mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-07 09:20:36 +00:00
ef8fb7f029
Closes #6975 Authored by: bashonly, Grub4K Co-authored-by: Simon Sawicki <contact@grub4k.xyz>
301 lines
12 KiB
Python
301 lines
12 KiB
Python
import base64
|
|
import binascii
|
|
import json
|
|
import time
|
|
import uuid
|
|
|
|
from .common import InfoExtractor
|
|
from ..dependencies import Cryptodome
|
|
from ..utils import (
|
|
ExtractorError,
|
|
int_or_none,
|
|
jwt_decode_hs256,
|
|
traverse_obj,
|
|
try_call,
|
|
url_or_none,
|
|
urlencode_postdata,
|
|
)
|
|
|
|
|
|
class WrestleUniverseBaseIE(InfoExtractor):
|
|
_NETRC_MACHINE = 'wrestleuniverse'
|
|
_VALID_URL_TMPL = r'https?://(?:www\.)?wrestle-universe\.com/(?:(?P<lang>\w{2})/)?%s/(?P<id>\w+)'
|
|
_API_PATH = None
|
|
_REAL_TOKEN = None
|
|
_TOKEN_EXPIRY = None
|
|
_REFRESH_TOKEN = None
|
|
_DEVICE_ID = None
|
|
_LOGIN_QUERY = {'key': 'AIzaSyCaRPBsDQYVDUWWBXjsTrHESi2r_F3RAdA'}
|
|
_LOGIN_HEADERS = {
|
|
'Accept': '*/*',
|
|
'Content-Type': 'application/json',
|
|
'X-Client-Version': 'Chrome/JsCore/9.9.4/FirebaseCore-web',
|
|
'X-Firebase-gmpid': '1:307308870738:web:820f38fe5150c8976e338b',
|
|
'Referer': 'https://www.wrestle-universe.com/',
|
|
'Origin': 'https://www.wrestle-universe.com',
|
|
}
|
|
|
|
@property
|
|
def _TOKEN(self):
|
|
if not self._REAL_TOKEN or not self._TOKEN_EXPIRY:
|
|
token = try_call(lambda: self._get_cookies('https://www.wrestle-universe.com/')['token'].value)
|
|
if not token and not self._REFRESH_TOKEN:
|
|
self.raise_login_required()
|
|
self._REAL_TOKEN = token
|
|
|
|
if not self._REAL_TOKEN or self._TOKEN_EXPIRY <= int(time.time()):
|
|
if not self._REFRESH_TOKEN:
|
|
raise ExtractorError(
|
|
'Expired token. Refresh your cookies in browser and try again', expected=True)
|
|
self._refresh_token()
|
|
|
|
return self._REAL_TOKEN
|
|
|
|
@_TOKEN.setter
|
|
def _TOKEN(self, value):
|
|
self._REAL_TOKEN = value
|
|
|
|
expiry = traverse_obj(value, ({jwt_decode_hs256}, 'exp', {int_or_none}))
|
|
if not expiry:
|
|
raise ExtractorError('There was a problem with the auth token')
|
|
self._TOKEN_EXPIRY = expiry
|
|
|
|
def _perform_login(self, username, password):
|
|
login = self._download_json(
|
|
'https://identitytoolkit.googleapis.com/v1/accounts:signInWithPassword', None,
|
|
'Logging in', query=self._LOGIN_QUERY, headers=self._LOGIN_HEADERS, data=json.dumps({
|
|
'returnSecureToken': True,
|
|
'email': username,
|
|
'password': password,
|
|
}, separators=(',', ':')).encode())
|
|
self._REFRESH_TOKEN = traverse_obj(login, ('refreshToken', {str}))
|
|
if not self._REFRESH_TOKEN:
|
|
self.report_warning('No refresh token was granted')
|
|
self._TOKEN = traverse_obj(login, ('idToken', {str}))
|
|
|
|
def _real_initialize(self):
|
|
if WrestleUniverseBaseIE._DEVICE_ID:
|
|
return
|
|
|
|
