result_name.replace('%RN', '%S.N.%A.D.%E.N-SickGear') result_name = result_name.replace('%rn', '%s.n.%A.D.%e.n-SickGear') elif anime_type != 3: result_name = result_name.replace('%RN', '%S.N.%AB.%E.N-SickGear') result_name = result_name.replace('%rn', '%s.n.%ab.%e.n-SickGear') else: result_name = result_name.replace('%RN', '%S.N.S%0SE%0E.%E.N-SickGear') result_name = result_name.replace('%rn', '%s.n.s%0se%0e.%e.n-SickGear') result_name = result_name.replace('%RG', 'SickGear') result_name = result_name.replace('%rg', 'SickGear') logger.log('Episode has no release name, replacing it with a generic one: %s' % result_name, logger.DEBUG) if not replace_map['%RT']: result_name = re.sub('([ _.-]*)%RT([ _.-]*)', r'\2', result_name) # split off ep name part only name_groups = re.split(r'[\\/]', result_name) # figure out the double-ep numbering style for each group, if applicable for cur_name_group in name_groups: season_format = sep = ep_sep = ep_format = None season_ep_regex = ''' (?P[ _.-]*) ((?:s(?:eason|eries)?\s*)?%0?S(?![._]?N)) (.*?) (%0?E(?![._]?N)) (?P[ _.-]*) ''' ep_only_regex = '(E?%0?E(?![._]?N))' # try the normal way season_ep_match = re.search(season_ep_regex, cur_name_group, re.I | re.X) ep_only_match = re.search(ep_only_regex, cur_name_group, re.I | re.X) # if we have a season and episode then collect the necessary data if season_ep_match: season_format = season_ep_match.group(2) ep_sep = season_ep_match.group(3) ep_format = season_ep_match.group(4) sep = season_ep_match.group('pre_sep') if not sep: sep = season_ep_match.group('post_sep') if not sep: sep = ' ' # force 2-3-4 format if they chose to extend if multi in (NAMING_EXTEND, NAMING_LIMITED_EXTEND, NAMING_LIMITED_EXTEND_E_PREFIXED): ep_sep = '-' regex_used = season_ep_regex # if there's no season then there's not much choice so we'll just force them to use 03-04-05 style elif ep_only_match: season_format = '' ep_sep = '-' ep_format = ep_only_match.group(1) sep = '' regex_used = ep_only_regex else: continue # we need at least this much info to continue if not ep_sep or not ep_format: continue # start with the ep string, eg. E03 ep_string = self._format_string(ep_format.upper(), replace_map) for other_ep in self.relatedEps: # for limited extend we only append the last ep if multi in (NAMING_LIMITED_EXTEND, NAMING_LIMITED_EXTEND_E_PREFIXED) and other_ep != self.relatedEps[ -1]: continue elif multi == NAMING_DUPLICATE: # add " - S01" ep_string += sep + season_format elif multi == NAMING_SEPARATED_REPEAT: ep_string += sep # add "E04" ep_string += ep_sep if multi == NAMING_LIMITED_EXTEND_E_PREFIXED: ep_string += 'E' ep_string += other_ep._format_string(ep_format.upper(), other_ep._replace_map()) if anime_type != 3: if self.absolute_number == 0: curAbsolute_number = self.episode else: curAbsolute_number = self.absolute_number if self.season != 0: # dont set absolute numbers if we are on specials ! if anime_type == 1: # this crazy person wants both ! (note: +=) ep_string += sep + '%(#)03d' % { '#': curAbsolute_number} elif anime_type == 2: # total anime freak only need the absolute number ! (note: =) ep_string = '%(#)03d' % {'#': curAbsolute_number} for relEp in self.relatedEps: if relEp.absolute_number != 0: ep_string += '-' + '%(#)03d' % {'#': relEp.absolute_number} else: ep_string += '-' + '%(#)03d' % {'#': relEp.episode} regex_replacement = None if anime_type == 2: regex_replacement = r'\g' + ep_string + r'\g' elif season_ep_match: regex_replacement = r'\g\g<2>\g<3>' + ep_string + r'\g' elif ep_only_match: regex_replacement = ep_string if regex_replacement: # fill out the template for this piece and then insert this piece into the actual pattern cur_name_group_result = re.sub('(?i)(?x)' + regex_used, regex_replacement, cur_name_group) # cur_name_group_result = cur_name_group.replace(ep_format, ep_string) # logger.log(u"found "+ep_format+" as the ep pattern using "+regex_used+" and replaced it with "+regex_replacement+" to result in "+cur_name_group_result+" from "+cur_name_group, logger.DEBUG) result_name = result_name.replace(cur_name_group, cur_name_group_result) result_name = self._format_string(result_name, replace_map) logger.log('formatting pattern: %s -> %s' % (pattern, result_name), logger.DEBUG) return result_name def proper_path(self): """ Figures out the path where this episode SHOULD live according to the renaming rules, relative from the show dir """ anime_type = sickbeard.NAMING_ANIME if not self.show.is_anime: anime_type = 3 result = self.formatted_filename(anime_type=anime_type) # if they want us to flatten it and we're allowed to flatten it then we will if self.show.flatten_folders and not sickbeard.NAMING_FORCE_FOLDERS: return result # if not we append the folder on and use that else: result = ek.ek(os.path.join, self.formatted_dir(), result) return result def formatted_dir(self, pattern=None, multi=None): """ Just the folder name of the episode """ if pattern == None: # we only use ABD if it's enabled, this is an ABD show, AND this is not a multi-ep if self.show.air_by_date and sickbeard.NAMING_CUSTOM_ABD and not self.relatedEps: pattern = sickbeard.NAMING_ABD_PATTERN elif self.show.sports and sickbeard.NAMING_CUSTOM_SPORTS and not self.relatedEps: pattern = sickbeard.NAMING_SPORTS_PATTERN elif self.show.anime and sickbeard.NAMING_CUSTOM_ANIME: pattern = sickbeard.NAMING_ANIME_PATTERN else: pattern = sickbeard.NAMING_PATTERN # split off the dirs only, if they exist name_groups = re.split(r'[\\/]', pattern) if len(name_groups) == 1: return '' else: return self._format_pattern(ek.ek(os.sep.join, name_groups[:-1]), multi) def formatted_filename(self, pattern=None, multi=None, anime_type=None): """ Just the filename of the episode, formatted based on the naming settings """ if pattern == None: # we only use ABD if it's enabled, this is an ABD show, AND this is not a multi-ep if self.show.air_by_date and sickbeard.NAMING_CUSTOM_ABD and not self.relatedEps: pattern = sickbeard.NAMING_ABD_PATTERN elif self.show.sports and sickbeard.NAMING_CUSTOM_SPORTS and not self.relatedEps: pattern = sickbeard.NAMING_SPORTS_PATTERN elif self.show.anime and sickbeard.NAMING_CUSTOM_ANIME: pattern = sickbeard.NAMING_ANIME_PATTERN else: pattern = sickbeard.NAMING_PATTERN # split off the dirs only, if they exist name_groups = re.split(r'[\\/]', pattern) return self._format_pattern(name_groups[-1], multi, anime_type) def rename(self): """ Renames an episode file and all related files to the location and filename as specified in the naming settings. """ if not ek.ek(os.path.isfile, self.location): logger.log('Can\'t perform rename on %s when it doesn\'t exist, skipping' % self.location, logger.WARNING) return proper_path = self.proper_path() absolute_proper_path = ek.ek(os.path.join, self.show.location, proper_path) absolute_current_path_no_ext, file_ext = ek.ek(os.path.splitext, self.location) absolute_current_path_no_ext_length = len(absolute_current_path_no_ext) related_subs = [] current_path = absolute_current_path_no_ext if absolute_current_path_no_ext.startswith(self.show.location): current_path = absolute_current_path_no_ext[len(self.show.location):] logger.log('Renaming/moving episode from the base path %s to %s' % (self.location, absolute_proper_path), logger.DEBUG) # if it's already named correctly then don't do anything if proper_path == current_path: logger.log('%s: File %s is already named correctly, skipping' % (self.indexerid, self.location), logger.DEBUG) return related_files = postProcessor.PostProcessor(self.location).list_associated_files( self.location, base_name_only=True) if self.show.subtitles and sickbeard.SUBTITLES_DIR != '': related_subs = postProcessor.PostProcessor(self.location).list_associated_files(sickbeard.SUBTITLES_DIR, subtitles_only=True) absolute_proper_subs_path = ek.ek(os.path.join, sickbeard.SUBTITLES_DIR, self.formatted_filename()) logger.log('Files associated to %s: %s' % (self.location, related_files), logger.DEBUG) # move the ep file result = helpers.rename_ep_file(self.location, absolute_proper_path, absolute_current_path_no_ext_length) # move related files for cur_related_file in related_files: cur_result = helpers.rename_ep_file(cur_related_file, absolute_proper_path, absolute_current_path_no_ext_length) if not cur_result: logger.log('%s: Unable to rename file %s' % (self.indexerid, cur_related_file), logger.ERROR) for cur_related_sub in related_subs: absolute_proper_subs_path = ek.ek(os.path.join, sickbeard.SUBTITLES_DIR, self.formatted_filename()) cur_result = helpers.rename_ep_file(cur_related_sub, absolute_proper_subs_path, absolute_current_path_no_ext_length) if not cur_result: logger.log('%s: Unable to rename file %s' % (self.indexerid, cur_related_sub), logger.ERROR) # save the ep with self.lock: if result: self.location = absolute_proper_path + file_ext for relEp in self.relatedEps: relEp.location = absolute_proper_path + file_ext # in case something changed with the metadata just do a quick check for curEp in [self] + self.relatedEps: curEp.checkForMetaFiles() # save any changes to the databas sql_l = [] with self.lock: for relEp in [self] + self.relatedEps: result = relEp.get_sql() if None is not result: sql_l.append(result) if 0 < len(sql_l): myDB = db.DBConnection() myDB.mass_action(sql_l) def airdateModifyStamp(self): """ Make the modify date and time of a file reflect the show air date and time. Note: Also called from postProcessor """ if not datetime.date == type(self.airdate) or 1 == self.airdate.year: logger.log('%s: Did not change modify date of %s because episode date is never aired or invalid' % (self.show.indexerid, ek.ek(os.path.basename, self.location)), logger.DEBUG) return hr, m = network_timezones.parse_time(self.show.airs) airtime = datetime.time(hr, m) aired_dt = datetime.datetime.combine(self.airdate, airtime) try: aired_epoch = helpers.datetime_to_epoch(aired_dt) filemtime = int(ek.ek(os.path.getmtime, self.location)) except (StandardError, Exception): return if filemtime != aired_epoch: result, loglevel = 'Changed', logger.MESSAGE if not helpers.touch_file(self.location, aired_epoch): result, loglevel = 'Error changing', logger.WARNING logger.log('%s: %s modify date of %s to show air date %s' % (self.show.indexerid, result, ek.ek(os.path.basename, self.location), aired_dt.strftime('%b %d,%Y (%H:%M)')), loglevel) def __getstate__(self): d = dict(self.__dict__) del d['lock'] return d def __setstate__(self, d): d['lock'] = threading.Lock() self.__dict__.update(d)