yt-dlp/yt_dlp/extractor/hotstar.py

284 lines
10 KiB
Python
Raw Normal View History

2015-12-24 17:59:56 -07:00
# coding: utf-8
from __future__ import unicode_literals
import hashlib
import hmac
import re
import time
import uuid
2020-05-28 08:22:25 -06:00
import json
import random
2015-12-24 17:59:56 -07:00
from .common import InfoExtractor
from ..compat import (
compat_HTTPError,
compat_str
)
2015-12-24 17:59:56 -07:00
from ..utils import (
determine_ext,
ExtractorError,
2015-12-24 17:59:56 -07:00
int_or_none,
str_or_none,
try_get,
url_or_none,
2015-12-24 17:59:56 -07:00
)
class HotStarBaseIE(InfoExtractor):
_AKAMAI_ENCRYPTION_KEY = b'\x05\xfc\x1a\x01\xca\xc9\x4b\xc4\x12\xfc\x53\x12\x07\x75\xf9\xee'
def _call_api_impl(self, path, video_id, query, st=None):
st = int_or_none(st) or int(time.time())
exp = st + 6000
auth = 'st=%d~exp=%d~acl=/*' % (st, exp)
auth += '~hmac=' + hmac.new(self._AKAMAI_ENCRYPTION_KEY, auth.encode(), hashlib.sha256).hexdigest()
def _generate_device_id():
"""
Reversed from javascript library.
JS function is generateUUID
"""
t = int(round(time.time() * 1000))
e = "xxxxxxxx-xxxx-4xxx-xxxx-xxxxxxxxxxxx" # 4 seems to be interchangeable
def _replacer():
n = int((t + 16 * random.random())) % 16 | 0
return hex(n if "x" == e else 3 & n | 8)[2:]
return "".join([_.replace('x', _replacer()) for _ in e])
2020-05-28 08:22:25 -06:00
token = self._download_json(
'https://api.hotstar.com/um/v3/users',
2020-05-28 08:22:25 -06:00
video_id, note='Downloading token',
data=json.dumps({"device_ids": [{"id": compat_str(uuid.uuid4()), "type": "device_id"}]}).encode('utf-8'),
2020-05-28 08:22:25 -06:00
headers={
'hotstarauth': auth,
'x-hs-platform': 'PCTV', # or 'web'
2020-05-28 08:22:25 -06:00
'Content-Type': 'application/json',
})['user_identity']
response = self._download_json(
'https://api.hotstar.com/' + path, video_id, headers={
'hotstarauth': auth,
2020-05-28 08:22:25 -06:00
'x-hs-appversion': '6.72.2',
'x-hs-platform': 'web',
'x-hs-usertoken': token,
}, query=query)
2020-05-28 08:22:25 -06:00
if response['message'] != "Playback URL's fetched successfully":
raise ExtractorError(
2020-05-28 08:22:25 -06:00
response['message'], expected=True)
return response['data']
def _call_api(self, path, video_id, query_name='contentId'):
return self._call_api_impl(path, video_id, {
query_name: video_id,
'tas': 10000,
})
def _call_api_v2(self, path, video_id, st=None):
return self._call_api_impl(
'%s/content/%s' % (path, video_id), video_id, st=st, query={
'desired-config': 'audio_channel:stereo|dynamic_range:sdr|encryption:plain|ladder:tv|package:dash|resolution:hd|subs-tag:HotstarVIP|video_codec:vp9',
2020-05-28 08:22:25 -06:00
'device-id': compat_str(uuid.uuid4()),
'os-name': 'Windows',
'os-version': '10',
})
class HotStarIE(HotStarBaseIE):
IE_NAME = 'hotstar'
_VALID_URL = r'''(?x)
https?://(?:www\.)?hotstar\.com(?:/in)?/(?!in/)
(?:
tv/(?:[^/?#]+/){3}|
(?!tv/)[^?#]+/
)?
(?P<id>\d{10})
'''
_TESTS = [{
# contentData
'url': 'https://www.hotstar.com/can-you-not-spread-rumours/1000076273',
2015-12-24 17:59:56 -07:00
'info_dict': {
'id': '1000076273',
'ext': 'mp4',
'title': 'Can You Not Spread Rumours?',
2015-12-24 17:59:56 -07:00
'description': 'md5:c957d8868e9bc793ccb813691cc4c434',
'timestamp': 1447248600,
2015-12-24 17:59:56 -07:00
'upload_date': '20151111',
'duration': 381,
},
'params': {
# m3u8 download
'skip_download': True,
}
}, {
# contentDetail
'url': 'https://www.hotstar.com/movies/radha-gopalam/1000057157',
'only_matching': True,
}, {
'url': 'http://www.hotstar.com/sports/cricket/rajitha-sizzles-on-debut-with-329/2001477583',
'only_matching': True,
}, {
'url': 'http://www.hotstar.com/1000000515',
'only_matching': True,
}, {
# only available via api v2
'url': 'https://www.hotstar.com/tv/ek-bhram-sarvagun-sampanna/s-2116/janhvi-targets-suman/1000234847',
'only_matching': True,
}]
_GEO_BYPASS = False
2015-12-24 17:59:56 -07:00
def _real_extract(self, url):
video_id = self._match_id(url)
webpage, urlh = self._download_webpage_handle(url, video_id)
st = urlh.headers.get('x-origin-date')
app_state = self._parse_json(self._search_regex(
r'<script>window\.APP_STATE\s*=\s*({.+?})</script>',
webpage, 'app state'), video_id)
video_data = {}
getters = list(
lambda x, k=k: x['initialState']['content%s' % k]['content']
for k in ('Data', 'Detail')
)
for v in app_state.values():
content = try_get(v, getters, dict)
if content and content.get('contentId') == video_id:
video_data = content
break
title = video_data['title']