WrestleUniverseBaseIE._DEVICE_ID = self._configuration_arg('device_id', [None], ie_key='WrestleUniverse')[0]
|
|
if not WrestleUniverseBaseIE._DEVICE_ID:
|
|
WrestleUniverseBaseIE._DEVICE_ID = self.cache.load(self._NETRC_MACHINE, 'device_id')
|
|
if WrestleUniverseBaseIE._DEVICE_ID:
|
|
return
|
|
WrestleUniverseBaseIE._DEVICE_ID = str(uuid.uuid4())
|
|
|
|
self.cache.store(self._NETRC_MACHINE, 'device_id', WrestleUniverseBaseIE._DEVICE_ID)
|
|
|
|
def _refresh_token(self):
|
|
refresh = self._download_json(
|
|
'https://securetoken.googleapis.com/v1/token', None, 'Refreshing token',
|
|
query=self._LOGIN_QUERY, data=urlencode_postdata({
|
|
'grant_type': 'refresh_token',
|
|
'refresh_token': self._REFRESH_TOKEN,
|
|
}), headers={
|
|
**self._LOGIN_HEADERS,
|
|
'Content-Type': 'application/x-www-form-urlencoded',
|
|
})
|
|
if traverse_obj(refresh, ('refresh_token', {str})):
|
|
self._REFRESH_TOKEN = refresh['refresh_token']
|
|
token = traverse_obj(refresh, 'access_token', 'id_token', expected_type=str)
|
|
if not token:
|
|
raise ExtractorError('No auth token returned from refresh request')
|
|
self._TOKEN = token
|
|
|
|
def _call_api(self, video_id, param='', msg='API', auth=True, data=None, query={}, fatal=True):
|
|
headers = {'CA-CID': ''}
|
|
if data:
|
|
headers['Content-Type'] = 'application/json;charset=utf-8'
|
|
data = json.dumps(data, separators=(',', ':')).encode()
|
|
if auth:
|
|
headers['Authorization'] = f'Bearer {self._TOKEN}'
|
|
return self._download_json(
|
|
f'https://api.wrestle-universe.com/v1/{self._API_PATH}/{video_id}{param}', video_id,
|
|
note=f'Downloading {msg} JSON', errnote=f'Failed to download {msg} JSON',
|
|
data=data, headers=headers, query=query, fatal=fatal)
|
|
|
|
def _call_encrypted_api(self, video_id, param='', msg='API', data={}, query={}, fatal=True):
|
|
if not Cryptodome.RSA:
|
|
raise ExtractorError('pycryptodomex not found. Please install', expected=True)
|
|
private_key = Cryptodome.RSA.generate(2048)
|
|
cipher = Cryptodome.PKCS1_OAEP.new(private_key, hashAlgo=Cryptodome.SHA1)
|
|
|
|
def decrypt(data):
|
|
if not data:
|
|
return None
|
|
try:
|
|
return cipher.decrypt(base64.b64decode(data)).decode()
|
|
except (ValueError, binascii.Error) as e:
|
|
raise ExtractorError(f'Could not decrypt data: {e}')
|
|
|
|
token = base64.b64encode(private_key.public_key().export_key('DER')).decode()
|
|
api_json = self._call_api(video_id, param, msg, data={
|
|
'deviceId': self._DEVICE_ID,
|
|
'token': token,
|
|
**data,
|
|
}, query=query, fatal=fatal)
|
|
return api_json, decrypt
|
|
|
|
def _download_metadata(self, url, video_id, lang, props_key):
|
|
metadata = self._call_api(video_id, msg='metadata', query={'al': lang or 'ja'}, auth=False, fatal=False)
|
|
if not metadata:
|
|
webpage = self._download_webpage(url, video_id)
|
|
nextjs_data = self._search_nextjs_data(webpage, video_id)
