[extractor/twitter] Fix unauthenticated extraction (#7476)

Closes #7473
Authored by: bashonly
This commit is contained in:
bashonly 2023-07-05 11:27:36 -05:00 committed by GitHub
parent 1cffd621cb
commit 49296437a8
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 70 additions and 113 deletions

View File

@ -1852,9 +1852,6 @@ #### tiktok
#### rokfinchannel
* `tab`: Which tab to download - one of `new`, `top`, `videos`, `podcasts`, `streams`, `stacks`
#### twitter
* `legacy_api`: Force usage of the legacy Twitter API instead of the GraphQL API for tweet extraction. Has no effect if login cookies are passed
#### stacommu, wrestleuniverse
* `device_id`: UUID value assigned by the website and used to enforce device limits for paid livestream content. Can be found in browser local storage

View File

@ -1,5 +1,6 @@
import json
import re
import urllib.error
from .common import InfoExtractor
from .periscope import PeriscopeBaseIE, PeriscopeIE
@ -34,7 +35,6 @@ class TwitterBaseIE(InfoExtractor):
_GRAPHQL_API_BASE = 'https://twitter.com/i/api/graphql/'
_BASE_REGEX = r'https?://(?:(?:www|m(?:obile)?)\.)?(?:twitter\.com|twitter3e4tixl4xyajtrzo62zg5vztmjuricljdp2c5kshju4avyoid\.onion)/'
_AUTH = {'Authorization': 'Bearer AAAAAAAAAAAAAAAAAAAAANRILgAAAAAAnNwIzUejRCOuH5E6I8xnZz4puTs%3D1Zv7ttfk8LF81IUq16cHjhLTvJu4FA33AGWWjCpTnA'}
_guest_token = None
_flow_token = None
_LOGIN_INIT_DATA = json.dumps({
@ -145,14 +145,6 @@ def _search_dimensions_in_video_url(a_format, video_url):
def is_logged_in(self):
return bool(self._get_cookies(self._API_BASE).get('auth_token'))
def _fetch_guest_token(self, headers, display_id):
headers.pop('x-guest-token', None)
self._guest_token = traverse_obj(self._download_json(
f'{self._API_BASE}guest/activate.json', display_id,
'Downloading guest token', data=b'', headers=headers), 'guest_token')
if not self._guest_token:
raise ExtractorError('Could not retrieve guest token')
def _set_base_headers(self):
headers = self._AUTH.copy()
csrf_token = try_call(lambda: self._get_cookies(self._API_BASE)['ct0'].value)
@ -183,12 +175,15 @@ def _perform_login(self, username, password):
if self.is_logged_in:
return
self._request_webpage('https://twitter.com/', None, 'Requesting cookies')
webpage = self._download_webpage('https://twitter.com/', None, 'Downloading login page')
headers = self._set_base_headers()
self._fetch_guest_token(headers, None)
guest_token = self._search_regex(
r'\.cookie\s*=\s*["\']gt=(\d+);', webpage, 'gt', default=None) or self._download_json(
f'{self._API_BASE}guest/activate.json', None, 'Downloading guest token',
data=b'', headers=headers)['guest_token']
headers.update({
'content-type': 'application/json',
'x-guest-token': self._guest_token,
'x-guest-token': guest_token,
'x-twitter-client-language': 'en',
'x-twitter-active-user': 'yes',
'Referer': 'https://twitter.com/',
@ -285,33 +280,20 @@ def input_dict(subtask_id, text):
self.report_login()
def _call_api(self, path, video_id, query={}, graphql=False):
headers = self._set_base_headers()
if self.is_logged_in:
headers.update({
if not self.is_logged_in:
self.raise_login_required()
result = self._download_json(
(self._GRAPHQL_API_BASE if graphql else self._API_BASE) + path, video_id,
f'Downloading {"GraphQL" if graphql else "legacy API"} JSON', headers={
**self._set_base_headers(),
'x-twitter-auth-type': 'OAuth2Session',
'x-twitter-client-language': 'en',
'x-twitter-active-user': 'yes',
})
for first_attempt in (True, False):
if not self.is_logged_in:
if not self._guest_token:
self._fetch_guest_token(headers, video_id)
headers['x-guest-token'] = self._guest_token
allowed_status = {400, 401, 403, 404} if graphql else {403}
result = self._download_json(
(self._GRAPHQL_API_BASE if graphql else self._API_BASE) + path,
video_id, headers=headers, query=query, expected_status=allowed_status,
note=f'Downloading {"GraphQL" if graphql else "legacy API"} JSON')
}, query=query, expected_status={400, 401, 403, 404} if graphql else {403})
if result.get('errors'):
errors = ', '.join(set(traverse_obj(result, ('errors', ..., 'message', {str}))))
if not self.is_logged_in and first_attempt and 'bad guest token' in errors.lower():
self.to_screen('Guest token has expired. Refreshing guest token')
