# # 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 . import datetime import fnmatch import os import copy import re from exceptions_helper import ex import sickgear from . import common, db, logger from .helpers import sanitize_scene_name from .name_parser.parser import InvalidNameException, InvalidShowException, NameParser from .scene_exceptions import get_scene_exceptions from sg_helpers import scantree from _23 import map_list, quote_plus from six import iterkeys, itervalues # noinspection PyUnreachableCode if False: from typing import AnyStr, List, Optional, Set, Union from .tv import TVShow # noinspection PyUnresolvedReferences from re import Pattern def pass_wordlist_checks(name, # type: AnyStr parse=True, # type: bool indexer_lookup=True, # type: bool show_obj=None # type: TVShow ): # type: (...) -> bool """ Filters out non-english and just all-around stupid releases by comparing the word list contents at boundaries or the end of name. :param name: the release name to check :type name: basestring :param parse: parse release name :type parse: bool :param indexer_lookup: use indexer lookup during paring :type indexer_lookup: bool :param show_obj: TVShow object :type show_obj: TVShow :return: True if the release name is OK, False if it's bad. :rtype: bool """ if parse: err_msg = u'Unable to parse the filename %s into a valid ' % name try: NameParser(indexer_lookup=indexer_lookup).parse(name) except InvalidNameException: logger.log(err_msg + 'episode', logger.DEBUG) return False except InvalidShowException: logger.log(err_msg + 'show', logger.DEBUG) return False word_list = {'sub(bed|ed|pack|s)', '(dk|fin|heb|kor|nor|nordic|pl|swe)sub(bed|ed|s)?', '(dir|sample|sub|nfo)fix', 'sample', '(dvd)?extras', 'dub(bed)?'} # if any of the bad strings are in the name then say no if sickgear.IGNORE_WORDS: word_list.update(sickgear.IGNORE_WORDS) req_word_list = copy.copy(sickgear.REQUIRE_WORDS) result = None if show_obj: if show_obj.rls_ignore_words and isinstance(show_obj.rls_ignore_words, set): if sickgear.IGNORE_WORDS_REGEX == show_obj.rls_ignore_words_regex: word_list.update(show_obj.rls_ignore_words) else: result = contains_any(name, show_obj.rls_ignore_words, rx=show_obj.rls_ignore_words_regex) if show_obj.rls_global_exclude_ignore and isinstance(show_obj.rls_global_exclude_ignore, set): word_list = word_list - show_obj.rls_global_exclude_ignore result = result or contains_any(name, word_list, rx=sickgear.IGNORE_WORDS_REGEX) if None is not result and result: logger.log(u'Ignored: %s for containing ignore word' % name, logger.DEBUG) return False result = None if show_obj: if show_obj.rls_require_words and isinstance(show_obj.rls_require_words, set): result = not contains_any(name, show_obj.rls_require_words, rx=show_obj.rls_require_words_regex) if show_obj.rls_global_exclude_require and isinstance(show_obj.rls_global_exclude_require, set): req_word_list = req_word_list - show_obj.rls_global_exclude_require # if any of the good strings aren't in the name then say no result = result or not_contains_any(name, req_word_list, rx=sickgear.REQUIRE_WORDS_REGEX) if None is not result and result: logger.log(u'Ignored: %s for not containing required word match' % name, logger.DEBUG) return False return True def not_contains_any(subject, # type: AnyStr lookup_words, # type: Union[AnyStr, Set[AnyStr]] rx=None, **kwargs ): # type: (...) -> bool return contains_any(subject, lookup_words, invert=True, rx=rx, **kwargs) def contains_any(subject, # type: AnyStr lookup_words, # type: Union[AnyStr, Set[AnyStr]] invert=False, # type: bool rx=None, **kwargs ): # type: (...) -> Optional[bool] """ Check if subject does or does not contain a match from a list or string of regular expression lookup words :param subject: word to test existence of :type subject: basestring :param lookup_words: List or comma separated string of words to search :type lookup_words: Union(list, set, basestring) :param re_prefix: insert string to all lookup words :type re_prefix: basestring :param re_suffix: append string to all lookup words :type re_suffix: basestring :param invert: invert function logic "contains any" into "does not contain any" :type invert: bool :param rx: lookup_words are regex :type rx: Union(NoneType, bool) :return: None if no checking was done. True for first match found, or if invert is False, :param lookup_words: List or comma separated string of words to search :param invert: invert function logic "contains any" into "does not contain any" :param kwargs: :return: None if no checking was done. True for first match found, or if invert is False, then True for first pattern that does not match, or False :rtype: Union(NoneType, bool) """ compiled_words = compile_word_list(lookup_words, rx=rx, **kwargs) if subject and compiled_words: for rc_filter in compiled_words: match = rc_filter.search(subject) if (match and not invert) or (not match and invert): msg = match and