bazarr/libs/subliminal_patch/providers/soustitreseu.py

293 lines
10 KiB
Python
Raw Normal View History

2020-07-21 04:28:34 +00:00
# -*- coding: utf-8 -*-
from __future__ import absolute_import
import io
import os
import logging
from urllib.parse import unquote
from zipfile import ZipFile, is_zipfile
from rarfile import RarFile, is_rarfile
from guessit import guessit
2021-01-17 13:58:01 +00:00
from subliminal_patch.http import RetryingCFSession
2020-07-21 04:28:34 +00:00
import chardet
from bs4 import NavigableString, UnicodeDammit
from subzero.language import Language
from subliminal_patch.providers import Provider
from subliminal_patch.providers.mixins import ProviderSubtitleArchiveMixin
from subliminal_patch.subtitle import Subtitle, guess_matches
2020-07-21 04:28:34 +00:00
from subliminal_patch.score import get_scores, framerate_equal
from subliminal.providers import ParserBeautifulSoup
from subliminal.subtitle import sanitize, SUBTITLE_EXTENSIONS
2020-07-21 04:28:34 +00:00
from subliminal.video import Episode, Movie
from .utils import FIRST_THOUSAND_OR_SO_USER_AGENTS as AGENT_LIST
logger = logging.getLogger(__name__)
class SoustitreseuSubtitle(Subtitle):
"""Sous-Titres.eu Subtitle."""
provider_name = 'soustitreseu'
def __init__(self, language, video, name, data, content, is_perfect_match):
self.language = language
self.srt_filename = name
self.release_info = name
self.page_link = None
self.download_link = None
self.data = data
self.video = video
self.matches = None
self.content = content
self.hearing_impaired = None
self.is_perfect_match = is_perfect_match
self._guessed_encoding = None
2020-07-21 04:28:34 +00:00
@property
def id(self):
return self.srt_filename
def get_matches(self, video):
matches = set()
if self.is_perfect_match:
if isinstance(video, Episode):
matches.add('series')
else:
matches.add('title')
# guess additional info from data
matches |= guess_matches(video, self.data)
self.matches = matches
self.data = None # removing this make the subtitles object unpickable
return matches
def guess_encoding(self):
# override default subtitle guess_encoding method to not include language-specific encodings guessing
# chardet encoding detection seem to yield better results
"""Guess encoding using chardet.
:return: the guessed encoding.
:rtype: str
"""
if self._guessed_encoding:
return self._guessed_encoding
logger.info('Guessing encoding for language %s', self.language)
# guess/detect encoding using chardet
encoding = chardet.detect(self.content)['encoding']
logger.info('Chardet found encoding %s', encoding)
if not encoding:
# fallback on bs4
logger.info('Falling back to bs4 detection')
a = UnicodeDammit(self.content)
logger.info("bs4 detected encoding: %s", a.original_encoding)
if a.original_encoding:
self._guessed_encoding = a.original_encoding
return a.original_encoding
raise ValueError(u"Couldn't guess the proper encoding for %s", self)
self._guessed_encoding = encoding
return encoding
class SoustitreseuProvider(Provider, ProviderSubtitleArchiveMixin):
"""Sous-Titres.eu Provider."""
subtitle_class = SoustitreseuSubtitle
languages = {Language(l) for l in ['fra', 'eng']}
video_types = (Episode, Movie)
2020-07-21 04:28:34 +00:00
server_url = 'https://www.sous-titres.eu/'
search_url = server_url + 'search.html'
def __init__(self):
self.session = None
self.is_perfect_match = False
def initialize(self):
2021-01-17 13:58:01 +00:00
self.session = RetryingCFSession()
2020-07-21 04:28:34 +00:00
self.session.headers['Referer'] = self.server_url
def terminate(self):
self.session.close()
def query_series(self, video, title):
subtitles = []
2021-01-17 13:58:01 +00:00
r = self.session.get(self.search_url, params={'q': title}, timeout=30)
2020-07-21 04:28:34 +00:00
r.raise_for_status()
soup = ParserBeautifulSoup(r.content.decode('utf-8', 'ignore'), ['html.parser'])
# loop over series name
self.is_perfect_match = False
series_url = []
series = soup.select('.serie > h3 > a')
for item in series:
# title
if title in item.text:
series_url.append(item.attrs['href'])
self.is_perfect_match = True
series_subs_archives_url = []
for series_page in series_url:
page_link = self.server_url + series_page
2021-01-17 13:58:01 +00:00
r = self.session.get(page_link, timeout=30)
2020-07-21 04:28:34 +00:00
r.raise_for_status()
soup = ParserBeautifulSoup(r.content.decode('utf-8', 'ignore'), ['html.parser'])
series_subs_archives = soup.select('a.subList')
for item in series_subs_archives:
matching_archive = False
