yt-dlp/yt_dlp/extractor/viu.py

390 lines
14 KiB
Python
Raw Normal View History

2016-09-09 19:49:56 +02:00
# coding: utf-8
from __future__ import unicode_literals
import re
import json
import uuid
import random
import urllib.parse
2016-09-09 19:49:56 +02:00
from .common import InfoExtractor
from ..compat import compat_str
2016-09-09 19:49:56 +02:00
from ..utils import (
ExtractorError,
int_or_none,
strip_or_none,
2021-04-16 02:00:46 +02:00
try_get,
smuggle_url,
unsmuggle_url,
url_or_none,
2016-09-09 19:49:56 +02:00
)
class ViuBaseIE(InfoExtractor):
def _call_api(self, path, *args, headers={}, **kwargs):
response = self._download_json(
f'https://www.viu.com/api/{path}', *args, **kwargs,
headers={**self.geo_verification_headers(), **headers})['response']
if response.get('status') != 'success':
raise ExtractorError(f'{self.IE_NAME} said: {response["message"]}', expected=True)
return response
2016-09-09 19:49:56 +02:00
class ViuIE(ViuBaseIE):
2017-03-22 16:26:59 +01:00
_VALID_URL = r'(?:viu:|https?://[^/]+\.viu\.com/[a-z]{2}/media/)(?P<id>\d+)'
2016-09-09 19:49:56 +02:00
_TESTS = [{
'url': 'https://www.viu.com/en/media/1116705532?containerId=playlist-22168059',
'info_dict': {
'id': '1116705532',
'ext': 'mp4',
'title': 'Citizen Khan - Ep 1',
2016-09-09 19:49:56 +02:00
'description': 'md5:d7ea1604f49e5ba79c212c551ce2110e',
},
'params': {
'skip_download': 'm3u8 download',
},
'skip': 'Geo-restricted to India',
}, {
'url': 'https://www.viu.com/en/media/1130599965',
'info_dict': {
'id': '1130599965',
'ext': 'mp4',
'title': 'Jealousy Incarnate - Episode 1',
'description': 'md5:d3d82375cab969415d2720b6894361e9',
},
'params': {
'skip_download': 'm3u8 download',
},
'skip': 'Geo-restricted to Indonesia',
2017-03-22 16:26:59 +01:00
}, {
'url': 'https://india.viu.com/en/media/1126286865',
'only_matching': True,
2016-09-09 19:49:56 +02:00
}]
def _real_extract(self, url):
video_id = self._match_id(url)
video_data = self._call_api(
'clip/load', video_id, 'Downloading video data', query={
'appid': 'viu_desktop',
'fmt': 'json',
'id': video_id
})['item'][0]
title = video_data['title']
m3u8_url = None
url_path = video_data.get('urlpathd') or video_data.get('urlpath')
tdirforwhole = video_data.get('tdirforwhole')
2016-12-18 18:24:01 +01:00
# #EXT-X-BYTERANGE is not supported by native hls downloader
# and ffmpeg (#10955)
# FIXME: It is supported in yt-dlp
2016-12-18 18:24:01 +01:00
# hls_file = video_data.get('hlsfile')
hls_file = video_data.get('jwhlsfile')
if url_path and tdirforwhole and hls_file:
m3u8_url = '%s/%s/%s' % (url_path, tdirforwhole, hls_file)
else:
2016-12-18 18:24:01 +01:00
# m3u8_url = re.sub(
# r'(/hlsc_)[a-z]+(\d+\.m3u8)',
# r'\1whe\2', video_data['href'])
m3u8_url = video_data['href']
formats = self._extract_m3u8_formats(m3u8_url, video_id, 'mp4')
2016-09-09 19:49:56 +02:00
self._sort_formats(formats)
subtitles = {}
for key, value in video_data.items():
mobj = re.match(r'^subtitle_(?P<lang>[^_]+)_(?P<ext>(vtt|srt))', key)
if not mobj:
continue
subtitles.setdefault(mobj.group('lang'), []).append({
'url': value,
'ext': mobj.group('ext')
})
2016-09-09 19:49:56 +02:00
return {
'id': video_id,
'title': title,
'description': video_data.get('description'),
'series': video_data.get('moviealbumshowname'),
'episode': title,
'episode_number': int_or_none(video_data.get('episodeno')),
'duration': int_or_none(video_data.get('duration')),
2016-09-09 19:49:56 +02:00
'formats': formats,
'subtitles': subtitles,
}
class ViuPlaylistIE(ViuBaseIE):
IE_NAME = 'viu:playlist'
_VALID_URL = r'https?://www\.viu\.com/[^/]+/listing/playlist-(?P<id>\d+)'
2016-09-09 19:49:56 +02:00
_TEST = {
'url': 'https://www.viu.com/en/listing/playlist-22461380',
'info_dict': {
'id': '22461380',
