mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-03 14:15:44 -05:00
[ie/crunchyroll] Remove initial state extraction (#7632)
Authored by: Grub4K
This commit is contained in:
parent
65cfa2b057
commit
9b16762f48
1 changed files with 43 additions and 30 deletions
|
@ -27,11 +27,24 @@ class CrunchyrollBaseIE(InfoExtractor):
|
||||||
_AUTH_HEADERS = None
|
_AUTH_HEADERS = None
|
||||||
_API_ENDPOINT = None
|
_API_ENDPOINT = None
|
||||||
_BASIC_AUTH = None
|
_BASIC_AUTH = None
|
||||||
_QUERY = {}
|
_CLIENT_ID = ('cr_web', 'noaihdevm_6iyg0a8l0q')
|
||||||
|
_LOCALE_LOOKUP = {
|
||||||
|
'ar': 'ar-SA',
|
||||||
|
'de': 'de-DE',
|
||||||
|
'': 'en-US',
|
||||||
|
'es': 'es-419',
|
||||||
|
'es-es': 'es-ES',
|
||||||
|
'fr': 'fr-FR',
|
||||||
|
'it': 'it-IT',
|
||||||
|
'pt-br': 'pt-BR',
|
||||||
|
'pt-pt': 'pt-PT',
|
||||||
|
'ru': 'ru-RU',
|
||||||
|
'hi': 'hi-IN',
|
||||||
|
}
|
||||||
|
|
||||||
@property
|
@property
|
||||||
def is_logged_in(self):
|
def is_logged_in(self):
|
||||||
return self._get_cookies(self._BASE_URL).get('etp_rt')
|
return bool(self._get_cookies(self._BASE_URL).get('etp_rt'))
|
||||||
|
|
||||||
def _perform_login(self, username, password):
|
def _perform_login(self, username, password):
|
||||||
if self.is_logged_in:
|
if self.is_logged_in:
|
||||||
|
@ -62,49 +75,49 @@ def _perform_login(self, username, password):
|
||||||
if not self.is_logged_in:
|
if not self.is_logged_in:
|
||||||
raise ExtractorError('Login succeeded but did not set etp_rt cookie')
|
raise ExtractorError('Login succeeded but did not set etp_rt cookie')
|
||||||
|
|
||||||
def _update_query(self, lang):
|
|
||||||
if lang in CrunchyrollBaseIE._QUERY:
|
|
||||||
return
|
|
||||||
|
|
||||||
webpage = self._download_webpage(
|
|
||||||
f'{self._BASE_URL}/{lang}', None, note=f'Retrieving main page (lang={lang or None})')
|
|
||||||
|
|
||||||
initial_state = self._search_json(r'__INITIAL_STATE__\s*=', webpage, 'initial state', None)
|
|
||||||
CrunchyrollBaseIE._QUERY[lang] = traverse_obj(initial_state, {
|
|
||||||
'locale': ('localization', 'locale'),
|
|
||||||
}) or None
|
|
||||||
|
|
||||||
if CrunchyrollBaseIE._BASIC_AUTH:
|
|
||||||
return
|
|
||||||
|
|
||||||
app_config = self._search_json(r'__APP_CONFIG__\s*=', webpage, 'app config', None)
|
|
||||||
cx_api_param = app_config['cxApiParams']['accountAuthClientId' if self.is_logged_in else 'anonClientId']
|
|
||||||
self.write_debug(f'Using cxApiParam={cx_api_param}')
|
|
||||||
CrunchyrollBaseIE._BASIC_AUTH = 'Basic ' + base64.b64encode(f'{cx_api_param}:'.encode()).decode()
|
|
||||||
|
|
||||||
def _update_auth(self):
|
def _update_auth(self):
|
||||||
if CrunchyrollBaseIE._AUTH_HEADERS and CrunchyrollBaseIE._AUTH_REFRESH > time_seconds():
|
if CrunchyrollBaseIE._AUTH_HEADERS and CrunchyrollBaseIE._AUTH_REFRESH > time_seconds():
|
||||||
return
|
return
|
||||||
|
|
||||||
assert CrunchyrollBaseIE._BASIC_AUTH, '_update_query needs to be called at least one time beforehand'
|
if not CrunchyrollBaseIE._BASIC_AUTH:
|
||||||
|
cx_api_param = self._CLIENT_ID[self.is_logged_in]
|
||||||
|
self.write_debug(f'Using cxApiParam={cx_api_param}')
|
||||||
|
CrunchyrollBaseIE._BASIC_AUTH = 'Basic ' + base64.b64encode(f'{cx_api_param}:'.encode()).decode()
|
||||||
|
|
||||||
grant_type = 'etp_rt_cookie' if self.is_logged_in else 'client_id'
|
grant_type = 'etp_rt_cookie' if self.is_logged_in else 'client_id'
|
||||||
auth_response = self._download_json(
|
try:
|
||||||
f'{self._BASE_URL}/auth/v1/token', None, note=f'Authenticating with grant_type={grant_type}',
|
auth_response = self._download_json(
|
||||||
headers={'Authorization': CrunchyrollBaseIE._BASIC_AUTH}, data=f'grant_type={grant_type}'.encode())
|
f'{self._BASE_URL}/auth/v1/token', None, note=f'Authenticating with grant_type={grant_type}',
