mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-18 07:01:45 +00:00
8b7539d27c
Closes #2526, #1614
323 lines
12 KiB
Python
323 lines
12 KiB
Python
# coding: utf-8
|
||
from __future__ import unicode_literals
|
||
|
||
import base64
|
||
from datetime import datetime
|
||
import itertools
|
||
import json
|
||
|
||
from .common import InfoExtractor
|
||
from ..utils import (
|
||
update_url_query,
|
||
random_uuidv4,
|
||
try_get,
|
||
float_or_none,
|
||
dict_get
|
||
)
|
||
from ..compat import (
|
||
compat_str,
|
||
)
|
||
|
||
|
||
class MildomBaseIE(InfoExtractor):
|
||
_GUEST_ID = None
|
||
_DISPATCHER_CONFIG = None
|
||
|
||
def _call_api(self, url, video_id, query=None, note='Downloading JSON metadata', init=False):
|
||
query = query or {}
|
||
if query:
|
||
query['__platform'] = 'web'
|
||
url = update_url_query(url, self._common_queries(query, init=init))
|
||
content = self._download_json(url, video_id, note=note)
|
||
if content['code'] == 0:
|
||
return content['body']
|
||
else:
|
||
self.raise_no_formats(
|
||
f'Video not found or premium content. {content["code"]} - {content["message"]}',
|
||
expected=True)
|
||
|
||
def _common_queries(self, query={}, init=False):
|
||
dc = self._fetch_dispatcher_config()
|
||
r = {
|
||
'timestamp': self.iso_timestamp(),
|
||
'__guest_id': '' if init else self.guest_id(),
|
||
'__location': dc['location'],
|
||
'__country': dc['country'],
|
||
'__cluster': dc['cluster'],
|
||
'__platform': 'web',
|
||
'__la': self.lang_code(),
|
||
'__pcv': 'v2.9.44',
|
||
'sfr': 'pc',
|
||
'accessToken': '',
|
||
}
|
||
r.update(query)
|
||
return r
|
||
|
||
def _fetch_dispatcher_config(self):
|
||
if not self._DISPATCHER_CONFIG:
|
||
tmp = self._download_json(
|
||
'https://disp.mildom.com/serverListV2', 'initialization',
|
||
note='Downloading dispatcher_config', data=json.dumps({
|
||
'protover': 0,
|
||
'data': base64.b64encode(json.dumps({
|
||
'fr': 'web',
|
||
'sfr': 'pc',
|
||
'devi': 'Windows',
|
||
'la': 'ja',
|
||
'gid': None,
|
||
'loc': '',
|
||
'clu': '',
|
||
'wh': '1919*810',
|
||
'rtm': self.iso_timestamp(),
|
||
'ua': self.get_param('http_headers')['User-Agent'],
|
||
}).encode('utf8')).decode('utf8').replace('\n', ''),
|
||
}).encode('utf8'))
|
||
self._DISPATCHER_CONFIG = self._parse_json(base64.b64decode(tmp['data']), 'initialization')
|
||
return self._DISPATCHER_CONFIG
|
||
|
||
@staticmethod
|
||
def iso_timestamp():
|
||
'new Date().toISOString()'
|
||
return datetime.utcnow().isoformat()[0:-3] + 'Z'
|
||
|
||
def guest_id(self):
|
||
'getGuestId'
|
||
if self._GUEST_ID:
|
||
return self._GUEST_ID
|
||
self._GUEST_ID = try_get(
|
||
self, (
|
||
lambda x: x._call_api(
|
||
'https://cloudac.mildom.com/nonolive/gappserv/guest/h5init', 'initialization',
|
||
note='Downloading guest token', init=True)['guest_id'] or None,
|
||
lambda x: x._get_cookies('https://www.mildom.com').get('gid').value,
|
||
lambda x: x._get_cookies('https://m.mildom.com').get('gid').value,
|
||
), compat_str) or ''
|
||
return self._GUEST_ID
|
||
|
||
def lang_code(self):
|
||
'getCurrentLangCode'
|
||
return 'ja'
|
||
|
||
|
||
class MildomIE(MildomBaseIE):
|
||
IE_NAME = 'mildom'
|
||
IE_DESC = 'Record ongoing live by specific user in Mildom'
|
||
_VALID_URL = r'https?://(?:(?:www|m)\.)mildom\.com/(?P<id>\d+)'
|
||
|
||
def _real_extract(self, url):
|
||
video_id = self._match_id(url)
|
||
url = 'https://www.mildom.com/%s' % video_id
|
||
|
||
