mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-27 03:03:01 +00:00
db77c49c84
Closes #1945 Authored by: Ashish0804
210 lines
9.3 KiB
Python
210 lines
9.3 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import datetime
|
|
import math
|
|
import random
|
|
import time
|
|
import uuid
|
|
|
|
from .common import InfoExtractor
|
|
from ..compat import compat_HTTPError
|
|
from ..utils import (
|
|
ExtractorError,
|
|
int_or_none,
|
|
try_get,
|
|
)
|
|
|
|
|
|
class SonyLIVIE(InfoExtractor):
|
|
_VALID_URL = r'''(?x)
|
|
(?:
|
|
sonyliv:|
|
|
https?://(?:www\.)?sonyliv\.com/(?:s(?:how|port)s/[^/]+|movies|clip|trailer|music-videos)/[^/?#&]+-
|
|
)
|
|
(?P<id>\d+)
|
|
'''
|
|
_TESTS = [{
|
|
'url': 'https://www.sonyliv.com/shows/bachelors-delight-1700000113/achaari-cheese-toast-1000022678?watch=true',
|
|
'info_dict': {
|
|
'title': 'Achaari Cheese Toast',
|
|
'id': '1000022678',
|
|
'ext': 'mp4',
|
|
'upload_date': '20200411',
|
|
'description': 'md5:3957fa31d9309bf336ceb3f37ad5b7cb',
|
|
'timestamp': 1586632091,
|
|
'duration': 185,
|
|
'season_number': 1,
|
|
'series': 'Bachelors Delight',
|
|
'episode_number': 1,
|
|
'release_year': 2016,
|
|
},
|
|
'params': {
|
|
'skip_download': True,
|
|
},
|
|
}, {
|
|
'url': 'https://www.sonyliv.com/movies/tahalka-1000050121?watch=true',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://www.sonyliv.com/clip/jigarbaaz-1000098925',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://www.sonyliv.com/trailer/sandwiched-forever-1000100286?watch=true',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://www.sonyliv.com/sports/india-tour-of-australia-2020-21-1700000286/cricket-hls-day-3-1st-test-aus-vs-ind-19-dec-2020-1000100959?watch=true',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://www.sonyliv.com/music-videos/yeh-un-dinon-ki-baat-hai-1000018779',
|
|
'only_matching': True,
|
|
}]
|
|
_GEO_COUNTRIES = ['IN']
|
|
_HEADERS = {}
|
|
_LOGIN_HINT = 'Use "--username <mobile_number>" to login using OTP or "--username token --password <auth_token>" to login using auth token.'
|
|
_NETRC_MACHINE = 'sonyliv'
|
|
|
|
def _get_device_id(self):
|
|
e = int(time.time() * 1000)
|
|
t = list('xxxxxxxxxxxx4xxxyxxxxxxxxxxxxxxx')
|
|
for i, c in enumerate(t):
|
|
n = int((e + 16 * random.random()) % 16) | 0
|
|
e = math.floor(e / 16)
|
|
if c == 'x':
|
|
t[i] = str(n)
|
|
elif c == 'y':
|
|
t[i] = '{:x}'.format(3 & n | 8)
|
|
return ''.join(t) + '-' + str(int(time.time() * 1000))
|
|
|
|
def _login(self, username, password):
|
|
if username.lower() == 'token' and len(password) > 1198:
|
|
return password
|
|
elif len(username) != 10 or not username.isdigit():
|
|
raise ExtractorError(f'Invalid username/password; {self._LOGIN_HINT}')
|
|
|
|
self.report_login()
|
|
data = '''{"mobileNumber":"%s","channelPartnerID":"MSMIND","country":"IN","timestamp":"%s",
|
|
"otpSize":6,"loginType":"REGISTERORSIGNIN","isMobileMandatory":true}
|
|
''' % (username, datetime.datetime.now().strftime("%Y-%m-%dT%H:%M:%S.%MZ"))
|
|
otp_request_json = self._download_json(
|
|
'https://apiv2.sonyliv.com/AGL/1.6/A/ENG/WEB/IN/HR/CREATEOTP-V2',
|
|
None, note='Sending OTP', data=data.encode(), headers=self._HEADERS)
|
|
if otp_request_json['resultCode'] == 'KO':
|
|
raise ExtractorError(otp_request_json['message'], expected=True)
|
|
otp_code = self._get_tfa_info('OTP')
|
|
data = '''{"channelPartnerID":"MSMIND","mobileNumber":"%s","country":"IN","otp":"%s",
|
|
"dmaId":"IN","ageConfirmation":true,"timestamp":"%s","isMobileMandatory":true}
|
|
''' % (username, otp_code, datetime.datetime.now().strftime("%Y-%m-%dT%H:%M:%S.%MZ"))
|
|
otp_verify_json = self._download_json(
|
|
'https://apiv2.sonyliv.com/AGL/2.0/A/ENG/WEB/IN/HR/CONFIRMOTP-V2',
|
|
None, note='Verifying OTP', data=data.encode(), headers=self._HEADERS)
|
|
if otp_verify_json['resultCode'] == 'KO':
|
|
raise ExtractorError(otp_request_json['message'], expected=True)
|
|
return otp_verify_json['resultObj']['accessToken']
|
|
|
|
def _call_api(self, version, path, video_id):
|
|
try:
|
|
return self._download_json(
|
|
'https://apiv2.sonyliv.com/AGL/%s/A/ENG/WEB/%s' % (version, path),
|
|
video_id, headers=self._HEADERS)['resultObj']
|
|
except ExtractorError as e:
|
|
if isinstance(e.cause, compat_HTTPError) and e.cause.code == 406 and self._parse_json(
