0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-25 02:45:12 +00:00

[ie/jiocinema] Add extractors (#10026)

Closes #5563, Closes #7759, Closes #8679, Closes #9349
Authored by: bashonly
This commit is contained in:
bashonly 2024-05-25 18:03:05 -05:00 committed by GitHub
parent c92e4e625e
commit 1463945ae5
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 410 additions and 216 deletions

View file

@ -1835,6 +1835,9 @@ #### nhkradirulive (NHK らじる★らじる LIVE)
#### nflplusreplay
* `type`: Type(s) of game replays to extract. Valid types are: `full_game`, `full_game_spanish`, `condensed_game` and `all_22`. You can use `all` to extract all available replay types, which is the default
#### jiocinema
* `refresh_token`: The `refreshToken` UUID from browser local storage can be passed to extend the life of your login session when logging in with `token` as username and the `accessToken` from browser local storage as password
#### jiosaavn
* `bitrate`: Audio bitrates to request. One or more of `16`, `32`, `64`, `128`, `320`. Default is `128,320`

View file

@ -872,6 +872,10 @@
SangiinIE,
)
from .jeuxvideo import JeuxVideoIE
from .jiocinema import (
JioCinemaIE,
JioCinemaSeriesIE,
)
from .jiosaavn import (
JioSaavnSongIE,
JioSaavnAlbumIE,
@ -2282,10 +2286,6 @@
VoicyChannelIE,
)
from .volejtv import VolejTVIE
from .voot import (
VootIE,
VootSeriesIE,
)
from .voxmedia import (
VoxMediaVolumeIE,
VoxMediaIE,

View file

@ -0,0 +1,403 @@
import base64
import itertools
import json
import random
import re
import string
import time
from .common import InfoExtractor
from ..utils import (
ExtractorError,
float_or_none,
int_or_none,
jwt_decode_hs256,
parse_age_limit,
try_call,
url_or_none,
)
from ..utils.traversal import traverse_obj
class JioCinemaBaseIE(InfoExtractor):
_NETRC_MACHINE = 'jiocinema'
_GEO_BYPASS = False
_ACCESS_TOKEN = None
_REFRESH_TOKEN = None
_GUEST_TOKEN = None
_USER_ID = None
_DEVICE_ID = None
_API_HEADERS = {'Origin': 'https://www.jiocinema.com', 'Referer': 'https://www.jiocinema.com/'}
_APP_NAME = {'appName': 'RJIL_JioCinema'}
_APP_VERSION = {'appVersion': '5.0.0'}
_API_SIGNATURES = 'o668nxgzwff'
_METADATA_API_BASE = 'https://content-jiovoot.voot.com/psapi'
_ACCESS_HINT = 'the `accessToken` from your browser local storage'
_LOGIN_HINT = (
'Log in with "-u phone -p <PHONE_NUMBER>" to authenticate with OTP, '
f'or use "-u token -p <ACCESS_TOKEN>" to log in with {_ACCESS_HINT}. '
'If you have previously logged in with yt-dlp and your session '
'has been cached, you can use "-u device -p <DEVICE_ID>"')
def _cache_token(self, token_type):
assert token_type in ('access', 'refresh', 'all')
if token_type in ('access', 'all'):
self.cache.store(
JioCinemaBaseIE._NETRC_MACHINE, f'{JioCinemaBaseIE._DEVICE_ID}-access', JioCinemaBaseIE._ACCESS_TOKEN)
if token_type in ('refresh', 'all'):
self.cache.store(
JioCinemaBaseIE._NETRC_MACHINE, f'{JioCinemaBaseIE._DEVICE_ID}-refresh', JioCinemaBaseIE._REFRESH_TOKEN)
def _call_api(self, url, video_id, note='Downloading API JSON', headers={}, data={}):
return self._download_json(
url, video_id, note, data=json.dumps(data, separators=(',', ':')).encode(), headers={
'Content-Type': 'application/json',