webpage = self._download_webpage(url, video_id)
|
||
|
||
enterstudio = self._call_api(
|
||
'https://cloudac.mildom.com/nonolive/gappserv/live/enterstudio', video_id,
|
||
note='Downloading live metadata', query={'user_id': video_id})
|
||
result_video_id = enterstudio.get('log_id', video_id)
|
||
|
||
title = try_get(
|
||
enterstudio, (
|
||
lambda x: self._html_search_meta('twitter:description', webpage),
|
||
lambda x: x['anchor_intro'],
|
||
), compat_str)
|
||
description = try_get(
|
||
enterstudio, (
|
||
lambda x: x['intro'],
|
||
lambda x: x['live_intro'],
|
||
), compat_str)
|
||
uploader = try_get(
|
||
enterstudio, (
|
||
lambda x: self._html_search_meta('twitter:title', webpage),
|
||
lambda x: x['loginname'],
|
||
), compat_str)
|
||
|
||
servers = self._call_api(
|
||
'https://cloudac.mildom.com/nonolive/gappserv/live/liveserver', result_video_id,
|
||
note='Downloading live server list', query={
|
||
'user_id': video_id,
|
||
'live_server_type': 'hls',
|
||
})
|
||
|
||
stream_query = self._common_queries({
|
||
'streamReqId': random_uuidv4(),
|
||
'is_lhls': '0',
|
||
})
|
||
m3u8_url = update_url_query(servers['stream_server'] + '/%s_master.m3u8' % video_id, stream_query)
|
||
formats = self._extract_m3u8_formats(m3u8_url, result_video_id, 'mp4', headers={
|
||
'Referer': 'https://www.mildom.com/',
|
||
'Origin': 'https://www.mildom.com',
|
||
}, note='Downloading m3u8 information')
|
||
|
||
del stream_query['streamReqId'], stream_query['timestamp']
|
||
for fmt in formats:
|
||
fmt.setdefault('http_headers', {})['Referer'] = 'https://www.mildom.com/'
|
||
|
||
self._sort_formats(formats)
|
||
|
||
return {
|
||
'id': result_video_id,
|
||
'title': title,
|
||
'description': description,
|
||
'timestamp': float_or_none(enterstudio.get('live_start_ms'), scale=1000),
|
||
'uploader': uploader,
|
||
'uploader_id': video_id,
|
||
'formats': formats,
|
||
'is_live': True,
|
||
}
|
||
|
||
|
||
class MildomVodIE(MildomBaseIE):
|
||
IE_NAME = 'mildom:vod'
|
||
IE_DESC = 'Download a VOD in Mildom'
|
||
_VALID_URL = r'https?://(?:(?:www|m)\.)mildom\.com/playback/(?P<user_id>\d+)/(?P<id>(?P=user_id)-[a-zA-Z0-9]+-?[0-9]*)'
|
||
_TESTS = [{
|
||
'url': 'https://www.mildom.com/playback/10882672/10882672-1597662269',
|
||
'info_dict': {
|
||
'id': '10882672-1597662269',
|
||
'ext': 'mp4',
|
||
'title': '始めてのミルダム配信じゃぃ!',
|
||
'thumbnail': r're:^https?://.*\.(png|jpg)$',
|
||
'upload_date': '20200817',
|
||
'duration': 4138.37,
|
||
'description': 'ゲームをしたくて!',
|
||
'timestamp': 1597662269.0,
|
||
'uploader_id': '10882672',
|
||
'uploader': 'kson組長(けいそん)',
|
||
},
|
||
}, {
|
||
'url': 'https://www.mildom.com/playback/10882672/10882672-1597758589870-477',
|
||
'info_dict': {
|
||
'id': '10882672-1597758589870-477',
|
||
'ext': 'mp4',
|
||
'title': '【kson】感染メイズ!麻酔銃で無双する',
|
||
'thumbnail': r're:^https?://.*\.(png|jpg)$',
|
||
'timestamp': 1597759093.0,
|
||
'uploader': 'kson組長(けいそん)',
|
||
'duration': 4302.58,
|
||
'uploader_id': '10882672',
|
||
'description': 'このステージ絶対乗り越えたい',
|
||
'upload_date': '20200818',
|
||
},
|
||
}, {
|
||
'url': 'https://www.mildom.com/playback/10882672/10882672-buha9td2lrn97fk2jme0',
|
||
'info_dict': {
|
||
'id': '10882672-buha9td2lrn97fk2jme0',
|
||
'ext': 'mp4',
|
||
'title': '【kson組長】CART RACER!!!',
|
||
'thumbnail': r're:^https?://.*\.(png|jpg)$',
|
||
'uploader_id': '10882672',
|
||
'uploader': 'kson組長(けいそん)',
|
||
'upload_date': '20201104',
|
||
'timestamp': 1604494797.0,
|
||
'duration': 4657.25,
|
||
'description': 'WTF',
|
||
},
|
||
}]
|
||
|
||
def _real_extract(self, url):
|
||
m = self._match_valid_url(url)
|
||
user_id, video_id = m.group('user_id'), m.group('id')
|
||
url = 'https://www.mildom.com/playback/%s/%s' % (user_id, video_id)
|
||
|
||
webpage = self._download_webpage(url, video_id)
|
||
|
||
autoplay = self._call_api(
|
||
'https://cloudac.mildom.com/nonolive/videocontent/playback/getPlaybackDetail', video_id,
|
||
note='Downloading playback metadata', query={
|
||
'v_id': video_id,
|
||
})['playback']
|
||
|
||
title = try_get(
|
||
autoplay, (
|
||
lambda x: self._html_search_meta('og:description', webpage),
|
||
lambda x: x['title'],
|
||
), compat_str)
|
||
description = try_get(
|
||
autoplay, (
|
||
lambda x: x['video_intro'],
|
||
), compat_str)
|
||
uploader = try_get(
|
||
autoplay, (
|
||
lambda x: x['author_info']['login_name'],
|
||
), compat_str)
|
||
|
||
formats = [{
|
||
'url': autoplay['audio_url'],
|
||
'format_id': 'audio',
|
||
'protocol': 'm3u8_native',
|
||
'vcodec': 'none',
|
||
'acodec': 'aac',
|
||
'ext': 'm4a'
|
||
}]
|
||
for fmt in autoplay['video_link']:
|
||
formats.append({
|
||
'format_id': 'video-%s' % fmt['name'],
|
||
'url': fmt['url'],
|
||
'protocol': 'm3u8_native',
|
||
'width': fmt['level'] * autoplay['video_width'] // autoplay['video_height'],
|
||
'height': fmt['level'],
|
||
'vcodec': 'h264',
|
||
'acodec': 'aac',
|
||
'ext': 'mp4'
|
||
})
|
||
|
||
self._sort_formats(formats)
|
||
|
||
return {
|
||
'id': video_id,
|
||
'title': title,
|
||
'description': description,
|
||
'timestamp': float_or_none(autoplay['publish_time'], scale=1000),
|
||
'duration': float_or_none(autoplay['video_length'], scale=1000),
|
||
'thumbnail': dict_get(autoplay, ('upload_pic', 'video_pic')),
|
||
'uploader': uploader,
|
||
'uploader_id': user_id,
|
||
'formats': formats,
|
||
}
|
||
|
||
|
||
class MildomUserVodIE(MildomBaseIE):
|
||
IE_NAME = 'mildom:user:vod'
|
||
IE_DESC = 'Download all VODs from specific user in Mildom'
|
||
_VALID_URL = r'https?://(?:(?:www|m)\.)mildom\.com/profile/(?P<id>\d+)'
|
||
_TESTS = [{
|
||
'url': 'https://www.mildom.com/profile/10093333',
|
||
'info_dict': {
|
||
'id': '10093333',
|
||
'title': 'Uploads from ねこばたけ',
|
||
},
|
||
'playlist_mincount': 351,
|
||
}, {
|
||
'url': 'https://www.mildom.com/profile/10882672',
|
||
'info_dict': {
|
||
'id': '10882672',
|
||
'title': 'Uploads from kson組長(けいそん)',
|
||
},
|
||
'playlist_mincount': 191,
|
||
}]
|
||
|
||
def _entries(self, user_id):
|
||
for page in itertools.count(1):
|
||
reply = self._call_api(
|
||
'https://cloudac.mildom.com/nonolive/videocontent/profile/playbackList',
|
||
user_id, note='Downloading page %d' % page, query={
|
||
'user_id': user_id,
|
||
'page': page,
|
||
'limit': '30',
|
||
})
|
||
if not reply:
|
||
break
|
||
for x in reply:
|
||
yield self.url_result('https://www.mildom.com/playback/%s/%s' % (user_id, x['v_id']))
|
||
|
||
def _real_extract(self, url):
|
||
user_id = self._match_id(url)
|
||
self.to_screen('This will download all VODs belonging to user. To download ongoing live video, use "https://www.mildom.com/%s" instead' % user_id)
|
||
|
||
profile = self._call_api(
|
||
'https://cloudac.mildom.com/nonolive/gappserv/user/profileV2', user_id,
|
||
query={'user_id': user_id}, note='Downloading user profile')['user_info']
|
||
|
||
return self.playlist_result(
|
||
self._entries(user_id), user_id, 'Uploads from %s' % profile['loginname'])
|