|
|
metadata = traverse_obj(nextjs_data, ('props', 'pageProps', props_key, {dict})) or {}
|
|
return metadata
|
|
|
|
def _get_formats(self, data, path, video_id=None):
|
|
hls_url = traverse_obj(data, path, get_all=False)
|
|
if not hls_url and not data.get('canWatch'):
|
|
self.raise_no_formats(
|
|
'This account does not have access to the requested content', expected=True)
|
|
elif not hls_url:
|
|
self.raise_no_formats('No supported formats found')
|
|
return self._extract_m3u8_formats(hls_url, video_id, 'mp4', m3u8_id='hls', live=True)
|
|
|
|
|
|
class WrestleUniverseVODIE(WrestleUniverseBaseIE):
|
|
_VALID_URL = WrestleUniverseBaseIE._VALID_URL_TMPL % 'videos'
|
|
_TESTS = [{
|
|
'url': 'https://www.wrestle-universe.com/en/videos/dp8mpjmcKfxzUhEHM2uFws',
|
|
'info_dict': {
|
|
'id': 'dp8mpjmcKfxzUhEHM2uFws',
|
|
'ext': 'mp4',
|
|
'title': 'The 3rd “Futari wa Princess” Max Heart Tournament',
|
|
'description': 'md5:318d5061e944797fbbb81d5c7dd00bf5',
|
|
'location': '埼玉・春日部ふれあいキューブ',
|
|
'channel': 'tjpw',
|
|
'duration': 7119,
|
|
'timestamp': 1674979200,
|
|
'upload_date': '20230129',
|
|
'thumbnail': 'https://image.asset.wrestle-universe.com/8FjD67P8rZc446RBQs5RBN/8FjD67P8rZc446RBQs5RBN',
|
|
'chapters': 'count:7',
|
|
'cast': 'count:21',
|
|
},
|
|
'params': {
|
|
'skip_download': 'm3u8',
|
|
},
|
|
}]
|
|
|
|
_API_PATH = 'videoEpisodes'
|
|
|
|
def _real_extract(self, url):
|
|
lang, video_id = self._match_valid_url(url).group('lang', 'id')
|
|
metadata = self._download_metadata(url, video_id, lang, 'videoEpisodeFallbackData')
|
|
video_data = self._call_api(video_id, ':watch', 'watch', data={
|
|
# 'deviceId' is required if ignoreDeviceRestriction is False
|
|
'ignoreDeviceRestriction': True,
|
|
})
|
|
|
|
return {
|
|
'id': video_id,
|
|
'formats': self._get_formats(video_data, (
|
|
(('protocolHls', 'url'), ('chromecastUrls', ...)), {url_or_none}), video_id),
|
|
**traverse_obj(metadata, {
|
|
'title': ('displayName', {str}),
|
|
'description': ('description', {str}),
|
|
'channel': ('labels', 'group', {str}),
|
|
'location': ('labels', 'venue', {str}),
|
|
'timestamp': ('watchStartTime', {int_or_none}),
|
|
'thumbnail': ('keyVisualUrl', {url_or_none}),
|
|
'cast': ('casts', ..., 'displayName', {str}),
|
|
'duration': ('duration', {int}),
|
|
'chapters': ('videoChapters', lambda _, v: isinstance(v.get('start'), int), {
|
|
'title': ('displayName', {str}),
|
|
'start_time': ('start', {int}),
|
|
'end_time': ('end', {int}),
|
|
}),
|
|
}),
|
|
}
|
|
|
|
|
|
class WrestleUniversePPVIE(WrestleUniverseBaseIE):
|
|
_VALID_URL = WrestleUniverseBaseIE._VALID_URL_TMPL % 'lives'
|
|
_TESTS = [{
|
|
'note': 'HLS AES-128 key obtained via API',
|
|
'url': 'https://www.wrestle-universe.com/en/lives/buH9ibbfhdJAY4GKZcEuJX',
|
|
'info_dict': {
|
|
'id': 'buH9ibbfhdJAY4GKZcEuJX',
|