'Accept': 'application/json',
**self._API_HEADERS,
**headers,
}, expected_status=(400, 403, 474))
def _call_auth_api(self, service, endpoint, note, headers={}, data={}):
return self._call_api(
f'https://auth-jiocinema.voot.com/{service}service/apis/v4/{endpoint}',
None, note=note, headers=headers, data=data)
def _refresh_token(self):
if not JioCinemaBaseIE._REFRESH_TOKEN or not JioCinemaBaseIE._DEVICE_ID:
raise ExtractorError('User token has expired', expected=True)
response = self._call_auth_api(
'token', 'refreshtoken', 'Refreshing token',
headers={'accesstoken': self._ACCESS_TOKEN}, data={
**self._APP_NAME,
'deviceId': self._DEVICE_ID,
'refreshToken': self._REFRESH_TOKEN,
**self._APP_VERSION,
})
refresh_token = response.get('refreshTokenId')
if refresh_token and refresh_token != JioCinemaBaseIE._REFRESH_TOKEN:
JioCinemaBaseIE._REFRESH_TOKEN = refresh_token
self._cache_token('refresh')
JioCinemaBaseIE._ACCESS_TOKEN = response['authToken']
self._cache_token('access')
def _fetch_guest_token(self):
JioCinemaBaseIE._DEVICE_ID = ''.join(random.choices(string.digits, k=10))
guest_token = self._call_auth_api(
'token', 'guest', 'Downloading guest token', data={
**self._APP_NAME,
'deviceType': 'phone',
'os': 'ios',
'deviceId': self._DEVICE_ID,
'freshLaunch': False,
'adId': self._DEVICE_ID,
**self._APP_VERSION,
})
self._GUEST_TOKEN = guest_token['authToken']
self._USER_ID = guest_token['userId']
def _call_login_api(self, endpoint, guest_token, data, note):
return self._call_auth_api(
'user', f'loginotp/{endpoint}', note, headers={
**self.geo_verification_headers(),
'accesstoken': self._GUEST_TOKEN,
**self._APP_NAME,
**traverse_obj(guest_token, 'data', {
'deviceType': ('deviceType', {str}),
'os': ('os', {str}),
})}, data=data)
def _is_token_expired(self, token):
return (try_call(lambda: jwt_decode_hs256(token)['exp']) or 0) <= int(time.time() - 180)
def _perform_login(self, username, password):
if self._ACCESS_TOKEN and not self._is_token_expired(self._ACCESS_TOKEN):
return
UUID_RE = r'[\da-f]{8}-(?:[\da-f]{4}-){3}[\da-f]{12}'
if username.lower() == 'token':
if try_call(lambda: jwt_decode_hs256(password)):
JioCinemaBaseIE._ACCESS_TOKEN = password
refresh_hint = 'the `refreshToken` UUID from your browser local storage'
refresh_token = self._configuration_arg('refresh_token', [''], ie_key=JioCinemaIE)[0]
if not refresh_token:
self.to_screen(
'To extend the life of your login session, in addition to your access token, '
'you can pass --extractor-args "jiocinema:refresh_token=REFRESH_TOKEN" '
f'where REFRESH_TOKEN is {refresh_hint}')
elif re.fullmatch(UUID_RE, refresh_token):
JioCinemaBaseIE._REFRESH_TOKEN = refresh_token
else:
self.report_warning(f'Invalid refresh_token value. Use {refresh_hint}')
else:
raise ExtractorError(
f'The password given could not be decoded as a token; use {self._ACCESS_HINT}', expected=True)
elif username.lower() == 'device' and re.fullmatch(rf'(?:{UUID_RE}|\d+)', password):
JioCinemaBaseIE._REFRESH_TOKEN = self.cache.load(JioCinemaBaseIE._NETRC_MACHINE, f'{password}-refresh')
JioCinemaBaseIE._ACCESS_TOKEN = self.cache.load(JioCinemaBaseIE._NETRC_MACHINE, f'{password}-access')