not invert and 'Found match' or '' msg = not match and invert and 'No match found' or msg logger.log(u'%s from pattern: %s in text: %s ' % (msg, rc_filter.pattern, subject), logger.DEBUG) return True return False return None def compile_word_list(lookup_words, # type: Union[AnyStr, Set[AnyStr]] re_prefix=r'(^|[\W_])', # type: AnyStr re_suffix=r'($|[\W_])', # type: AnyStr rx=None ): # type: (...) -> List[Pattern[AnyStr]] result = [] if lookup_words: if None is rx: search_raw = isinstance(lookup_words, list) if not search_raw: # noinspection PyUnresolvedReferences search_raw = not lookup_words.startswith('regex:') # noinspection PyUnresolvedReferences lookup_words = lookup_words[(6, 0)[search_raw]:].split(',') lookup_words = [x.strip() for x in lookup_words if x.strip()] else: search_raw = not rx for word in lookup_words: try: # !0 == regex and subject = s / 'what\'s the "time"' / what\'s\ the\ \"time\" subject = search_raw and re.escape(word) or re.sub(r'([\" \'])', r'\\\1', word) result.append(re.compile('(?i)%s%s%s' % (re_prefix, subject, re_suffix))) except re.error as e: logger.log(u'Failure to compile filter expression: %s ... Reason: %s' % (word, ex(e)), logger.DEBUG) diff = len(lookup_words) - len(result) if diff: logger.log(u'From %s expressions, %s was discarded during compilation' % (len(lookup_words), diff), logger.DEBUG) return result def url_encode(show_names, spacer='.'): # type: (List[AnyStr], AnyStr) -> List[AnyStr] """ :param show_names: show name :param spacer: spacer :return: """ return [quote_plus(n.replace('.', spacer).encode('utf-8', errors='replace')) for n in show_names] def get_show_names(ep_obj, spacer='.'): # type: (sickgear.tv.TVEpisode, AnyStr) -> List[AnyStr] """ :param ep_obj: episode object :param spacer: spacer :return: """ return get_show_names_all_possible(ep_obj.show_obj, season=ep_obj.season, spacer=spacer, force_anime=True) def get_show_names_all_possible(show_obj, season=-1, scenify=True, spacer='.', force_anime=False): # type: (sickgear.tv.TVShow, int, bool, AnyStr, bool) -> List[AnyStr] """ :param show_obj: show object :param season: season :param scenify: :param spacer: spacer :param force_anime: :return: """ show_names = list(set( all_possible_show_names(show_obj, season=season, force_anime=force_anime))) # type: List[AnyStr] if scenify: show_names = map_list(sanitize_scene_name, show_names) return url_encode(show_names, spacer) def make_scene_season_search_string(show_obj, # type: sickgear.tv.TVShow ep_obj, # type: sickgear.tv.TVEpisode ignore_allowlist=False, # type: bool extra_search_type=None ): # type: (...) -> List[AnyStr] """ :param show_obj: show object :param ep_obj: episode object :param ignore_allowlist: :param extra_search_type: :return: list of search strings """ if show_obj.air_by_date or show_obj.sports: numseasons = 0 # the search string for air by date shows is just season_strings = [str(ep_obj.airdate).split('-')[0]] elif show_obj.is_anime: numseasons = 0 ep_obj_list = show_obj.get_all_episodes(ep_obj.season) # get show qualities any_qualities, best_qualities = common.Quality.splitQuality(show_obj.quality) # compile a list of all the episode numbers we need in this 'season' season_strings = [] for episode in ep_obj_list: # get quality of the episode cur_composite_status = episode.status cur_status, cur_quality = common.Quality.splitCompositeStatus(cur_composite_status) if best_qualities: highest_best_quality = max(best_qualities) else: highest_best_quality = 0 # if we need a better one then add it to the list of episodes to fetch if (cur_status in ( common.DOWNLOADED, common.SNATCHED) and cur_quality < highest_best_quality) or cur_status == common.WANTED: ab_number = episode.scene_absolute_number if 0 < ab_number: season_strings.append("%02d" % ab_number) else: my_db = db.DBConnection() sql_result = my_db.select( 'SELECT COUNT(DISTINCT season) AS numseasons' ' FROM tv_episodes' ' WHERE indexer = ? AND showid = ?' ' AND season != 0', [show_obj.tvid, show_obj.prodid]) numseasons = int(sql_result[0][0]) season_strings = ["S%02d" % int(ep_obj.scene_season)] show_names = get_show_names_all_possible(show_obj, ep_obj.scene_season) to_return = [] # search each show name for cur_name in show_names: # most providers all work the same way if not extra_search_type: # if there's only one season then we can just use the show name straight up if 1 == numseasons: to_return.append(cur_name) # for providers that don't allow multiple searches in one request we only search for Sxx style stuff else: for cur_season in season_strings: if not ignore_allowlist and show_obj.is_anime \ and None is not show_obj.release_groups and show_obj.release_groups.allowlist: for keyword in show_obj.release_groups.allowlist: to_return.append(keyword + '.' + cur_name + "." + cur_season) else: to_return.append(cur_name + "." + cur_season) return to_return def make_scene_search_string(show_obj, # type: sickgear.tv.TVShow ep_obj, # type: sickgear.tv.TVEpisode ignore_allowlist=False # type: bool ): # type: (...) -> List[AnyStr] """ :param show_obj: show object :param ep_obj: episode object :param ignore_allowlist: :return: list or search strings """ my_db = db.DBConnection() sql_result = my_db.select( 'SELECT COUNT(DISTINCT season) AS numseasons' ' FROM tv_episodes' ' WHERE indexer = ? AND showid = ? AND season != 0', [show_obj.tvid, show_obj.prodid]) num_seasons = int(sql_result[0][0]) # see if we should use dates instead of episodes if (show_obj.air_by_date or show_obj.sports) and ep_obj.airdate != datetime.date.fromordinal(1): ep_strings = [str(ep_obj.airdate)] elif show_obj.is_anime: ep_strings = ['%02i' % int(ep_obj.scene_absolute_number if 0 < ep_obj.scene_absolute_number else ep_obj.scene_episode)] else: ep_strings = ['S%02iE%02i' % (int(ep_obj.scene_season), int(ep_obj.scene_episode)), '%ix%02i' % (int(ep_obj.scene_season), int(ep_obj.scene_episode))] # for single-season shows just search for the show name -- if total ep count (exclude s0) is less than 11 # due to the amount of qualities and releases, it is easy to go over the 50 result limit on rss feeds otherwise if 1 == num_seasons and not ep_obj.show_obj.is_anime: ep_strings = [''] show_names = get_show_names_all_possible(show_obj, ep_obj.scene_season) to_return = [] for cur_show_obj in show_names: for cur_ep_string in ep_strings: if not ignore_allowlist and ep_obj.show_obj.is_anime and \ None is not ep_obj.show_obj.release_groups and ep_obj.show_obj.release_groups.allowlist: for keyword in ep_obj.show_obj.release_groups.allowlist: to_return.append(keyword + '.' + cur_show_obj + '.' + cur_ep_string) else: to_return.append(cur_show_obj + '.' + cur_ep_string) return to_return def all_possible_show_names(show_obj, season=-1, force_anime=False): # type: (sickgear.tv.TVShow, int, bool) -> List[AnyStr] """ Figures out every possible variation of the name for a particular show. Includes TVDB name, TVRage name, country codes on the end, eg. "Show Name (AU)", and any scene exception names. :param show_obj: a TVShow object that we should get the names of :param season: season :param force_anime: :return: a list of all the possible show names """ show_names = get_scene_exceptions(show_obj.tvid, show_obj.prodid, season=season)[:] if not show_names: # if we dont have any season specific exceptions fallback to generic exceptions season = -1 show_names = get_scene_exceptions(show_obj.tvid, show_obj.prodid, season=season)[:] if -1 == season: show_names.append(show_obj.name) if not show_obj.is_anime and not force_anime: new_show_names = [] country_list = common.countryList country_list.update(dict(zip(itervalues(common.countryList), iterkeys(common.countryList)))) for cur_name in set(show_names): if not cur_name: continue # if we have "Show Name Australia" or "Show Name (Australia)" this will add "Show Name (AU)" for # any countries defined in common.countryList # (and vice versa) for cur_country in country_list: if cur_name.endswith(' ' + cur_country): new_show_names.append(cur_name.replace(' ' + cur_country, ' (' + country_list[cur_country] + ')')) elif cur_name.endswith(' (' + cur_country + ')'): new_show_names.append(cur_name.replace(' (' + cur_country + ')', ' (' + country_list[cur_country] + ')')) # if we have "Show Name (2013)" this will strip the (2013) show year from the show name # newShowNames.append(re.sub('\(\d{4}\)','',curName)) show_names += new_show_names return show_names def determine_release_name(dir_name=None, nzb_name=None): # type: (AnyStr, AnyStr) -> Union[AnyStr, None] """Determine a release name from a nzb and/or folder name :param dir_name: dir name :param nzb_name: nzb name :return: None or release name """ if None is not nzb_name: logger.log(u'Using nzb name for release name.') return nzb_name.rpartition('.')[0] if not dir_name or not os.path.isdir(dir_name): return None # try to get the release name from nzb/nfo file_types = ['*.nzb', '*.nfo'] for search in file_types: results = [direntry.name for direntry in scantree(dir_name, include=[fnmatch.translate(search)], filter_kind=False, recurse=False)] if 1 == len(results): found_file = results[0].rpartition('.')[0] if pass_wordlist_checks(found_file): logger.log(u'Release name (%s) found from file (%s)' % (found_file, results[0])) return found_file.rpartition('.')[0] # If that fails, we try the folder folder = os.path.basename(dir_name) if pass_wordlist_checks(folder): # NOTE: Multiple failed downloads will change the folder name. # (e.g., appending #s) # Should we handle that? logger.log(u'Folder name (%s) appears to be a valid release name. Using it.' % folder) return folder return None def abbr_showname(name): # type: (AnyStr) -> AnyStr result = name for cur_from, cur_to in ( (r'^Star Trek\s*:\s*', r'ST: '), (r'^The Walking Dead\s*:\s*', r'TWD: '), ): result = re.sub('(?i)%s' % cur_from, cur_to, result) if name != result: break return result