|
|
e.cause.read().decode(), video_id)['message'] == 'Please subscribe to watch this content':
|
|
self.raise_login_required(self._LOGIN_HINT, method=None)
|
|
if isinstance(e.cause, compat_HTTPError) and e.cause.code == 403:
|
|
message = self._parse_json(
|
|
e.cause.read().decode(), video_id)['message']
|
|
if message == 'Geoblocked Country':
|
|
self.raise_geo_restricted(countries=self._GEO_COUNTRIES)
|
|
raise ExtractorError(message)
|
|
raise
|
|
|
|
def _real_initialize(self):
|
|
self._HEADERS['security_token'] = self._call_api('1.4', 'ALL/GETTOKEN', None)
|
|
username, password = self._get_login_info()
|
|
if username:
|
|
self._HEADERS['device_id'] = self._get_device_id()
|
|
self._HEADERS['content-type'] = 'application/json'
|
|
self._HEADERS['authorization'] = self._login(username, password)
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
content = self._call_api(
|
|
'1.5', 'IN/CONTENT/VIDEOURL/VOD/' + video_id, video_id)
|
|
if not self.get_param('allow_unplayable_formats') and content.get('isEncrypted'):
|
|
self.report_drm(video_id)
|
|
dash_url = content['videoURL']
|
|
headers = {
|
|
'x-playback-session-id': '%s-%d' % (uuid.uuid4().hex, time.time() * 1000)
|
|
}
|
|
formats = self._extract_mpd_formats(
|
|
dash_url, video_id, mpd_id='dash', headers=headers, fatal=False)
|
|
formats.extend(self._extract_m3u8_formats(
|
|
dash_url.replace('.mpd', '.m3u8').replace('/DASH/', '/HLS/'),
|
|
video_id, 'mp4', m3u8_id='hls', headers=headers, fatal=False))
|
|
for f in formats:
|
|
f.setdefault('http_headers', {}).update(headers)
|
|
self._sort_formats(formats)
|
|
|
|
metadata = self._call_api(
|
|
'1.6', 'IN/DETAIL/' + video_id, video_id)['containers'][0]['metadata']
|
|
title = metadata['episodeTitle']
|
|
subtitles = {}
|
|
for sub in content.get('subtitle', []):
|
|
sub_url = sub.get('subtitleUrl')
|
|
if not sub_url:
|
|
continue
|
|
subtitles.setdefault(sub.get('subtitleLanguageName', 'ENG'), []).append({
|
|
'url': sub_url,
|
|
})
|
|
return {
|
|
'id': video_id,
|
|
'title': title,
|
|
'formats': formats,
|
|
'thumbnail': content.get('posterURL'),
|
|
'description': metadata.get('longDescription') or metadata.get('shortDescription'),
|
|
'timestamp': int_or_none(metadata.get('creationDate'), 1000),
|
|
'duration': int_or_none(metadata.get('duration')),
|
|
'season_number': int_or_none(metadata.get('season')),
|
|
'series': metadata.get('title'),
|
|
'episode_number': int_or_none(metadata.get('episodeNumber')),
|
|
'release_year': int_or_none(metadata.get('year')),
|
|
'subtitles': subtitles,
|
|
}
|
|
|
|
|
|
class SonyLIVSeriesIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?sonyliv\.com/shows/[^/?#&]+-(?P<id>\d{10})$'
|
|
_TESTS = [{
|
|
'url': 'https://www.sonyliv.com/shows/adaalat-1700000091',
|
|
'playlist_mincount': 456,
|
|
'info_dict': {
|
|
'id': '1700000091',
|
|
},
|
|
}]
|
|
_API_SHOW_URL = "https://apiv2.sonyliv.com/AGL/1.9/R/ENG/WEB/IN/DL/DETAIL/{}?kids_safe=false&from=0&to=49"
|
|
_API_EPISODES_URL = "https://apiv2.sonyliv.com/AGL/1.4/R/ENG/WEB/IN/CONTENT/DETAIL/BUNDLE/{}?from=0&to=1000&orderBy=episodeNumber&sortOrder=asc"
|
|
_API_SECURITY_URL = 'https://apiv2.sonyliv.com/AGL/1.4/A/ENG/WEB/ALL/GETTOKEN'
|
|
|
|
def _entries(self, show_id):
|
|
headers = {
|
|
'Accept': 'application/json, text/plain, */*',
|
|
'Referer': 'https://www.sonyliv.com',
|
|
}
|
|
headers['security_token'] = self._download_json(
|
|
self._API_SECURITY_URL, video_id=show_id, headers=headers,
|
|
note='Downloading security token')['resultObj']
|
|
seasons = try_get(
|
|
self._download_json(self._API_SHOW_URL.format(show_id), video_id=show_id, headers=headers),
|
|
lambda x: x['resultObj']['containers'][0]['containers'], list)
|
|
for season in seasons or []:
|
|
season_id = season['id']
|
|
episodes = try_get(
|
|
self._download_json(self._API_EPISODES_URL.format(season_id), video_id=season_id, headers=headers),
|
|
lambda x: x['resultObj']['containers'][0]['containers'], list)
|
|
for episode in episodes or []:
|
|
video_id = episode.get('id')
|
|
yield self.url_result('sonyliv:%s' % video_id, ie=SonyLIVIE.ie_key(), video_id=video_id)
|
|
|
|
def _real_extract(self, url):
|
|
show_id = self._match_id(url)
|
|
return self.playlist_result(self._entries(show_id), playlist_id=show_id)
|