if not JioCinemaBaseIE._REFRESH_TOKEN or not JioCinemaBaseIE._ACCESS_TOKEN:
raise ExtractorError(f'Failed to load cached tokens for device ID "{password}"', expected=True)
elif username.lower() == 'phone' and re.fullmatch(r'\+?\d+', password):
self._fetch_guest_token()
guest_token = jwt_decode_hs256(self._GUEST_TOKEN)
initial_data = {
'number': base64.b64encode(password.encode()).decode(),
**self._APP_VERSION,
}
response = self._call_login_api('send', guest_token, initial_data, 'Requesting OTP')
if not traverse_obj(response, ('OTPInfo', {dict})):
raise ExtractorError('There was a problem with the phone number login attempt')
is_iphone = guest_token.get('os') == 'ios'
response = self._call_login_api('verify', guest_token, {
'deviceInfo': {
'consumptionDeviceName': 'iPhone' if is_iphone else 'Android',
'info': {
'platform': {'name': 'iPhone OS' if is_iphone else 'Android'},
'androidId': self._DEVICE_ID,
'type': 'iOS' if is_iphone else 'Android'
}
},
**initial_data,
'otp': self._get_tfa_info('the one-time password sent to your phone')
}, 'Submitting OTP')
if traverse_obj(response, 'code') == 1043:
raise ExtractorError('Wrong OTP', expected=True)
JioCinemaBaseIE._REFRESH_TOKEN = response['refreshToken']
JioCinemaBaseIE._ACCESS_TOKEN = response['authToken']
else:
raise ExtractorError(self._LOGIN_HINT, expected=True)
user_token = jwt_decode_hs256(JioCinemaBaseIE._ACCESS_TOKEN)['data']
JioCinemaBaseIE._USER_ID = user_token['userId']
JioCinemaBaseIE._DEVICE_ID = user_token['deviceId']
if JioCinemaBaseIE._REFRESH_TOKEN and username != 'device':
self._cache_token('all')
if self.get_param('cachedir') is not False:
self.to_screen(
f'NOTE: For subsequent logins you can use "-u device -p {JioCinemaBaseIE._DEVICE_ID}"')
elif not JioCinemaBaseIE._REFRESH_TOKEN:
JioCinemaBaseIE._REFRESH_TOKEN = self.cache.load(
JioCinemaBaseIE._NETRC_MACHINE, f'{JioCinemaBaseIE._DEVICE_ID}-refresh')
if JioCinemaBaseIE._REFRESH_TOKEN:
self._cache_token('access')
self.to_screen(f'Logging in as device ID "{JioCinemaBaseIE._DEVICE_ID}"')
if self._is_token_expired(JioCinemaBaseIE._ACCESS_TOKEN):
self._refresh_token()
class JioCinemaIE(JioCinemaBaseIE):
IE_NAME = 'jiocinema'
_VALID_URL = r'https?://(?:www\.)?jiocinema\.com/?(?:movies?/[^/?#]+/|tv-shows/(?:[^/?#]+/){3})(?P<id>\d{3,})'
_TESTS = [{
'url': 'https://www.jiocinema.com/tv-shows/agnisakshi-ek-samjhauta/1/pradeep-to-stop-the-wedding/3759931',
'info_dict': {
'id': '3759931',
'ext': 'mp4',
'title': 'Pradeep to stop the wedding?',
'description': 'md5:75f72d1d1a66976633345a3de6d672b1',
'episode': 'Pradeep to stop the wedding?',
'episode_number': 89,
'season': 'Agnisakshi…Ek Samjhauta-S1',
'season_number': 1,
'series': 'Agnisakshi Ek Samjhauta',
'duration': 1238.0,
'thumbnail': r're:https?://.+\.jpg',
'age_limit': 13,
'season_id': '3698031',
'upload_date': '20230606',
'timestamp': 1686009600,
'release_date': '20230607',
'genres': ['Drama'],
},
'params': {'skip_download': 'm3u8'},
}, {
'url': 'https://www.jiocinema.com/movies/bhediya/3754021/watch',
'info_dict': {
'id': '3754021',
'ext': 'mp4',
'title': 'Bhediya',
'description': 'md5:a6bf2900371ac2fc3f1447401a9f7bb0',
'episode': 'Bhediya',
'duration': 8500.0,
'thumbnail': r're:https?://.+\.jpg',