if not self.get_param('allow_unplayable_formats') and video_data.get('drmProtected'):
raise ExtractorError('This video is DRM protected.', expected=True)
2015-12-24 17:59:56 -07:00
headers = {'Referer': url}
2015-12-24 17:59:56 -07:00
formats = []
geo_restricted = False
# change to v2 in the future
playback_sets = self._call_api_v2('play/v1/playback', video_id, st=st)['playBackSets']
for playback_set in playback_sets:
if not isinstance(playback_set, dict):
continue
format_url = url_or_none(playback_set.get('playbackUrl'))
if not format_url:
continue
format_url = re.sub(
r'(?<=//staragvod)(\d)', r'web\1', format_url)
tags = str_or_none(playback_set.get('tagsCombination')) or ''
if tags and 'encryption:plain' not in tags:
continue
ext = determine_ext(format_url)
try:
if 'package:hls' in tags or ext == 'm3u8':
formats.extend(self._extract_m3u8_formats(
format_url, video_id, 'mp4',
entry_protocol='m3u8_native',
m3u8_id='hls', headers=headers))
elif 'package:dash' in tags or ext == 'mpd':
formats.extend(self._extract_mpd_formats(
format_url, video_id, mpd_id='dash', headers=headers))
elif ext == 'f4m':
# produce broken files
pass
else:
formats.append({
'url': format_url,
'width': int_or_none(playback_set.get('width')),
'height': int_or_none(playback_set.get('height')),
})
except ExtractorError as e:
if isinstance(e.cause, compat_HTTPError) and e.cause.code == 403:
geo_restricted = True
continue
if not formats and geo_restricted:
self.raise_geo_restricted(countries=['IN'], metadata_available=True)
2015-12-24 17:59:56 -07:00
self._sort_formats(formats)
for f in formats:
f.setdefault('http_headers', {}).update(headers)
2015-12-24 17:59:56 -07:00
return {
'id': video_id,
'title': title,
2015-12-24 17:59:56 -07:00
'description': video_data.get('description'),
'duration': int_or_none(video_data.get('duration')),
'timestamp': int_or_none(video_data.get('broadcastDate') or video_data.get('startDate')),
2015-12-24 17:59:56 -07:00
'formats': formats,
'channel': video_data.get('channelName'),
'channel_id': video_data.get('channelId'),
'series': video_data.get('showName'),
'season': video_data.get('seasonName'),
'season_number': int_or_none(video_data.get('seasonNo')),
'season_id': video_data.get('seasonId'),
'episode': title,
'episode_number': int_or_none(video_data.get('episodeNo')),
2015-12-24 17:59:56 -07:00
}
2017-03-16 10:30:11 -06:00
class HotStarPlaylistIE(HotStarBaseIE):
2017-03-16 10:30:11 -06:00
IE_NAME = 'hotstar:playlist'
_VALID_URL = r'https?://(?:www\.)?hotstar\.com/tv/[^/]+/s-\w+/list/[^/]+/t-(?P<id>\w+)'
2017-03-16 10:30:11 -06:00
_TESTS = [{
'url': 'https://www.hotstar.com/tv/savdhaan-india/s-26/list/popular-clips/t-3_2_26',
2017-03-16 10:30:11 -06:00
'info_dict': {
'id': '3_2_26',
2017-03-16 10:30:11 -06:00
},
'playlist_mincount': 20,
2017-03-16 10:30:11 -06:00
}, {
'url': 'https://www.hotstar.com/tv/savdhaan-india/s-26/list/extras/t-2480',
2017-03-16 10:30:11 -06:00
'only_matching': True,
}]
def _real_extract(self, url):
playlist_id = self._match_id(url)
collection = self._call_api('o/v1/tray/find', playlist_id, 'uqId')
2017-03-16 10:30:11 -06:00
entries = [
self.url_result(
'https://www.hotstar.com/%s' % video['contentId'],
ie=HotStarIE.ie_key(), video_id=video['contentId'])
for video in collection['assets']['items']
if video.get('contentId')]
return self.playlist_result(entries, playlist_id)
class HotStarSeriesIE(HotStarBaseIE):
IE_NAME = 'hotstar:series'
_VALID_URL = r'(?:https?://)(?:www\.)?hotstar\.com(?:/in)?/tv/[^/]+/(?P<id>\d{10})$'
_TESTS = [{
'url': 'https://www.hotstar.com/in/tv/radhakrishn/1260000646',
'info_dict': {
'id': '1260000646',
},
'playlist_mincount': 690,
}, {
'url': 'https://www.hotstar.com/tv/dancee-/1260050431',
'info_dict': {
'id': '1260050431',
},
'playlist_mincount': 43,
}]
def _real_extract(self, url):
series_id = self._match_id(url)
headers = {
'x-country-code': 'IN',
'x-platform-code': 'PCTV',
}
detail_json = self._download_json('https://api.hotstar.com/o/v1/show/detail?contentId=' + series_id,
video_id=series_id, headers=headers)
id = compat_str(try_get(detail_json, lambda x: x['body']['results']['item']['id'], int))
item_json = self._download_json('https://api.hotstar.com/o/v1/tray/g/1/items?etid=0&tao=0&tas=10000&eid=' + id,
video_id=series_id, headers=headers)
entries = [
self.url_result(
'https://www.hotstar.com/%d' % video['contentId'],
ie=HotStarIE.ie_key(), video_id=video['contentId'])
for video in item_json['body']['results']['items']
if video.get('contentId')]
return self.playlist_result(entries, series_id)