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.
142 lines
5.7 KiB
142 lines
5.7 KiB
# coding=utf-8
|
|
from __future__ import absolute_import
|
|
import logging
|
|
|
|
from subliminal.providers.napiprojekt import NapiProjektProvider as _NapiProjektProvider, \
|
|
NapiProjektSubtitle as _NapiProjektSubtitle, get_subhash
|
|
from subzero.language import Language
|
|
from subliminal.subtitle import guess_matches
|
|
from subliminal.video import Episode, Movie
|
|
from subliminal_patch.utils import fix_inconsistent_naming as _fix_inconsistent_naming
|
|
from bs4 import BeautifulSoup
|
|
from guessit import guessit
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
def fix_inconsistent_naming(title):
|
|
return _fix_inconsistent_naming(title, {}, True)
|
|
|
|
|
|
class NapiProjektSubtitle(_NapiProjektSubtitle):
|
|
def __init__(self, language, hash, release_info, matches=None):
|
|
super(NapiProjektSubtitle, self).__init__(language, hash)
|
|
self.release_info = release_info
|
|
self.matches = matches
|
|
|
|
def __repr__(self):
|
|
return '<%s %r [%s]>' % (
|
|
self.__class__.__name__, self.release_info, self.language)
|
|
|
|
def get_matches(self, video):
|
|
matches = super().get_matches(video)
|
|
if self.matches is not None:
|
|
matches |= self.matches
|
|
return matches
|
|
|
|
|
|
class NapiProjektProvider(_NapiProjektProvider):
|
|
languages = {Language.fromalpha2(l) for l in ['pl']}
|
|
video_types = (Episode, Movie)
|
|
subtitle_class = NapiProjektSubtitle
|
|
|
|
def query(self, language, hash):
|
|
params = {
|
|
'v': 'dreambox',
|
|
'kolejka': 'false',
|
|
'nick': '',
|
|
'pass': '',
|
|
'napios': 'Linux',
|
|
'l': language.alpha2.upper(),
|
|
'f': hash,
|
|
't': get_subhash(hash)}
|
|
logger.info('Searching subtitle %r', params)
|
|
r = self.session.get(self.server_url, params=params, timeout=10)
|
|
r.raise_for_status()
|
|
|
|
# handle subtitles not found and errors
|
|
if r.content[:4] == b'NPc0':
|
|
logger.debug('No subtitles found')
|
|
return None
|
|
|
|
subtitle = self.subtitle_class(language, hash, release_info=hash)
|
|
subtitle.content = r.content
|
|
logger.debug('Found subtitle %r', subtitle)
|
|
|
|
return subtitle
|
|
|
|
def list_subtitles(self, video, languages):
|
|
def flatten(l):
|
|
return [item for sublist in l for item in sublist]
|
|
return [s for s in [self.query(l, video.hashes['napiprojekt']) for l in languages] if s is not None] + \
|
|
flatten([self._scrape(video, l) for l in languages])
|
|
|
|
def download_subtitle(self, subtitle):
|
|
if subtitle.content is not None:
|
|
return
|
|
subtitle.content = self.query(subtitle.language, subtitle.hash).content
|
|
|
|
def _scrape(self, video, language):
|
|
if language.alpha2 != 'pl':
|
|
return []
|
|
title, matches = self._find_title(video)
|
|
if title == None:
|
|
return []
|
|
episode = f'-s{video.season:02d}e{video.episode:02d}' if isinstance(
|
|
video, Episode) else ''
|
|
response = self.session.get(
|
|
f'https://www.napiprojekt.pl/napisy1,7,0-dla-{title}{episode}')
|
|
response.raise_for_status()
|
|
soup = BeautifulSoup(response.content, 'html.parser')
|
|
subtitles = []
|
|
for link in soup.find_all('a'):
|
|
if 'class' in link.attrs and 'tableA' in link.attrs['class']:
|
|
hash = link.attrs['href'][len('napiprojekt:'):]
|
|
subtitles.append(
|
|
NapiProjektSubtitle(language,
|
|
hash,
|
|
release_info=str(link.contents[0]),
|
|
matches=matches | ({'season', 'episode'} if episode else set())))
|
|
|
|
logger.debug(f'Found subtitles {subtitles}')
|
|
return subtitles
|
|
|
|
def _find_title(self, video):
|
|
search_response = self.session.post('https://www.napiprojekt.pl/ajax/search_catalog.php', {
|
|
'queryString': video.series if isinstance(video, Episode) else video.title,
|
|
'queryKind': 1 if isinstance(video, Episode) else 2,
|
|
'queryYear': str(video.year) if video.year is not None else '',
|
|
'associate': '',
|
|
})
|
|
search_response.raise_for_status()
|
|
soup = BeautifulSoup(search_response.content, 'html.parser')
|
|
imdb_id = video.series_imdb_id if isinstance(
|
|
video, Episode) else video.imdb_id
|
|
|
|
def match_title_tag(
|
|
tag): return tag.name == 'a' and 'class' in tag.attrs and 'movieTitleCat' in tag.attrs['class'] and 'href' in tag.attrs
|
|
|
|
if imdb_id:
|
|
for entry in soup.find_all(lambda tag: tag.name == 'div' and 'greyBoxCatcher' in tag['class']):
|
|
if entry.find_all(href=lambda href: href and href.startswith(f'https://www.imdb.com/title/{imdb_id}')):
|
|
for link in entry.find_all(match_title_tag):
|
|
return link.attrs['href'][len('napisy-'):], \
|
|
{'series', 'year', 'series_imdb_id'} if isinstance(
|
|
video, Episode) else {'title', 'year', 'imdb_id'}
|
|
|
|
type = 'episode' if isinstance(video, Episode) else 'movie'
|
|
for link in soup.find_all(match_title_tag):
|
|
title = fix_inconsistent_naming(str(link.contents[0].string))
|
|
matches = guess_matches(video, guessit(title, {'type': type}))
|
|
if video.year:
|
|
matches |= {'year'}
|
|
if isinstance(video, Episode):
|
|
if title == fix_inconsistent_naming(video.series):
|
|
matches |= {'series'}
|
|
else:
|
|
if title == fix_inconsistent_naming(video.title):
|
|
matches |= {'title'}
|
|
return link.attrs['href'][len('napisy-'):], matches
|
|
|
|
return None, None
|