You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
bazarr/libs/subliminal_patch/providers/xsubs.py

304 lines
11 KiB

# -*- coding: utf-8 -*-
5 years ago
from __future__ import absolute_import
import logging
import re
from subzero.language import Language
from guessit import guessit
from requests import Session
from subliminal.providers import ParserBeautifulSoup, Provider
from subliminal import __short_version__
from subliminal.cache import SHOW_EXPIRATION_TIME, region
from subliminal.exceptions import AuthenticationError, ConfigurationError
from subliminal.score import get_equivalent_release_groups
from subliminal.subtitle import Subtitle, fix_line_ending, guess_matches
from subliminal.utils import sanitize, sanitize_release_group
from subliminal.video import Episode
logger = logging.getLogger(__name__)
article_re = re.compile(r'^([A-Za-z]{1,3}) (.*)$')
episode_re = re.compile(r'^(\d+)(-(\d+))*$')
class XSubsSubtitle(Subtitle):
"""XSubs Subtitle."""
provider_name = 'xsubs'
def __init__(self, language, page_link, series, season, episode, year, title, version, download_link):
super(XSubsSubtitle, self).__init__(language, page_link=page_link)
self.series = series
self.season = season
self.episode = episode
self.year = year
self.title = title
self.version = version
self.download_link = download_link
self.hearing_impaired = None
self.encoding = 'windows-1253'
@property
def id(self):
return self.download_link
def get_matches(self, video):
matches = set()
if isinstance(video, Episode):
# series name
if video.series and sanitize(self.series) in (
sanitize(name) for name in [video.series] + video.alternative_series):
matches.add('series')
# season
if video.season and self.season == video.season:
matches.add('season')
# episode
if video.episode and self.episode == video.episode:
matches.add('episode')
# title of the episode
if video.title and sanitize(self.title) == sanitize(video.title):
matches.add('title')
# year
if video.original_series and self.year is None or video.year and video.year == self.year:
matches.add('year')
# release_group
if (video.release_group and self.version and
any(r in sanitize_release_group(self.version)
for r in get_equivalent_release_groups(sanitize_release_group(video.release_group)))):
matches.add('release_group')
# other properties
matches |= guess_matches(video, guessit(self.version, {'type': 'episode'}), partial=True)
return matches
class XSubsProvider(Provider):
"""XSubs Provider."""
languages = {Language(l) for l in ['ell']}
video_types = (Episode,)
server_url = 'http://xsubs.tv'
sign_in_url = '/xforum/account/signin/'
sign_out_url = '/xforum/account/signout/'
all_series_url = '/series/all.xml'
series_url = '/series/{:d}/main.xml'
season_url = '/series/{show_id:d}/{season:d}.xml'
page_link = '/ice/xsw.xml?srsid={show_id:d}#{season_id:d};{season:d};{episode:d}'
download_link = '/xthru/getsub/{:d}'
subtitle_class = XSubsSubtitle
def __init__(self, username=None, password=None):
if any((username, password)) and not all((username, password)):
raise ConfigurationError('Username and password must be specified')
self.username = username
self.password = password
self.logged_in = False
self.session = None
def initialize(self):
self.session = Session()
self.session.headers['User-Agent'] = 'Subliminal/{}'.format(__short_version__)
# login
if self.username and self.password:
logger.info('Logging in')
self.session.get(self.server_url + self.sign_in_url)
data = {'username': self.username,
'password': self.password,
'csrfmiddlewaretoken': self.session.cookies['csrftoken']}
r = self.session.post(self.server_url + self.sign_in_url, data, allow_redirects=False, timeout=10)
if r.status_code != 302:
raise AuthenticationError(self.username)
logger.debug('Logged in')
self.logged_in = True
def terminate(self):
# logout
if self.logged_in:
logger.info('Logging out')
r = self.session.get(self.server_url + self.sign_out_url, timeout=10)
r.raise_for_status()
logger.debug('Logged out')
self.logged_in = False
self.session.close()
@region.cache_on_arguments(expiration_time=SHOW_EXPIRATION_TIME, should_cache_fn=lambda value: value)
def _get_show_ids(self):
# get the shows page
logger.info('Getting show ids')
r = self.session.get(self.server_url + self.all_series_url, timeout=10)
r.raise_for_status()
if not r.content:
logger.debug('No data returned from provider')
return []
soup = ParserBeautifulSoup(r.content, ['lxml', 'html.parser'])
# populate the show ids
show_ids = {}
for show_category in soup.findAll('seriesl'):
if show_category.attrs['category'] == u'Σειρές':