'age_limit': 13,
'upload_date': '20230525',
'timestamp': 1685026200,
'release_date': '20230524',
'genres': ['Comedy'],
},
'params': {'skip_download': 'm3u8'},
}]
def _extract_formats_and_subtitles(self, playback, video_id):
m3u8_url = traverse_obj(playback, (
'data', 'playbackUrls', lambda _, v: v['streamtype'] == 'hls', 'url', {url_or_none}, any))
if not m3u8_url: # DRM-only content only serves dash urls
self.report_drm(video_id)
formats, subtitles = self._extract_m3u8_formats_and_subtitles(m3u8_url, video_id, m3u8_id='hls')
self._remove_duplicate_formats(formats)
return {
# '/_definst_/smil:vod/' m3u8 manifests claim to have 720p+ formats but max out at 480p
'formats': traverse_obj(formats, (
lambda _, v: '/_definst_/smil:vod/' not in v['url'] or v['height'] <= 480)),
'subtitles': subtitles,
}
def _real_extract(self, url):
video_id = self._match_id(url)
if not self._ACCESS_TOKEN and self._is_token_expired(self._GUEST_TOKEN):
self._fetch_guest_token()
elif self._ACCESS_TOKEN and self._is_token_expired(self._ACCESS_TOKEN):
self._refresh_token()
playback = self._call_api(
f'https://apis-jiovoot.voot.com/playbackjv/v3/{video_id}', video_id,
'Downloading playback JSON', headers={
**self.geo_verification_headers(),
'accesstoken': self._ACCESS_TOKEN or self._GUEST_TOKEN,
**self._APP_NAME,
'deviceid': self._DEVICE_ID,
'uniqueid': self._USER_ID,
'x-apisignatures': self._API_SIGNATURES,
'x-platform': 'androidweb',
'x-platform-token': 'web',
}, data={
'4k': False,
'ageGroup': '18+',
'appVersion': '3.4.0',
'bitrateProfile': 'xhdpi',
'capability': {
'drmCapability': {
'aesSupport': 'yes',
'fairPlayDrmSupport': 'none',
'playreadyDrmSupport': 'none',
'widevineDRMSupport': 'none'
},
'frameRateCapability': [{
'frameRateSupport': '30fps',
'videoQuality': '1440p'
}]
},
'continueWatchingRequired': False,
'dolby': False,
'downloadRequest': False,
'hevc': False,
'kidsSafe': False,
'manufacturer': 'Windows',
'model': 'Windows',
'multiAudioRequired': True,
'osVersion': '10',
'parentalPinValid': True,
'x-apisignatures': self._API_SIGNATURES
})
status_code = traverse_obj(playback, ('code', {int}))
if status_code == 474:
self.raise_geo_restricted(countries=['IN'])
elif status_code == 1008:
error_msg = 'This content is only available for premium users'
if self._ACCESS_TOKEN:
raise ExtractorError(error_msg, expected=True)
self.raise_login_required(f'{error_msg}. {self._LOGIN_HINT}', method=None)
elif status_code == 400:
raise ExtractorError('The requested content is not available', expected=True)
elif status_code is not None and status_code != 200:
raise ExtractorError(
f'JioCinema says: {traverse_obj(playback, ("message", {str})) or status_code}')
metadata = self._download_json(
f'{self._METADATA_API_BASE}/voot/v1/voot-web/content/query/asset-details',
video_id, fatal=False, query={
'ids': f'include:{video_id}',
'responseType': 'common',
'devicePlatformType': 'desktop',
})
return {
'id': video_id,
'http_headers': self._API_HEADERS,
**self._extract_formats_and_subtitles(playback, video_id),
**traverse_obj(playback, ('data', {
# fallback metadata
'title': ('name', {str}),
'description': ('fullSynopsis', {str}),
'series': ('show', 'name', {str}, {lambda x: x or None}),