|
'ext': 'mp4',
|
|
'title': '【PPV】Beyond the origins, into the future',
|
|
'description': 'md5:9a872db68cd09be4a1e35a3ee8b0bdfc',
|
|
'channel': 'tjpw',
|
|
'location': '東京・Twin Box AKIHABARA',
|
|
'duration': 10098,
|
|
'timestamp': 1675076400,
|
|
'upload_date': '20230130',
|
|
'thumbnail': 'https://image.asset.wrestle-universe.com/rJs2m7cBaLXrwCcxMdQGRM/rJs2m7cBaLXrwCcxMdQGRM',
|
|
'thumbnails': 'count:3',
|
|
'hls_aes': {
|
|
'key': '5633184acd6e43f1f1ac71c6447a4186',
|
|
'iv': '5bac71beb33197d5600337ce86de7862',
|
|
},
|
|
},
|
|
'params': {
|
|
'skip_download': 'm3u8',
|
|
},
|
|
'skip': 'No longer available',
|
|
}, {
|
|
'note': 'unencrypted HLS',
|
|
'url': 'https://www.wrestle-universe.com/en/lives/wUG8hP5iApC63jbtQzhVVx',
|
|
'info_dict': {
|
|
'id': 'wUG8hP5iApC63jbtQzhVVx',
|
|
'ext': 'mp4',
|
|
'title': 'GRAND PRINCESS \'22',
|
|
'description': 'md5:e4f43d0d4262de3952ff34831bc99858',
|
|
'channel': 'tjpw',
|
|
'location': '東京・両国国技館',
|
|
'duration': 18044,
|
|
'timestamp': 1647665400,
|
|
'upload_date': '20220319',
|
|
'thumbnail': 'https://image.asset.wrestle-universe.com/i8jxSTCHPfdAKD4zN41Psx/i8jxSTCHPfdAKD4zN41Psx',
|
|
'thumbnails': 'count:3',
|
|
},
|
|
'params': {
|
|
'skip_download': 'm3u8',
|
|
},
|
|
}]
|
|
|
|
_API_PATH = 'events'
|
|
|
|
def _real_extract(self, url):
|
|
lang, video_id = self._match_valid_url(url).group('lang', 'id')
|
|
metadata = self._download_metadata(url, video_id, lang, 'eventFallbackData')
|
|
|
|
info = {
|
|
'id': video_id,
|
|
**traverse_obj(metadata, {
|
|
'title': ('displayName', {str}),
|
|
'description': ('description', {str}),
|
|
'channel': ('labels', 'group', {str}),
|
|
'location': ('labels', 'venue', {str}),
|
|
'timestamp': ('startTime', {int_or_none}),
|
|
'thumbnails': (('keyVisualUrl', 'alterKeyVisualUrl', 'heroKeyVisualUrl'), {'url': {url_or_none}}),
|
|
}),
|
|
}
|
|
|
|
ended_time = traverse_obj(metadata, ('endedTime', {int_or_none}))
|
|
if info.get('timestamp') and ended_time:
|
|
info['duration'] = ended_time - info['timestamp']
|
|
|
|
video_data, decrypt = self._call_encrypted_api(
|
|
video_id, ':watchArchive', 'watch archive', data={'method': 1})
|
|
info['formats'] = self._get_formats(video_data, (
|
|
('hls', None), ('urls', 'chromecastUrls'), ..., {url_or_none}), video_id)
|
|
for f in info['formats']:
|
|
# bitrates are exaggerated in PPV playlists, so avoid wrong/huge filesize_approx values
|
|
if f.get('tbr'):
|
|
f['tbr'] = int(f['tbr'] / 2.5)
|
|
|
|
hls_aes_key = traverse_obj(video_data, ('hls', 'key', {decrypt}))
|
|
if hls_aes_key:
|
|
info['hls_aes'] = {
|
|
'key': hls_aes_key,
|
|
'iv': traverse_obj(video_data, ('hls', 'iv', {decrypt})),
|
|
},
|
|
elif traverse_obj(video_data, ('hls', 'encryptType', {int})):
|
|
self.report_warning('HLS AES-128 key was not found in API response')
|
|
|
|
return info
|