Added subdl provider initial implementation

pull/2572/head
morpheus65535 5 months ago committed by GitHub
parent c94de67b44
commit 40985fdee3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

@ -321,6 +321,9 @@ validators = [
Validator('karagarga.f_username', must_exist=True, default='', is_type_of=str, cast=str),
Validator('karagarga.f_password', must_exist=True, default='', is_type_of=str, cast=str),
# subdl section
Validator('subdl.api_key', must_exist=True, default='', is_type_of=str, cast=str),
# subsync section
Validator('subsync.use_subsync', must_exist=True, default=False, is_type_of=bool),
Validator('subsync.use_subsync_threshold', must_exist=True, default=False, is_type_of=bool),

@ -322,6 +322,9 @@ def get_providers_auth():
},
"animetosho": {
'search_threshold': settings.animetosho.search_threshold,
},
"subdl": {
'api_key': settings.subdl.api_key,
}
}

@ -0,0 +1,90 @@
# -*- coding: utf-8 -*-
from __future__ import absolute_import
from babelfish import LanguageReverseConverter
from subliminal.exceptions import ConfigurationError
class SubdlConverter(LanguageReverseConverter):
def __init__(self):
self.from_subdl = {
"AR": ("ara", None, None), # Arabic
"DA": ("dan", None, None), # Danish
"NL": ("nld", None, None), # Dutch
"EN": ("eng", None, None), # English
"FA": ("fas", None, None), # Farsi_Persian
"FI": ("fin", None, None), # Finnish
"FR": ("fra", None, None), # French
"ID": ("ind", None, None), # Indonesian
"IT": ("ita", None, None), # Italian
"NO": ("nor", None, None), # Norwegian
"RO": ("ron", None, None), # Romanian
"ES": ("spa", None, None), # Spanish
"SV": ("swe", None, None), # Swedish
"VI": ("vie", None, None), # Vietnamese
"SQ": ("sqi", None, None), # Albanian
"AZ": ("aze", None, None), # Azerbaijani
"BE": ("bel", None, None), # Belarusian
"BN": ("ben", None, None), # Bengali
"BS": ("bos", None, None), # Bosnian
"BG": ("bul", None, None), # Bulgarian
"MY": ("mya", None, None), # Burmese
"CA": ("cat", None, None), # Catalan
"ZH": ("zho", None, None), # Chinese BG code
"HR": ("hrv", None, None), # Croatian
"CS": ("ces", None, None), # Czech
"EO": ("epo", None, None), # Esperanto
"ET": ("est", None, None), # Estonian
"KA": ("kat", None, None), # Georgian
"DE": ("deu", None, None), # German
"EL": ("ell", None, None), # Greek
"KL": ("kal", None, None), # Greenlandic
"HE": ("heb", None, None), # Hebrew
"HI": ("hin", None, None), # Hindi
"HU": ("hun", None, None), # Hungarian
"IS": ("isl", None, None), # Icelandic
"JA": ("jpn", None, None), # Japanese
"KO": ("kor", None, None), # Korean
"KU": ("kur", None, None), # Kurdish
"LV": ("lav", None, None), # Latvian
"LT": ("lit", None, None), # Lithuanian
"MK": ("mkd", None, None), # Macedonian
"MS": ("msa", None, None), # Malay
"ML": ("mal", None, None), # Malayalam
"PL": ("pol", None, None), # Polish
"PT": ("por", None, None), # Portuguese
"RU": ("rus", None, None), # Russian
"SR": ("srp", None, None), # Serbian
"SI": ("sin", None, None), # Sinhala
"SK": ("slk", None, None), # Slovak
"SL": ("slv", None, None), # Slovenian
"TL": ("tgl", None, None), # Tagalog
"TA": ("tam", None, None), # Tamil
"TE": ("tel", None, None), # Telugu
"TH": ("tha", None, None), # Thai
"TR": ("tur", None, None), # Turkish
"UK": ("ukr", None, None), # Ukrainian
"UR": ("urd", None, None), # Urdu
# custom languages
"BR_PT": ("por", "BR", None), # Brazilian Portuguese
"ZH_BG": ("zho", None, "Hant"), # Big 5 code
# unsupported language in Bazarr
# "BG_EN": "Bulgarian_English",
# "NL_EN": "Dutch_English",
# "EN_DE": "English_German",
# "HU_EN": "Hungarian_English",
# "MNI": "Manipuri",
}
self.to_subdl = {v: k for k, v in self.from_subdl.items()}
self.codes = set(self.from_subdl.keys())
def convert(self, alpha3, country=None, script=None):
if (alpha3, country, script) in self.to_subdl:
return self.to_subdl[(alpha3, country, script)]
raise ConfigurationError('Unsupported language for subdl: %s, %s, %s' % (alpha3, country, script))
def reverse(self, subdl):
if subdl in self.from_subdl:
return self.from_subdl[subdl]
raise ConfigurationError('Unsupported language code for subdl: %s' % subdl)