'season': ('tournamentName', {str}, {lambda x: x if x != 'Season 0' else None}),
'season_number': ('episode', 'season', {int_or_none}, {lambda x: x or None}),
'episode': ('fullTitle', {str}),
'episode_number': ('episode', 'episodeNo', {int_or_none}, {lambda x: x or None}),
'age_limit': ('ageNemonic', {parse_age_limit}),
'duration': ('totalDuration', {float_or_none}),
'thumbnail': ('images', {url_or_none}),
})),
**traverse_obj(metadata, ('result', 0, {
'title': ('fullTitle', {str}),
'description': ('fullSynopsis', {str}),
'series': ('showName', {str}, {lambda x: x or None}),
'season': ('seasonName', {str}, {lambda x: x or None}),
'season_number': ('season', {int_or_none}),
'season_id': ('seasonId', {str}, {lambda x: x or None}),
'episode': ('fullTitle', {str}),
'episode_number': ('episode', {int_or_none}),
'timestamp': ('uploadTime', {int_or_none}),
'release_date': ('telecastDate', {str}),
'age_limit': ('ageNemonic', {parse_age_limit}),
'duration': ('duration', {float_or_none}),
'genres': ('genres', ..., {str}),
'thumbnail': ('seo', 'ogImage', {url_or_none}),
})),
}
class JioCinemaSeriesIE(JioCinemaBaseIE):
IE_NAME = 'jiocinema:series'
_VALID_URL = r'https?://(?:www\.)?jiocinema\.com/tv-shows/(?P<slug>[\w-]+)/(?P<id>\d{3,})'
_TESTS = [{
'url': 'https://www.jiocinema.com/tv-shows/naagin/3499917',
'info_dict': {
'id': '3499917',
'title': 'naagin',
},
'playlist_mincount': 120,
}]
def _entries(self, series_id):
seasons = self._download_json(
f'{self._METADATA_API_BASE}/voot/v1/voot-web/content/generic/season-by-show', series_id,
'Downloading series metadata JSON', query={
'sort': 'season:asc',
'id': series_id,
'responseType': 'common',
})
for season_num, season in enumerate(traverse_obj(seasons, ('result', lambda _, v: v['id'])), 1):
season_id = season['id']
label = season.get('season') or season_num
for page_num in itertools.count(1):
episodes = traverse_obj(self._download_json(
f'{self._METADATA_API_BASE}/voot/v1/voot-web/content/generic/series-wise-episode',
season_id, f'Downloading season {label} page {page_num} JSON', query={
'sort': 'episode:asc',
'id': season_id,
'responseType': 'common',
'page': page_num,
}), ('result', lambda _, v: v['id'] and url_or_none(v['slug'])))
if not episodes:
break
for episode in episodes:
yield self.url_result(
episode['slug'], JioCinemaIE, **traverse_obj(episode, {
'video_id': 'id',
'video_title': ('fullTitle', {str}),
'season_number': ('season', {int_or_none}),
'episode_number': ('episode', {int_or_none}),
}))
def _real_extract(self, url):
slug, series_id = self._match_valid_url(url).group('slug', 'id')
return self.playlist_result(self._entries(series_id), series_id, slug)

View file

@ -1,212 +0,0 @@
import json
import time
import uuid
from .common import InfoExtractor
from ..compat import compat_str
from ..networking.exceptions import HTTPError
from ..utils import (
ExtractorError,
float_or_none,
int_or_none,
jwt_decode_hs256,
parse_age_limit,
traverse_obj,
try_call,
try_get,
unified_strdate,
)
class VootBaseIE(InfoExtractor):
_NETRC_MACHINE = 'voot'
_GEO_BYPASS = False
_LOGIN_HINT = 'Log in with "-u <email_address> -p <password>", or use "-u token -p <auth_token>" to login with auth token.'