2016-09-09 19:49:56 +02:00
'title': 'The Good Wife',
},
'playlist_count': 16,
'skip': 'Geo-restricted to Indonesia',
}
def _real_extract(self, url):
playlist_id = self._match_id(url)
playlist_data = self._call_api(
'container/load', playlist_id,
'Downloading playlist info', query={
'appid': 'viu_desktop',
'fmt': 'json',
'id': 'playlist-' + playlist_id
})['container']
entries = []
for item in playlist_data.get('item', []):
item_id = item.get('id')
if not item_id:
continue
item_id = compat_str(item_id)
entries.append(self.url_result(
'viu:' + item_id, 'Viu', item_id))
return self.playlist_result(
entries, playlist_id, playlist_data.get('title'))
class ViuOTTIE(InfoExtractor):
IE_NAME = 'viu:ott'
_NETRC_MACHINE = 'viu'
_VALID_URL = r'https?://(?:www\.)?viu\.com/ott/(?P<country_code>[a-z]{2})/(?P<lang_code>[a-z]{2}-[a-z]{2})/vod/(?P<id>\d+)'
_TESTS = [{
'url': 'http://www.viu.com/ott/sg/en-us/vod/3421/The%20Prime%20Minister%20and%20I',
'info_dict': {
'id': '3421',
'ext': 'mp4',
'title': 'A New Beginning',
'description': 'md5:1e7486a619b6399b25ba6a41c0fe5b2c',
},
'params': {
'skip_download': 'm3u8 download',
'noplaylist': True,
},
'skip': 'Geo-restricted to Singapore',
}, {
'url': 'http://www.viu.com/ott/hk/zh-hk/vod/7123/%E5%A4%A7%E4%BA%BA%E5%A5%B3%E5%AD%90',
'info_dict': {
'id': '7123',
'ext': 'mp4',
'title': '這就是我的生活之道',
'description': 'md5:4eb0d8b08cf04fcdc6bbbeb16043434f',
},
'params': {
'skip_download': 'm3u8 download',
'noplaylist': True,
},
'skip': 'Geo-restricted to Hong Kong',
}, {
'url': 'https://www.viu.com/ott/hk/zh-hk/vod/68776/%E6%99%82%E5%B0%9A%E5%AA%BD%E5%92%AA',
'playlist_count': 12,
'info_dict': {
'id': '3916',
'title': '時尚媽咪',
},
'params': {
'skip_download': 'm3u8 download',
'noplaylist': False,
},
'skip': 'Geo-restricted to Hong Kong',
}]
2018-07-19 19:46:50 +02:00
_AREA_ID = {
'HK': 1,
'SG': 2,
'TH': 4,
'PH': 5,
}
_LANGUAGE_FLAG = {
'zh-hk': 1,
'zh-cn': 2,
'en-us': 3,
}
_user_token = None
_auth_codes = {}
def _detect_error(self, response):
code = try_get(response, lambda x: x['status']['code'])
if code and code > 0:
message = try_get(response, lambda x: x['status']['message'])
raise ExtractorError(f'{self.IE_NAME} said: {message} ({code})', expected=True)
return response.get('data') or {}
def _login(self, country_code, video_id):
if self._user_token is None:
username, password = self._get_login_info()
if username is None:
return
headers = {
'Authorization': f'Bearer {self._auth_codes[country_code]}',
'Content-Type': 'application/json'
}
data = self._download_json(
'https://api-gateway-global.viu.com/api/account/validate',
video_id, 'Validating email address', headers=headers,
data=json.dumps({
'principal': username,
'provider': 'email'
}).encode())
if not data.get('exists'):
raise ExtractorError('Invalid email address')
data = self._download_json(
'https://api-gateway-global.viu.com/api/auth/login',
video_id, 'Logging in', headers=headers,
data=json.dumps({
'email': username,
'password': password,
'provider': 'email',
}).encode())
self._detect_error(data)
self._user_token = data.get('identity')
# need to update with valid user's token else will throw an error again
self._auth_codes[country_code] = data.get('token')
return self._user_token
def _get_token(self, country_code, video_id):
rand = ''.join(random.choice('0123456789') for _ in range(10))
return self._download_json(
f'https://api-gateway-global.viu.com/api/auth/token?v={rand}000', video_id,
headers={'Content-Type': 'application/json'}, note='Getting bearer token',
data=json.dumps({
'countryCode': country_code.upper(),
'platform': 'browser',
'platformFlagLabel': 'web',
'language': 'en',
'uuid': str(uuid.uuid4()),
'carrierId': '0'
}).encode('utf-8'))['token']
2018-07-19 19:46:50 +02:00