@ -0,0 +1,278 @@
# -*- coding: utf-8 -*-
import logging
import os
import time
import io
from zipfile import ZipFile, is_zipfile
from urllib.parse import urljoin
from requests import Session
from babelfish import language_converters
from subzero.language import Language
from subliminal import Episode, Movie
from subliminal.exceptions import ConfigurationError, ProviderError, DownloadLimitExceeded
from subliminal_patch.exceptions import APIThrottled
from .mixins import ProviderRetryMixin
from subliminal_patch.subtitle import Subtitle
from subliminal.subtitle import fix_line_ending
from subliminal_patch.providers import Provider
from subliminal_patch.subtitle import guess_matches
from guessit import guessit
logger = logging.getLogger(__name__)
retry_amount = 3
retry_timeout = 5
language_converters.register('subdl = subliminal_patch.converters.subdl:SubdlConverter')
supported_languages = list(language_converters['subdl'].to_subdl.keys())
class SubdlSubtitle(Subtitle):
provider_name = 'subdl'
hash_verifiable = False
hearing_impaired_verifiable = True
def __init__(self, language, forced, hearing_impaired, page_link, download_link, file_id, release_names, uploader,
season=None, episode=None):
super().__init__(language)
language = Language.rebuild(language, hi=hearing_impaired, forced=forced)
self.season = season
self.episode = episode
self.releases = release_names
self.release_info = ', '.join(release_names)
self.language = language
self.forced = forced
self.hearing_impaired = hearing_impaired
self.file_id = file_id
self.page_link = page_link
self.download_link = download_link
self.uploader = uploader
self.matches = None
@property
def id(self):
return self.file_id
def get_matches(self, video):
matches = set()
type_ = "movie" if isinstance(video, Movie) else "episode"
# handle movies and series separately
if isinstance(video, Episode):
# series
matches.add('series')
# season
if video.season == self.season:
matches.add('season')
# episode
if video.episode == self.episode:
matches.add('episode')
# imdb
matches.add('series_imdb_id')
else:
# title
matches.add('title')
# imdb
matches.add('imdb_id')
# other properties
matches |= guess_matches(video, guessit(self.release_info, {"type": type_}))
self.matches = matches
return matches
class SubdlProvider(ProviderRetryMixin, Provider):
"""Subdl Provider"""
server_hostname = 'api.subdl.com'
languages = {Language(*lang) for lang in supported_languages}
languages.update(set(Language.rebuild(lang, forced=True) for lang in languages))
languages.update(set(Language.rebuild(l, hi=True) for l in languages))
video_types = (Episode, Movie)
def __init__(self, api_key=None):
if not api_key:
raise ConfigurationError('Api_key must be specified')
self.session = Session()
self.session.headers = {'User-Agent': os.environ.get("SZ_USER_AGENT", "Sub-Zero/2")}
self.api_key = api_key
self.video = None
self._started = None
def initialize(self):
self._started = time.time()
def terminate(self):
self.session.close()
def server_url(self):
return f'https://{self.server_hostname}/api/v1/'
def query(self, languages, video):
self.video = video
if isinstance(self.video, Episode):
title = self.video.series
else:
title = self.video.title
imdb_id = None
if isinstance(self.video, Episode) and self.video.series_imdb_id:
imdb_id = self.video.series_imdb_id
elif isinstance(self.video, Movie) and self.video.imdb_id:
imdb_id = self.video.imdb_id
# be sure to remove duplicates using list(set())
langs_list = sorted(list(set([lang.basename.upper() for lang in languages])))
langs = ','.join(langs_list)
logger.debug(f'Searching for those languages: {langs}')
# query the server
if isinstance(self.video, Episode):
res = self.retry(
lambda: self.session.get(self.server_url() + 'subtitles',
params=(('api_key', self.api_key),
('episode_number', self.video.episode),
('film_name', title if not imdb_id else None),
('imdb_id', imdb_id if imdb_id else None),
('languages', langs),
('season_number', self.video.season),
('subs_per_page', 30),
('type', 'tv'),
('comment', 1),