for show in show_category.findAll('series'):
show_ids[sanitize(show.text)] = int(show['srsid'])
break
logger.debug('Found %d show ids', len(show_ids))
return show_ids
def get_show_id(self, series_names, year=None):
series_sanitized_names = []
for name in series_names:
sanitized_name = sanitize(name)
series_sanitized_names.append(sanitized_name)
alternative_name = _get_alternative_name(sanitized_name)
if alternative_name:
series_sanitized_names.append(alternative_name)
show_ids = self._get_show_ids()
show_id = None
for series_sanitized in series_sanitized_names:
# attempt with year
if year:
logger.debug('Getting show id with year')
show_id = show_ids.get('{series} {year:d}'.format(series=series_sanitized, year=year))
# attempt with article at the end
if not show_id and year:
logger.debug('Getting show id with year in brackets')
show_id = show_ids.get('{series} [{year:d}]'.format(series=series_sanitized, year=year))
# attempt clean
if not show_id:
logger.debug('Getting show id')
show_id = show_ids.get(series_sanitized)
if show_id:
break
return int(show_id) if show_id else None
def query(self, show_id, series, season, year=None, country=None):
# get the season list of the show
logger.info('Getting the season list of show id %d', show_id)
r = self.session.get(self.server_url + self.series_url.format(show_id), timeout=10)
r.raise_for_status()
if not r.content:
logger.debug('No data returned from provider')
return []
soup = ParserBeautifulSoup(r.content, ['lxml', 'html.parser'])
series = soup.find('name').text
# loop over season rows
seasons = soup.findAll('series_group')
season_id = None
for season_row in seasons:
try:
parsed_season = int(season_row['ssnnum'])
if parsed_season == season:
season_id = int(season_row['ssnid'])
break
except (ValueError, TypeError):
continue
if season_id is None:
logger.debug('Season not found in provider')
return []
# get the subtitle list of the season
logger.info('Getting the subtitle list of season %d', season)
r = self.session.get(self.server_url + self.season_url.format(show_id=show_id, season=season_id), timeout=10)
r.raise_for_status()
if not r.content:
logger.debug('No data returned from provider')
return []
soup = ParserBeautifulSoup(r.content, ['lxml', 'html.parser'])
subtitles = []
# loop over episode rows
for subtitle_group in soup.findAll('subg'):
# read the episode info
episode_info = subtitle_group.find('etitle')
if episode_info is None:
continue
episodes = []
episode_match = episode_re.match(episode_info['number'])
if episode_match:
episodes = [int(e) for e in [episode_match.group(1), episode_match.group(3)] if e]
subtitle_info = subtitle_group.find('sgt')
if subtitle_info is None:
continue
season = int(subtitle_info['ssnnum'])
episode_id = int(subtitle_info['epsid'])
# filter out unreleased subtitles
for subs_tag in subtitle_group.findAll('sr'):
if subs_tag['published_on'] == '':
continue
page_link = self.server_url + self.page_link.format(show_id=show_id, season_id=season_id,
season=season, episode=episode_id)
title = episode_info['title']
version = subs_tag.fmt.text + ' ' + subs_tag.team.text
download_link = self.server_url + self.download_link.format(int(subs_tag['rlsid']))
for episode in episodes:
subtitle = self.subtitle_class(Language.fromalpha2('el'), page_link, series, season, episode, year,
title, version, download_link)
logger.debug('Found subtitle %r', subtitle)
subtitles.append(subtitle)
return subtitles
def list_subtitles(self, video, languages):
if isinstance(video, Episode):
# lookup show_id
titles = [video.series] + video.alternative_series
show_id = self.get_show_id(titles, video.year)
# query for subtitles with the show_id
if show_id:
subtitles = [s for s in self.query(show_id, video.series, video.season, video.year)
if s.language in languages and s.season == video.season and s.episode == video.episode]
if subtitles:
return subtitles
else:
logger.error('No show id found for %r (%r)', video.series, {'year': video.year})
return []
def download_subtitle(self, subtitle):
if isinstance(subtitle, XSubsSubtitle):
# download the subtitle
logger.info('Downloading subtitle %r', subtitle)
r = self.session.get(subtitle.download_link, headers={'Referer': subtitle.page_link},
timeout=10)
r.raise_for_status()
if not r.content:
logger.debug('Unable to download subtitle. No data returned from provider')
return
subtitle.content = fix_line_ending(r.content)
def _get_alternative_name(series):
article_match = article_re.match(series)
if article_match:
return '{series} {article}'.format(series=article_match.group(2), article=article_match.group(1))
return None