_TOKEN = None
_EXPIRY = 0
_API_HEADERS = {'Origin': 'https://www.voot.com', 'Referer': 'https://www.voot.com/'}
def _perform_login(self, username, password):
if self._TOKEN and self._EXPIRY:
return
if username.lower() == 'token' and try_call(lambda: jwt_decode_hs256(password)):
VootBaseIE._TOKEN = password
VootBaseIE._EXPIRY = jwt_decode_hs256(password)['exp']
self.report_login()
# Mobile number as username is not supported
elif not username.isdigit():
check_username = self._download_json(
'https://userauth.voot.com/usersV3/v3/checkUser', None, data=json.dumps({
'type': 'email',
'email': username
}, separators=(',', ':')).encode(), headers={
**self._API_HEADERS,
'Content-Type': 'application/json;charset=utf-8',
}, note='Checking username', expected_status=403)
if not traverse_obj(check_username, ('isExist', {bool})):
if traverse_obj(check_username, ('status', 'code', {int})) == 9999:
self.raise_geo_restricted(countries=['IN'])
raise ExtractorError('Incorrect username', expected=True)
auth_token = traverse_obj(self._download_json(
'https://userauth.voot.com/usersV3/v3/login', None, data=json.dumps({
'type': 'traditional',
'deviceId': str(uuid.uuid4()),
'deviceBrand': 'PC/MAC',
'data': {
'email': username,
'password': password
}
}, separators=(',', ':')).encode(), headers={
**self._API_HEADERS,
'Content-Type': 'application/json;charset=utf-8',
}, note='Logging in', expected_status=400), ('data', 'authToken', {dict}))
if not auth_token:
raise ExtractorError('Incorrect password', expected=True)
VootBaseIE._TOKEN = auth_token['accessToken']
VootBaseIE._EXPIRY = auth_token['expirationTime']
else:
raise ExtractorError(self._LOGIN_HINT, expected=True)
def _check_token_expiry(self):
if int(time.time()) >= self._EXPIRY:
raise ExtractorError('Access token has expired', expected=True)
def _real_initialize(self):
if not self._TOKEN:
self.raise_login_required(self._LOGIN_HINT, method=None)
self._check_token_expiry()
class VootIE(VootBaseIE):
_WORKING = False
_VALID_URL = r'''(?x)
(?:
voot:|
https?://(?:www\.)?voot\.com/?