('releases', 1)),
timeout=30),
amount=retry_amount,
retry_timeout=retry_timeout
)
else:
res = self.retry(
lambda: self.session.get(self.server_url() + 'subtitles',
params=(('api_key', self.api_key),
('film_name', title if not imdb_id else None),
('imdb_id', imdb_id if imdb_id else None),
('languages', langs),
('subs_per_page', 30),
('type', 'movie'),
('comment', 1),
('releases', 1)),
timeout=30),
amount=retry_amount,
retry_timeout=retry_timeout
)
if res.status_code == 429:
raise APIThrottled("Too many requests")
elif res.status_code == 403:
raise ConfigurationError("Invalid API key")
elif res.status_code != 200:
res.raise_for_status()
subtitles = []
result = res.json()
if ('success' in result and not result['success']) or ('status' in result and not result['status']):
raise ProviderError(result['error'])
logger.debug(f"Query returned {len(result['subtitles'])} subtitles")
if len(result['subtitles']):
for item in result['subtitles']:
if item.get('episode_from', False) == item.get('episode_end', False): # ignore season packs
subtitle = SubdlSubtitle(
language=Language.fromsubdl(item['language']),
forced=self._is_forced(item),
hearing_impaired=item.get('hi', False) or self._is_hi(item),
page_link=urljoin("https://subdl.com", item.get('subtitlePage', '')),
download_link=item['url'],
file_id=item['name'],
release_names=item.get('releases', []),
uploader=item.get('author', ''),
season=item.get('season', None),
episode=item.get('episode', None),
)
subtitle.get_matches(self.video)
if subtitle.language in languages: # make sure only desired subtitles variants are returned
subtitles.append(subtitle)
return subtitles
@staticmethod
def _is_hi(item):
# Comments include specific mention of removed or non HI
non_hi_tag = ['hi remove', 'non hi', 'nonhi', 'non-hi', 'non-sdh', 'non sdh', 'nonsdh', 'sdh remove']
for tag in non_hi_tag:
if tag in item.get('comment', '').lower():
return False
# Archive filename include _HI_
if '_hi_' in item.get('name', '').lower():
return True
# Comments or release names include some specific strings
hi_keys = [item.get('comment', '').lower(), [x.lower() for x in item.get('releases', [])]]
hi_tag = ['_hi_', ' hi ', '.hi.', 'hi ', ' hi', 'sdh', '𝓢𝓓𝓗']
for key in hi_keys:
if any(x in key for x in hi_tag):
return True
# nothing match so we consider it as non-HI
return False
@staticmethod
def _is_forced(item):
# Comments include specific mention of forced subtitles
forced_tags = ['forced', 'foreign']
for tag in forced_tags:
if tag in item.get('comment', '').lower():
return True
# nothing match so we consider it as normal subtitles
return False
def list_subtitles(self, video, languages):
return self.query(languages, video)
def download_subtitle(self, subtitle):
logger.debug('Downloading subtitle %r', subtitle)
download_link = urljoin("https://dl.subdl.com", subtitle.download_link)
r = self.retry(
lambda: self.session.get(download_link, timeout=30),
amount=retry_amount,
retry_timeout=retry_timeout
)
if r.status_code == 429:
raise DownloadLimitExceeded("Daily download limit exceeded")
elif r.status_code == 403:
raise ConfigurationError("Invalid API key")
elif r.status_code != 200:
r.raise_for_status()
if not r:
logger.error(f'Could not download subtitle from {download_link}')
subtitle.content = None
return
else:
archive_stream = io.BytesIO(r.content)
if is_zipfile(archive_stream):
archive = ZipFile(archive_stream)
for name in archive.namelist():
# TODO when possible, deal with season pack / multiple files archive
subtitle_content = archive.read(name)
subtitle.content = fix_line_ending(subtitle_content)
return
else:
logger.error(f'Could not unzip subtitle from {download_link}')
subtitle.content = None
return

@ -371,6 +371,15 @@ export const ProviderList: Readonly<ProviderInfo[]> = [
description: "Mostly French Subtitles Provider",
},
{ key: "subdivx", description: "LATAM Spanish / Spanish Subtitles Provider" },
{
key: "subdl",
inputs: [
{
type: "text",
key: "api_key",
},
],
},
{
key: "subf2m",
name: "subf2m.co",

Loading…
Cancel
Save