|
||||||
|
headers={'Authorization': CrunchyrollBaseIE._BASIC_AUTH}, data=f'grant_type={grant_type}'.encode())
|
||||||
|
except ExtractorError as error:
|
||||||
|
if isinstance(error.cause, HTTPError) and error.cause.status == 403:
|
||||||
|
raise ExtractorError(
|
||||||
|
'Request blocked by Cloudflare; navigate to Crunchyroll in your browser, '
|
||||||
|
'then pass the fresh cookies (with --cookies-from-browser or --cookies) '
|
||||||
|
'and your browser\'s User-Agent (with --user-agent)', expected=True)
|
||||||
|
raise
|
||||||
|
|
||||||
CrunchyrollBaseIE._AUTH_HEADERS = {'Authorization': auth_response['token_type'] + ' ' + auth_response['access_token']}
|
CrunchyrollBaseIE._AUTH_HEADERS = {'Authorization': auth_response['token_type'] + ' ' + auth_response['access_token']}
|
||||||
CrunchyrollBaseIE._AUTH_REFRESH = time_seconds(seconds=traverse_obj(auth_response, ('expires_in', {float_or_none}), default=300) - 10)
|
CrunchyrollBaseIE._AUTH_REFRESH = time_seconds(seconds=traverse_obj(auth_response, ('expires_in', {float_or_none}), default=300) - 10)
|
||||||
|
|
||||||
|
def _locale_from_language(self, language):
|
||||||
|
config_locale = self._configuration_arg('metadata', ie_key=CrunchyrollBetaIE, casesense=True)
|
||||||
|
return config_locale[0] if config_locale else self._LOCALE_LOOKUP.get(language)
|
||||||
|
|
||||||
def _call_base_api(self, endpoint, internal_id, lang, note=None, query={}):
|
def _call_base_api(self, endpoint, internal_id, lang, note=None, query={}):
|
||||||
self._update_query(lang)
|
|
||||||
self._update_auth()
|
self._update_auth()
|
||||||
|
|
||||||
if not endpoint.startswith('/'):
|
if not endpoint.startswith('/'):
|
||||||
endpoint = f'/{endpoint}'
|
endpoint = f'/{endpoint}'
|
||||||
|
|
||||||
|
query = query.copy()
|
||||||
|
locale = self._locale_from_language(lang)
|
||||||
|
if locale:
|
||||||
|
query['locale'] = locale
|
||||||
|
|
||||||
return self._download_json(
|
return self._download_json(
|
||||||
f'{self._BASE_URL}{endpoint}', internal_id, note or f'Calling API: {endpoint}',
|
f'{self._BASE_URL}{endpoint}', internal_id, note or f'Calling API: {endpoint}',
|
||||||
headers=CrunchyrollBaseIE._AUTH_HEADERS, query={**CrunchyrollBaseIE._QUERY[lang], **query})
|
headers=CrunchyrollBaseIE._AUTH_HEADERS, query=query)
|
||||||
|
|
||||||
def _call_api(self, path, internal_id, lang, note='api', query={}):
|
def _call_api(self, path, internal_id, lang, note='api', query={}):
|
||||||
if not path.startswith(f'/content/v2/{self._API_ENDPOINT}/'):
|
if not path.startswith(f'/content/v2/{self._API_ENDPOINT}/'):
|
||||||
|
@ -206,7 +219,7 @@ class CrunchyrollBetaIE(CrunchyrollCmsBaseIE):
|
||||||
IE_NAME = 'crunchyroll'
|
IE_NAME = 'crunchyroll'
|
||||||
_VALID_URL = r'''(?x)
|
_VALID_URL = r'''(?x)
|
||||||
https?://(?:beta\.|www\.)?crunchyroll\.com/
|
https?://(?:beta\.|www\.)?crunchyroll\.com/
|
||||||
(?P<lang>(?:\w{2}(?:-\w{2})?/)?)
|
(?:(?P<lang>\w{2}(?:-\w{2})?)/)?
|
||||||
watch/(?!concert|musicvideo)(?P<id>\w+)'''
|
watch/(?!concert|musicvideo)(?P<id>\w+)'''
|
||||||
_TESTS = [{
|
_TESTS = [{
|
||||||
# Premium only
|
# Premium only
|
||||||
|
@ -304,7 +317,7 @@ class CrunchyrollBetaIE(CrunchyrollCmsBaseIE):
|
||||||
},
|
},
|
||||||
'playlist_mincount': 5,
|
'playlist_mincount': 5,
|
||||||
}, {
|
}, {
|
||||||
'url': 'https://www.crunchyroll.com/watch/GY2P1Q98Y',
|
'url': 'https://www.crunchyroll.com/de/watch/GY2P1Q98Y',
|
||||||
'only_matching': True,
|
'only_matching': True,
|
||||||
}, {
|
}, {
|
||||||
'url': 'https://beta.crunchyroll.com/pt-br/watch/G8WUN8VKP/the-ruler-of-conspiracy',
|
'url': 'https://beta.crunchyroll.com/pt-br/watch/G8WUN8VKP/the-ruler-of-conspiracy',
|
||||||
|
|
Loading…
Reference in a new issue