subtitles_archive_name = unquote(item.attrs['href'].split('/')[-1:][0][:-4])
guessed_subs = guessit(subtitles_archive_name, {'type': 'episode'})
try:
season, episode = item.select_one('.episodenum').text.split('×')
guessed_subs.update({'season': int(season), 'episode': int(episode)})
except ValueError:
season = item.select_one('.episodenum').text[1:]
episode = None
guessed_subs.update({'season': int(season)})
if guessed_subs['season'] == video.season:
if 'episode' in guessed_subs:
if guessed_subs['episode'] == video.episode:
matching_archive = True
else:
matching_archive = True
if guessed_subs['season'] == 16:
print('test')
if matching_archive:
download_link = self.server_url + 'series/' + item.attrs['href']
2021-01-17 13:58:01 +00:00
res = self.session.get(download_link, timeout=30)
2020-07-21 04:28:34 +00:00
res.raise_for_status()
archive = self._get_archive(res.content)
# extract the subtitle
if archive:
subtitles_from_archive = self._get_subtitle_from_archive(archive, video)
for subtitle in subtitles_from_archive:
subtitle.page_link = page_link
subtitle.download_link = download_link
subtitles.append(subtitle)
return subtitles
def query_movies(self, video, title):
subtitles = []
2021-01-17 13:58:01 +00:00
r = self.session.get(self.search_url, params={'q': title}, timeout=30)
2020-07-21 04:28:34 +00:00
r.raise_for_status()
soup = ParserBeautifulSoup(r.content.decode('utf-8', 'ignore'), ['html.parser'])
# loop over movies name
movies_url = []
self.is_perfect_match = False
movies = soup.select('.film > h3 > a')
for item in movies:
# title
if title.lower() in item.text.lower():
movies_url.append(item.attrs['href'])
self.is_perfect_match = True
series_subs_archives_url = []
for movies_page in movies_url:
page_link = self.server_url + movies_page
2021-01-17 13:58:01 +00:00
r = self.session.get(page_link, timeout=30)
2020-07-21 04:28:34 +00:00
r.raise_for_status()
soup = ParserBeautifulSoup(r.content.decode('utf-8', 'ignore'), ['html.parser'])
movies_subs_archives = soup.select('a.subList')
for item in movies_subs_archives:
download_link = self.server_url + 'films/' + item.attrs['href']
2021-01-17 13:58:01 +00:00
res = self.session.get(download_link, timeout=30)
2020-07-21 04:28:34 +00:00
res.raise_for_status()
archive = self._get_archive(res.content)
# extract the subtitle
if archive:
subtitles_from_archive = self._get_subtitle_from_archive(archive, video)
for subtitle in subtitles_from_archive:
subtitle.page_link = page_link
subtitle.download_link = download_link
subtitles.append(subtitle)
return subtitles
def list_subtitles(self, video, languages):
subtitles = []
2020-07-21 04:28:34 +00:00
# query for subtitles
if isinstance(video, Episode):
subtitles += [s for s in self.query_series(video, video.series) if s.language in languages]
else:
subtitles += [s for s in self.query_movies(video, video.title) if s.language in languages]
2020-07-21 04:28:34 +00:00
return subtitles
def download_subtitle(self, subtitle):
return subtitle
def _get_archive(self, content):
# open the archive
archive_stream = io.BytesIO(content)
if is_rarfile(archive_stream):
logger.debug('Sous-Titres.eu: Identified rar archive')
archive = RarFile(archive_stream)
elif is_zipfile(archive_stream):
logger.debug('Sous-Titres.eu: Identified zip archive')
archive = ZipFile(archive_stream)
else:
logger.error('Sous-Titres.eu: Unsupported compressed format')
return None
return archive
def _get_subtitle_from_archive(self, archive, video):
subtitles = []
# some files have a non subtitle with .txt extension
_tmp = list(SUBTITLE_EXTENSIONS)
_tmp.remove('.txt')
_subtitle_extensions = tuple(_tmp)
_scores = get_scores(video)
for name in archive.namelist():
# discard hidden files
if os.path.split(name)[-1].startswith('.'):
continue
# discard non-subtitle files
if not name.lower().endswith(_subtitle_extensions):
continue
# get subtitles language
if '.en.' in name.lower():
language = Language.fromopensubtitles('eng')
else:
language = Language.fromopensubtitles('fre')
release = name[:-4].lower().rstrip('tag').rstrip('en').rstrip('fr')
_guess = guessit(release)
if isinstance(video, Episode):
if video.episode != _guess['episode'] or video.season != _guess['season']:
continue
matches = set()
matches |= guess_matches(video, _guess)
_score = sum((_scores.get(match, 0) for match in matches))
content = archive.read(name)
subtitles.append(SoustitreseuSubtitle(language, video, name, _guess, content, self.is_perfect_match))
return subtitles