Improved support for series for Titrari.ro provider

This commit is contained in:
morpheus65535 2021-10-11 23:44:06 -04:00
parent 294d3eb0ee
commit b1af4caeeb
1 changed files with 63 additions and 25 deletions

View File

@ -43,11 +43,13 @@ logger = logging.getLogger(__name__)
# Configure :mod:`rarfile` to use the same path separator as :mod:`zipfile`
rarfile.PATH_SEP = '/'
class TitrariSubtitle(Subtitle):
provider_name = 'titrari'
def __init__(self, language, download_link, sid, releases, title, imdb_id, year=None, download_count=None, comments=None):
def __init__(self, language, download_link, sid, releases, title, imdb_id, year=None, download_count=None,
comments=None, is_episode=False, desired_episode=None):
super(TitrariSubtitle, self).__init__(language)
self.sid = sid
self.title = title
@ -57,6 +59,9 @@ class TitrariSubtitle(Subtitle):
self.download_count = download_count
self.releases = self.release_info = releases
self.comments = comments
self.matches = None
self.is_episode = is_episode
self.desired_episode = desired_episode
@property
def id(self):
@ -71,38 +76,43 @@ class TitrariSubtitle(Subtitle):
def get_matches(self, video):
matches = set()
if video.year and self.year == video.year:
matches.add('year')
if video.release_group and video.release_group in self.comments:
matches.add('release_group')
if isinstance(video, Movie):
# title
if video.title and sanitize(self.title) == fix_inconsistent_naming(video.title):
matches.add('title')
if video.year and self.year == video.year:
matches.add('year')
# imdb
if video.imdb_id and self.imdb_id == video.imdb_id:
matches.add('imdb_id')
if video.release_group and video.release_group in self.comments:
matches.add('release_group')
# guess match others
matches |= guess_matches(video, guessit(self.comments, {"type": "movie"}))
else:
# title
if video.series and fix_inconsistent_naming(video.series) in sanitize(self.title):
seasonless_title = re.sub(r'\s-\sSezonul\s\d+$', '', self.title.rstrip())
if video.series and fix_inconsistent_naming(video.series) == sanitize(seasonless_title):
matches.add('series')
if video.year and self.year == video.year:
matches.add('year')
# imdb
if video.series_imdb_id and self.imdb_id == video.series_imdb_id:
matches.add('imdb_id')
if video.release_group and video.release_group in self.comments:
matches.add('release_group')
# season
if f"Sezonul {video.season}" in self.title:
matches.add('season')
# episode
if {"imdb_id", "season"}.issubset(matches):
matches.add('episode')
# guess match others
matches |= guess_matches(video, guessit(self.comments, {"type": "episode"}))
self.matches = matches
@ -112,8 +122,8 @@ class TitrariSubtitle(Subtitle):
class TitrariProvider(Provider, ProviderSubtitleArchiveMixin):
subtitle_class = TitrariSubtitle
languages = {Language(l) for l in ['ron', 'eng']}
languages.update(set(Language.rebuild(l, forced=True) for l in languages))
languages = {Language(lang) for lang in ['ron', 'eng']}
languages.update(set(Language.rebuild(lang, forced=True) for lang in languages))
api_url = 'https://www.titrari.ro/'
query_advanced_search = 'cautarepreaavansata'
@ -123,7 +133,8 @@ class TitrariProvider(Provider, ProviderSubtitleArchiveMixin):
def initialize(self):
self.session = Session()
# Hardcoding the UA to bypass the 30s throttle that titrari.ro uses for IP/UA pair
self.session.headers['User-Agent'] = 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/93.0.4535.2 Safari/537.36'
self.session.headers['User-Agent'] = 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, ' \
'like Gecko) Chrome/93.0.4535.2 Safari/537.36'
# self.session.headers['User-Agent'] = AGENT_LIST[randint(0, len(AGENT_LIST) - 1)]
def terminate(self):
@ -154,13 +165,14 @@ class TitrariProvider(Provider, ProviderSubtitleArchiveMixin):
fullTitle = row.parent.find("h1").find("a").text
#Get title
# Get title
try:
title = fullTitle.split("(")[0]
except:
logger.error("[#### Provider: titrari.ro] Error parsing title.")
# Get downloads count
downloads = 0
try:
downloads = int(row.parent.parent.select("span")[index].text[12:])
except:
@ -176,25 +188,31 @@ class TitrariProvider(Provider, ProviderSubtitleArchiveMixin):
# Get imdbId
sub_imdb_id = self.getImdbIdFromSubtitle(row)
comments = ''
try:
comments = row.parent.parent.find_all("td", class_=re.compile("comment"))[index*2+1].text
except:
logger.error("Error parsing comments.")
subtitle = self.subtitle_class(next(iter(languages)), download_link, index, None, title, sub_imdb_id, year, downloads, comments)
episode_number = video.episode if isinstance(video, Episode) else None
subtitle = self.subtitle_class(next(iter(languages)), download_link, index, None, title, sub_imdb_id,
year, downloads, comments, isinstance(video, Episode), episode_number)
logger.debug('[#### Provider: titrari.ro] Found subtitle %r', str(subtitle))
subtitles.append(subtitle)
ordered_subs = self.order(subtitles, video)
ordered_subs = self.order(subtitles)
return ordered_subs
def order(self, subtitles, video):
@staticmethod
def order(subtitles):
logger.debug("[#### Provider: titrari.ro] Sorting by download count...")
sorted_subs = sorted(subtitles, key=lambda s: s.download_count, reverse=True)
return sorted_subs
def getImdbIdFromSubtitle(self, row):
@staticmethod
def getImdbIdFromSubtitle(row):
imdbId = None
try:
imdbId = row.parent.parent.find_all(src=re.compile("imdb"))[0].parent.get('href').split("tt")[-1]
except:
@ -245,8 +263,9 @@ class TitrariProvider(Provider, ProviderSubtitleArchiveMixin):
except:
logger.error("[#### Provider: titrari.ro] Error parsing video.imdb_id.")
return [s for s in
self.query(languages, title, imdb_id, video)]
subtitles = [s for s in
self.query(languages, title, imdb_id, video)]
return subtitles
def download_subtitle(self, subtitle):
r = self.session.get(subtitle.download_link, headers={'Referer': self.api_url}, timeout=10)
@ -269,7 +288,26 @@ class TitrariProvider(Provider, ProviderSubtitleArchiveMixin):
raise ProviderError('[#### Provider: titrari.ro] Unidentified archive type')
subtitle.releases = _get_releases_from_archive(archive)
subtitle.content = self.get_subtitle_from_archive(subtitle, archive)
if subtitle.is_episode:
subtitle.content = self._get_subtitle_from_archive(subtitle, archive)
else:
subtitle.content = self.get_subtitle_from_archive(subtitle, archive)
def _get_subtitle_from_archive(self, subtitle, archive):
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
_guess = guessit(name)
if subtitle.desired_episode == _guess['episode']:
return archive.read(name)
return None
def _get_releases_from_archive(archive):