From 9d7fde89a40360396f0baa2ee8bf507f92108b32 Mon Sep 17 00:00:00 2001 From: bashonly <88596187+bashonly@users.noreply.github.com> Date: Sun, 11 Jun 2023 12:15:05 -0500 Subject: [PATCH] [extractor/zee5] Fix extraction of new content (#7280) Authored by: bashonly --- yt_dlp/extractor/zee5.py | 26 +++++++++++++++++++------- 1 file changed, 19 insertions(+), 7 deletions(-) diff --git a/yt_dlp/extractor/zee5.py b/yt_dlp/extractor/zee5.py index a64eb9ed0d..b4734cc8f1 100644 --- a/yt_dlp/extractor/zee5.py +++ b/yt_dlp/extractor/zee5.py @@ -1,14 +1,16 @@ import json -import random -import string +import time +import uuid from .common import InfoExtractor from ..compat import compat_str from ..utils import ( ExtractorError, int_or_none, + jwt_decode_hs256, parse_age_limit, str_or_none, + try_call, try_get, unified_strdate, unified_timestamp, @@ -94,12 +96,12 @@ class Zee5IE(InfoExtractor): 'url': 'https://www.zee5.com/music-videos/details/adhento-gaani-vunnapaatuga-jersey-nani-shraddha-srinath/0-0-56973', 'only_matching': True }] - _DETAIL_API_URL = 'https://spapi.zee5.com/singlePlayback/getDetails/secure?content_id={}&device_id={}&platform_name=desktop_web&country=IN&check_parental_control=false' - _DEVICE_ID = ''.join(random.choices(string.ascii_letters + string.digits, k=20)).ljust(32, '0') + _DEVICE_ID = str(uuid.uuid4()) _USER_TOKEN = None _LOGIN_HINT = 'Use "--username " to login using otp or "--username token" and "--password " to login using user token.' _NETRC_MACHINE = 'zee5' _GEO_COUNTRIES = ['IN'] + _USER_COUNTRY = None def _perform_login(self, username, password): if len(username) == 10 and username.isdigit() and self._USER_TOKEN is None: @@ -118,11 +120,16 @@ class Zee5IE(InfoExtractor): self._USER_TOKEN = otp_verify_json.get('token') if not self._USER_TOKEN: raise ExtractorError(otp_request_json['message'], expected=True) - elif username.lower() == 'token' and len(password) > 1198: + elif username.lower() == 'token' and try_call(lambda: jwt_decode_hs256(password)): self._USER_TOKEN = password else: raise ExtractorError(self._LOGIN_HINT, expected=True) + token = jwt_decode_hs256(self._USER_TOKEN) + if token.get('exp', 0) <= int(time.time()): + raise ExtractorError('User token has expired', expected=True) + self._USER_COUNTRY = token.get('current_country') + def _real_extract(self, url): video_id, display_id = self._match_valid_url(url).group('id', 'display_id') access_token_request = self._download_json( @@ -137,8 +144,13 @@ class Zee5IE(InfoExtractor): data['X-Z5-Guest-Token'] = self._DEVICE_ID json_data = self._download_json( - self._DETAIL_API_URL.format(video_id, self._DEVICE_ID), - video_id, headers={'content-type': 'application/json'}, data=json.dumps(data).encode('utf-8')) + 'https://spapi.zee5.com/singlePlayback/getDetails/secure', video_id, query={ + 'content_id': video_id, + 'device_id': self._DEVICE_ID, + 'platform_name': 'desktop_web', + 'country': self._USER_COUNTRY or self.get_param('geo_bypass_country') or 'IN', + 'check_parental_control': False, + }, headers={'content-type': 'application/json'}, data=json.dumps(data).encode('utf-8')) asset_data = json_data['assetDetails'] show_data = json_data.get('showDetails', {}) if 'premium' in asset_data['business_type']: