mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-14 12:23:19 +00:00
4cca2eb1bf
Closes #4917 Authored by: tannertechnology
324 lines
14 KiB
Python
324 lines
14 KiB
Python
import itertools
|
||
import json
|
||
import time
|
||
import urllib.error
|
||
import urllib.parse
|
||
|
||
from .common import InfoExtractor
|
||
from ..utils import ExtractorError, parse_iso8601, try_get
|
||
|
||
_BASE_URL_RE = r'https?://(?:www\.)?(?:watchnebula\.com|nebula\.app|nebula\.tv)'
|
||
|
||
|
||
class NebulaBaseIE(InfoExtractor):
|
||
_NETRC_MACHINE = 'watchnebula'
|
||
|
||
_nebula_api_token = None
|
||
_nebula_bearer_token = None
|
||
_zype_access_token = None
|
||
|
||
def _perform_nebula_auth(self, username, password):
|
||
if not username or not password:
|
||
self.raise_login_required()
|
||
|
||
data = json.dumps({'email': username, 'password': password}).encode('utf8')
|
||
response = self._download_json(
|
||
'https://api.watchnebula.com/api/v1/auth/login/',
|
||
data=data, fatal=False, video_id=None,
|
||
headers={
|
||
'content-type': 'application/json',
|
||
# Submitting the 'sessionid' cookie always causes a 403 on auth endpoint
|
||
'cookie': ''
|
||
},
|
||
note='Logging in to Nebula with supplied credentials',
|
||
errnote='Authentication failed or rejected')
|
||
if not response or not response.get('key'):
|
||
self.raise_login_required()
|
||
|
||
# save nebula token as cookie
|
||
self._set_cookie(
|
||
'nebula.app', 'nebula-auth',
|
||
urllib.parse.quote(
|
||
json.dumps({
|
||
"apiToken": response["key"],
|
||
"isLoggingIn": False,
|
||
"isLoggingOut": False,
|
||
}, separators=(",", ":"))),
|
||
expire_time=int(time.time()) + 86400 * 365,
|
||
)
|
||
|
||
return response['key']
|
||
|
||
def _retrieve_nebula_api_token(self, username=None, password=None):
|
||
"""
|
||
Check cookie jar for valid token. Try to authenticate using credentials if no valid token
|
||
can be found in the cookie jar.
|
||
"""
|
||
nebula_cookies = self._get_cookies('https://nebula.app')
|
||
nebula_cookie = nebula_cookies.get('nebula-auth')
|
||
if nebula_cookie:
|
||
self.to_screen('Authenticating to Nebula with token from cookie jar')
|
||
nebula_cookie_value = urllib.parse.unquote(nebula_cookie.value)
|
||
nebula_api_token = self._parse_json(nebula_cookie_value, None).get('apiToken')
|
||
if nebula_api_token:
|
||
return nebula_api_token
|
||
|
||
return self._perform_nebula_auth(username, password)
|
||
|
||
def _call_nebula_api(self, url, video_id=None, method='GET', auth_type='api', note=''):
|
||
assert method in ('GET', 'POST',)
|
||
assert auth_type in ('api', 'bearer',)
|
||
|
||
def inner_call():
|
||
authorization = f'Token {self._nebula_api_token}' if auth_type == 'api' else f'Bearer {self._nebula_bearer_token}'
|
||
return self._download_json(
|
||
url, video_id, note=note, headers={'Authorization': authorization},
|
||
data=b'' if method == 'POST' else None)
|
||
|
||
try:
|
||
return inner_call()
|
||
except ExtractorError as exc:
|
||
# if 401 or 403, attempt credential re-auth and retry
|
||
if exc.cause and isinstance(exc.cause, urllib.error.HTTPError) and exc.cause.code in (401, 403):
|
||
self.to_screen(f'Reauthenticating to Nebula and retrying, because last {auth_type} call resulted in error {exc.cause.code}')
|
||
self._perform_login()
|
||
return inner_call()
|
||
else:
|
||
raise
|
||
|
||
def _fetch_nebula_bearer_token(self):
|
||
"""
|
||
Get a Bearer token for the Nebula API. This will be required to fetch video meta data.
|
||
"""
|
||
response = self._call_nebula_api('https://api.watchnebula.com/api/v1/authorization/',
|
||
method='POST',
|
||
note='Authorizing to Nebula')
|
||
return response['token']
|
||
|
||
def _fetch_zype_access_token(self):
|
||
"""
|
||
Get a Zype access token, which is required to access video streams -- in our case: to
|
||
generate video URLs.
|
||
"""
|
||
user_object = self._call_nebula_api('https://api.watchnebula.com/api/v1/auth/user/', note='Retrieving Zype access token')
|
||
|
||
access_token = try_get(user_object, lambda x: x['zype_auth_info']['access_token'], str)
|
||
if not access_token:
|
||
if try_get(user_object, lambda x: x['is_subscribed'], bool):
|
||
# TODO: Reimplement the same Zype token polling the Nebula frontend implements
|
||
# see https://github.com/ytdl-org/youtube-dl/pull/24805#issuecomment-749231532
|
||
raise ExtractorError(
|
||
'Unable to extract Zype access token from Nebula API authentication endpoint. '
|
||
'Open an arbitrary video in a browser with this account to generate a token',
|
||
expected=True)
|
||
raise ExtractorError('Unable to extract Zype access token from Nebula API authentication endpoint')
|
||
return access_token
|
||
|
||
def _build_video_info(self, episode):
|
||
zype_id = episode['zype_id']
|
||
zype_video_url = f'https://player.zype.com/embed/{zype_id}.html?access_token={self._zype_access_token}'
|
||
channel_slug = episode['channel_slug']
|
||
return {
|
||
'id': episode['zype_id'],
|
||
'display_id': episode['slug'],
|
||
'_type': 'url_transparent',
|
||
'ie_key': 'Zype',
|
||
'url': zype_video_url,
|
||
'title': episode['title'],
|
||
'description': episode['description'],
|
||
'timestamp': parse_iso8601(episode['published_at']),
|
||
'thumbnails': [{
|
||
# 'id': tn.get('name'), # this appears to be null
|
||
'url': tn['original'],
|
||
'height': key,
|
||
} for key, tn in episode['assets']['thumbnail'].items()],
|
||
'duration': episode['duration'],
|
||
'channel': episode['channel_title'],
|
||
'channel_id': channel_slug,
|
||
'channel_url': f'https://nebula.app/{channel_slug}',
|
||
'uploader': episode['channel_title'],
|
||
'uploader_id': channel_slug,
|
||
'uploader_url': f'https://nebula.app/{channel_slug}',
|
||
'series': episode['channel_title'],
|
||
'creator': episode['channel_title'],
|
||
}
|
||
|
||
def _perform_login(self, username=None, password=None):
|
||
self._nebula_api_token = self._retrieve_nebula_api_token(username, password)
|
||
self._nebula_bearer_token = self._fetch_nebula_bearer_token()
|
||
self._zype_access_token = self._fetch_zype_access_token()
|
||
|
||
|
||
class NebulaIE(NebulaBaseIE):
|
||
_VALID_URL = rf'{_BASE_URL_RE}/videos/(?P<id>[-\w]+)'
|
||
_TESTS = [
|
||
{
|
||
'url': 'https://nebula.app/videos/that-time-disney-remade-beauty-and-the-beast',
|
||
'md5': '14944cfee8c7beeea106320c47560efc',
|
||
'info_dict': {
|
||
'id': '5c271b40b13fd613090034fd',
|
||
'ext': 'mp4',
|
||
'title': 'That Time Disney Remade Beauty and the Beast',
|
||
'description': 'Note: this video was originally posted on YouTube with the sponsor read included. We weren’t able to remove it without reducing video quality, so it’s presented here in its original context.',
|
||
'upload_date': '20180731',
|
||
'timestamp': 1533009600,
|
||
'channel': 'Lindsay Ellis',
|
||
'channel_id': 'lindsayellis',
|
||
'uploader': 'Lindsay Ellis',
|
||
'uploader_id': 'lindsayellis',
|
||
'timestamp': 1533009600,
|
||
'uploader_url': 'https://nebula.app/lindsayellis',
|
||
'series': 'Lindsay Ellis',
|
||
'average_rating': int,
|
||
'display_id': 'that-time-disney-remade-beauty-and-the-beast',
|
||
'channel_url': 'https://nebula.app/lindsayellis',
|
||
'creator': 'Lindsay Ellis',
|
||
'duration': 2212,
|
||
'view_count': int,
|
||
'thumbnail': r're:https://\w+\.cloudfront\.net/[\w-]+\.jpeg?.*',
|
||
},
|
||
},
|
||
{
|
||
'url': 'https://nebula.app/videos/the-logistics-of-d-day-landing-craft-how-the-allies-got-ashore',
|
||
'md5': 'd05739cf6c38c09322422f696b569c23',
|
||
'info_dict': {
|
||
'id': '5e7e78171aaf320001fbd6be',
|
||
'ext': 'mp4',
|
||
'title': 'Landing Craft - How The Allies Got Ashore',
|
||
'description': r're:^In this episode we explore the unsung heroes of D-Day, the landing craft.',
|
||
'upload_date': '20200327',
|
||
'timestamp': 1585348140,
|
||
'channel': 'Real Engineering',
|
||
'channel_id': 'realengineering',
|
||
'uploader': 'Real Engineering',
|
||
'uploader_id': 'realengineering',
|
||
'view_count': int,
|
||
'series': 'Real Engineering',
|
||
'average_rating': int,
|
||
'display_id': 'the-logistics-of-d-day-landing-craft-how-the-allies-got-ashore',
|
||
'creator': 'Real Engineering',
|
||
'duration': 841,
|
||
'channel_url': 'https://nebula.app/realengineering',
|
||
'uploader_url': 'https://nebula.app/realengineering',
|
||
'thumbnail': r're:https://\w+\.cloudfront\.net/[\w-]+\.jpeg?.*',
|
||
},
|
||
},
|
||
{
|
||
'url': 'https://nebula.app/videos/money-episode-1-the-draw',
|
||
'md5': 'ebe28a7ad822b9ee172387d860487868',
|
||
'info_dict': {
|
||
'id': '5e779ebdd157bc0001d1c75a',
|
||
'ext': 'mp4',
|
||
'title': 'Episode 1: The Draw',
|
||
'description': r'contains:There’s free money on offer… if the players can all work together.',
|
||
'upload_date': '20200323',
|
||
'timestamp': 1584980400,
|
||
'channel': 'Tom Scott Presents: Money',
|
||
'channel_id': 'tom-scott-presents-money',
|
||
'uploader': 'Tom Scott Presents: Money',
|
||
'uploader_id': 'tom-scott-presents-money',
|
||
'uploader_url': 'https://nebula.app/tom-scott-presents-money',
|
||
'duration': 825,
|
||
'channel_url': 'https://nebula.app/tom-scott-presents-money',
|
||
'view_count': int,
|
||
'series': 'Tom Scott Presents: Money',
|
||
'display_id': 'money-episode-1-the-draw',
|
||
'thumbnail': r're:https://\w+\.cloudfront\.net/[\w-]+\.jpeg?.*',
|
||
'average_rating': int,
|
||
'creator': 'Tom Scott Presents: Money',
|
||
},
|
||
},
|
||
{
|
||
'url': 'https://watchnebula.com/videos/money-episode-1-the-draw',
|
||
'only_matching': True,
|
||
},
|
||
]
|
||
|
||
def _fetch_video_metadata(self, slug):
|
||
return self._call_nebula_api(f'https://content.watchnebula.com/video/{slug}/',
|
||
video_id=slug,
|
||
auth_type='bearer',
|
||
note='Fetching video meta data')
|
||
|
||
def _real_extract(self, url):
|
||
slug = self._match_id(url)
|
||
video = self._fetch_video_metadata(slug)
|
||
return self._build_video_info(video)
|
||
|
||
|
||
class NebulaSubscriptionsIE(NebulaBaseIE):
|
||
IE_NAME = 'nebula:subscriptions'
|
||
_VALID_URL = rf'{_BASE_URL_RE}/myshows'
|
||
_TESTS = [
|
||
{
|
||
'url': 'https://nebula.app/myshows',
|
||
'playlist_mincount': 1,
|
||
'info_dict': {
|
||
'id': 'myshows',
|
||
},
|
||
},
|
||
]
|
||
|
||
def _generate_playlist_entries(self):
|
||
next_url = 'https://content.watchnebula.com/library/video/?page_size=100'
|
||
page_num = 1
|
||
while next_url:
|
||
channel = self._call_nebula_api(next_url, 'myshows', auth_type='bearer',
|
||
note=f'Retrieving subscriptions page {page_num}')
|
||
for episode in channel['results']:
|
||
yield self._build_video_info(episode)
|
||
next_url = channel['next']
|
||
page_num += 1
|
||
|
||
def _real_extract(self, url):
|
||
return self.playlist_result(self._generate_playlist_entries(), 'myshows')
|
||
|
||
|
||
class NebulaChannelIE(NebulaBaseIE):
|
||
IE_NAME = 'nebula:channel'
|
||
_VALID_URL = rf'{_BASE_URL_RE}/(?!myshows|videos/)(?P<id>[-\w]+)'
|
||
_TESTS = [
|
||
{
|
||
'url': 'https://nebula.app/tom-scott-presents-money',
|
||
'info_dict': {
|
||
'id': 'tom-scott-presents-money',
|
||
'title': 'Tom Scott Presents: Money',
|
||
'description': 'Tom Scott hosts a series all about trust, negotiation and money.',
|
||
},
|
||
'playlist_count': 5,
|
||
}, {
|
||
'url': 'https://nebula.app/lindsayellis',
|
||
'info_dict': {
|
||
'id': 'lindsayellis',
|
||
'title': 'Lindsay Ellis',
|
||
'description': 'Enjoy these hottest of takes on Disney, Transformers, and Musicals.',
|
||
},
|
||
'playlist_mincount': 100,
|
||
},
|
||
]
|
||
|
||
def _generate_playlist_entries(self, collection_id, channel):
|
||
episodes = channel['episodes']['results']
|
||
for page_num in itertools.count(2):
|
||
for episode in episodes:
|
||
yield self._build_video_info(episode)
|
||
next_url = channel['episodes']['next']
|
||
if not next_url:
|
||
break
|
||
channel = self._call_nebula_api(next_url, collection_id, auth_type='bearer',
|
||
note=f'Retrieving channel page {page_num}')
|
||
episodes = channel['episodes']['results']
|
||
|
||
def _real_extract(self, url):
|
||
collection_id = self._match_id(url)
|
||
channel_url = f'https://content.watchnebula.com/video/channels/{collection_id}/'
|
||
channel = self._call_nebula_api(channel_url, collection_id, auth_type='bearer', note='Retrieving channel')
|
||
channel_details = channel['details']
|
||
|
||
return self.playlist_result(
|
||
entries=self._generate_playlist_entries(collection_id, channel),
|
||
playlist_id=collection_id,
|
||
playlist_title=channel_details['title'],
|
||
playlist_description=channel_details['description']
|
||
)
|