def _real_extract(self, url):
url, idata = unsmuggle_url(url, {})
country_code, lang_code, video_id = self._match_valid_url(url).groups()
2018-07-19 19:46:50 +02:00
query = {
'r': 'vod/ajax-detail',
'platform_flag_label': 'web',
'product_id': video_id,
}
area_id = self._AREA_ID.get(country_code.upper())
if area_id:
query['area_id'] = area_id
product_data = self._download_json(
f'http://www.viu.com/ott/{country_code}/index.php', video_id,
2018-07-19 19:46:50 +02:00
'Downloading video info', query=query)['data']
video_data = product_data.get('current_product')
if not video_data:
self.raise_geo_restricted()
series_id = video_data.get('series_id')
if self._yes_playlist(series_id, video_id, idata):
series = product_data.get('series') or {}
product = series.get('product')
if product:
entries = []
for entry in sorted(product, key=lambda x: int_or_none(x.get('number', 0))):
item_id = entry.get('product_id')
if not item_id:
continue
entries.append(self.url_result(
smuggle_url(f'http://www.viu.com/ott/{country_code}/{lang_code}/vod/{item_id}/',
{'force_noplaylist': True}),
ViuOTTIE, str(item_id), entry.get('synopsis', '').strip()))
return self.playlist_result(entries, series_id, series.get('name'), series.get('description'))
duration_limit = False
query = {
'ccs_product_id': video_data['ccs_product_id'],
'language_flag_id': self._LANGUAGE_FLAG.get(lang_code.lower()) or '3',
}
def download_playback():
stream_data = self._download_json(
'https://api-gateway-global.viu.com/api/playback/distribute',
video_id=video_id, query=query, fatal=False, note='Downloading stream info',
headers={
'Authorization': f'Bearer {self._auth_codes[country_code]}',
'Referer': url,
'Origin': url
})
return self._detect_error(stream_data).get('stream')
if not self._auth_codes.get(country_code):
self._auth_codes[country_code] = self._get_token(country_code, video_id)
stream_data = None
try:
stream_data = download_playback()
except (ExtractorError, KeyError):
token = self._login(country_code, video_id)
if token is not None:
query['identity'] = token
else:
# preview is limited to 3min for non-members. But we can try to bypass it
duration_limit, query['duration'] = True, '180'
try:
stream_data = download_playback()
except (ExtractorError, KeyError):
if token is not None:
raise
self.raise_login_required(method='password')
if not stream_data:
raise ExtractorError('Cannot get stream info', expected=True)
formats = []
for vid_format, stream_url in (stream_data.get('url') or {}).items():
height = int(self._search_regex(r's(\d+)p', vid_format, 'height', default=None))
# bypass preview duration limit
if duration_limit:
stream_url = urllib.parse.urlparse(stream_url)
query.update({
'duration': video_data.get('time_duration') or '9999999',
'duration_start': '0',
})
stream_url = stream_url._replace(query=urllib.parse.urlencode(dict(
urllib.parse.parse_qsl(stream_url.query, keep_blank_values=True)))).geturl()
formats.append({
'format_id': vid_format,
'url': stream_url,
'height': height,
'ext': 'mp4',
'filesize': try_get(stream_data, lambda x: x['size'][vid_format], int)
})
self._sort_formats(formats)
subtitles = {}
for sub in video_data.get('subtitle') or []:
sub_url = sub.get('url')
if not sub_url:
continue
subtitles.setdefault(sub.get('name'), []).append({
'url': sub_url,
'ext': 'srt',
})
title = strip_or_none(video_data.get('synopsis'))
return {
'id': video_id,
'title': title,
'description': video_data.get('description'),
'series': try_get(product_data, lambda x: x['series']['name']),
'episode': title,
'episode_number': int_or_none(video_data.get('number')),
'duration': int_or_none(stream_data.get('duration')),
'thumbnail': url_or_none(video_data.get('cover_image_url')),
'formats': formats,
'subtitles': subtitles,
}