(?:
movies?/[^/]+/|
(?:shows|kids)/(?:[^/]+/){4}
)
)
(?P<id>\d{3,})
'''
_TESTS = [{
'url': 'https://www.voot.com/shows/ishq-ka-rang-safed/1/360558/is-this-the-end-of-kamini-/441353',
'info_dict': {
'id': '441353',
'ext': 'mp4',
'title': 'Is this the end of Kamini?',
'description': 'md5:06291fbbbc4dcbe21235c40c262507c1',
'timestamp': 1472103000,
'upload_date': '20160825',
'series': 'Ishq Ka Rang Safed',
'season_number': 1,
'episode': 'Is this the end of Kamini?',
'episode_number': 340,
'release_date': '20160825',
'season': 'Season 1',
'age_limit': 13,
'duration': 1146.0,
},
'params': {'skip_download': 'm3u8'},
}, {
'url': 'https://www.voot.com/kids/characters/mighty-cat-masked-niyander-e-/400478/school-bag-disappears/440925',
'only_matching': True,
}, {
'url': 'https://www.voot.com/movies/pandavas-5/424627',
'only_matching': True,
}, {
'url': 'https://www.voot.com/movie/fight-club/621842',
'only_matching': True,
}]
def _real_extract(self, url):
video_id = self._match_id(url)
media_info = self._download_json(
'https://psapi.voot.com/jio/voot/v1/voot-web/content/query/asset-details', video_id,
query={'ids': f'include:{video_id}', 'responseType': 'common'}, headers={'accesstoken': self._TOKEN})
try:
m3u8_url = self._download_json(
'https://vootapi.media.jio.com/playback/v1/playbackrights', video_id,
'Downloading playback JSON', data=b'{}', headers={
**self.geo_verification_headers(),
**self._API_HEADERS,
'Content-Type': 'application/json;charset=utf-8',
'platform': 'androidwebdesktop',
'vootid': video_id,
'voottoken': self._TOKEN,
})['m3u8']
except ExtractorError as e:
if isinstance(e.cause, HTTPError) and e.cause.status == 400:
self._check_token_expiry()
raise
formats = self._extract_m3u8_formats(m3u8_url, video_id, 'mp4', m3u8_id='hls')
self._remove_duplicate_formats(formats)
return {
'id': video_id,
# '/_definst_/smil:vod/' m3u8 manifests claim to have 720p+ formats but max out at 480p
'formats': traverse_obj(formats, (
lambda _, v: '/_definst_/smil:vod/' not in v['url'] or v['height'] <= 480)),
'http_headers': self._API_HEADERS,
**traverse_obj(media_info, ('result', 0, {
'title': ('fullTitle', {str}),
'description': ('fullSynopsis', {str}),
'series': ('showName', {str}),
'season_number': ('season', {int_or_none}),
'episode': ('fullTitle', {str}),
'episode_number': ('episode', {int_or_none}),
'timestamp': ('uploadTime', {int_or_none}),
'release_date': ('telecastDate', {unified_strdate}),
'age_limit': ('ageNemonic', {parse_age_limit}),
'duration': ('duration', {float_or_none}),
})),
}
class VootSeriesIE(VootBaseIE):
_WORKING = False
_VALID_URL = r'https?://(?:www\.)?voot\.com/shows/[^/]+/(?P<id>\d{3,})'
_TESTS = [{
'url': 'https://www.voot.com/shows/chakravartin-ashoka-samrat/100002',
'playlist_mincount': 442,
'info_dict': {
'id': '100002',
},
}, {
'url': 'https://www.voot.com/shows/ishq-ka-rang-safed/100003',
'playlist_mincount': 341,
'info_dict': {
'id': '100003',
},
}]
_SHOW_API = 'https://psapi.voot.com/media/voot/v1/voot-web/content/generic/season-by-show?sort=season%3Aasc&id={}&responseType=common'
_SEASON_API = 'https://psapi.voot.com/media/voot/v1/voot-web/content/generic/series-wise-episode?sort=episode%3Aasc&id={}&responseType=common&page={:d}'
def _entries(self, show_id):
show_json = self._download_json(self._SHOW_API.format(show_id), video_id=show_id)
for season in show_json.get('result', []):
page_num = 1
season_id = try_get(season, lambda x: x['id'], compat_str)
season_json = self._download_json(self._SEASON_API.format(season_id, page_num),
video_id=season_id,
note='Downloading JSON metadata page %d' % page_num)
episodes_json = season_json.get('result', [])
while episodes_json:
page_num += 1
for episode in episodes_json:
video_id = episode.get('id')
yield self.url_result(
'voot:%s' % video_id, ie=VootIE.ie_key(), video_id=video_id)
episodes_json = self._download_json(self._SEASON_API.format(season_id, page_num),
video_id=season_id,
note='Downloading JSON metadata page %d' % page_num)['result']
def _real_extract(self, url):
show_id = self._match_id(url)
return self.playlist_result(self._entries(show_id), playlist_id=show_id)