self._guest_token = None
continue
raise ExtractorError(
f'Error(s) while querying API: {errors or "Unknown error"}', expected=True)
@ -457,6 +439,7 @@ class TwitterIE(TwitterBaseIE):
_VALID_URL = TwitterBaseIE._BASE_REGEX + r'(?:(?:i/web|[^/]+)/status|statuses)/(?P<id>\d+)(?:/(?:video|photo)/(?P<index>\d+))?'
_TESTS = [{
# comment_count, repost_count, view_count are only available with auth (applies to all tests)
'url': 'https://twitter.com/freethenipple/status/643211948184596480',
'info_dict': {
'id': '643211870443208704',
@ -471,10 +454,7 @@ class TwitterIE(TwitterBaseIE):
'timestamp': 1442188653,
'upload_date': '20150913',
'uploader_url': 'https://twitter.com/freethenipple',
'comment_count': int,
'repost_count': int,
'like_count': int,
'view_count': int,
'tags': [],
'age_limit': 18,
},
@ -505,8 +485,6 @@ class TwitterIE(TwitterBaseIE):
'timestamp': 1447395772,
'upload_date': '20151113',
'uploader_url': 'https://twitter.com/starwars',
'comment_count': int,
'repost_count': int,
'like_count': int,
'tags': ['TV', 'StarWars', 'TheForceAwakens'],
'age_limit': 0,
@ -550,10 +528,7 @@ class TwitterIE(TwitterBaseIE):
'timestamp': 1455777459,
'upload_date': '20160218',
'uploader_url': 'https://twitter.com/jaydingeer',
'comment_count': int,
'repost_count': int,
'like_count': int,
'view_count': int,
'tags': ['Damndaniel'],
'age_limit': 0,
},
@ -591,10 +566,7 @@ class TwitterIE(TwitterBaseIE):
'upload_date': '20160412',
'uploader_url': 'https://twitter.com/CaptainAmerica',
'thumbnail': r're:^https?://.*\.jpg',
'comment_count': int,
'repost_count': int,
'like_count': int,
'view_count': int,
'tags': [],
'age_limit': 0,
},
@ -641,10 +613,7 @@ class TwitterIE(TwitterBaseIE):
'timestamp': 1505803395,
'upload_date': '20170919',
'uploader_url': 'https://twitter.com/Prefet971',
'comment_count': int,
'repost_count': int,
'like_count': int,
'view_count': int,
'tags': ['Maria'],
'age_limit': 0,
},
@ -667,10 +636,7 @@ class TwitterIE(TwitterBaseIE):
'timestamp': 1527623489,
'upload_date': '20180529',
'uploader_url': 'https://twitter.com/LisPower1',
'comment_count': int,
'repost_count': int,
'like_count': int,
'view_count': int,
'tags': [],
'age_limit': 0,
},
@ -692,10 +658,7 @@ class TwitterIE(TwitterBaseIE):
'timestamp': 1548184644,
'upload_date': '20190122',
'uploader_url': 'https://twitter.com/Twitter',
'comment_count': int,
'repost_count': int,
'like_count': int,
'view_count': int,
'tags': [],
'age_limit': 0,
},
@ -713,6 +676,7 @@ class TwitterIE(TwitterBaseIE):
'view_count': int,
},
'add_ie': ['TwitterBroadcast'],
'skip': 'Requires authentication',
}, {
# unified card
'url': 'https://twitter.com/BrooklynNets/status/1349794411333394432?s=20',
@ -729,8 +693,6 @@ class TwitterIE(TwitterBaseIE):
'timestamp': 1610651040,
'upload_date': '20210114',
'uploader_url': 'https://twitter.com/BrooklynNets',
'comment_count': int,
'repost_count': int,
'like_count': int,
'tags': [],
'age_limit': 0,
@ -753,10 +715,7 @@ class TwitterIE(TwitterBaseIE):
'thumbnail': r're:^https?://.*\.jpg',
'duration': 30.03,
'timestamp': 1665025050,
'comment_count': int,
'repost_count': int,
'like_count': int,
'view_count': int,
'tags': [],
'age_limit': 0,
},
@ -765,15 +724,13 @@ class TwitterIE(TwitterBaseIE):
'url': 'https://twitter.com/UltimaShadowX/status/1577719286659006464',
'info_dict': {
'id': '1577719286659006464',
'title': 'Ultima | #\u0432\u029f\u043c - Test',
'title': 'Ultima📛 | #вʟм - Test',
'description': 'Test https://t.co/Y3KEZD7Dad',
'uploader': 'Ultima | #\u0432\u029f\u043c',
'uploader': 'Ultima📛 | #вʟм',
'uploader_id': 'UltimaShadowX',
'uploader_url': 'https://twitter.com/UltimaShadowX',
'upload_date': '20221005',
'timestamp': 1664992565,
'comment_count': int,
'repost_count': int,
'like_count': int,
'tags': [],
'age_limit': 0,
@ -795,10 +752,7 @@ class TwitterIE(TwitterBaseIE):
'duration': 21.321,
'timestamp': 1664477766,
'upload_date': '20220929',
'comment_count': int,
'repost_count': int,
'like_count': int,
'view_count': int,
'tags': ['HurricaneIan'],
'age_limit': 0,
},
@ -825,6 +779,20 @@ class TwitterIE(TwitterBaseIE):
},
'skip': 'Requires authentication',
}, {
# Single Vimeo video result without auth
'url': 'https://twitter.com/Srirachachau/status/1395079556562706435',
'info_dict': {
'id': '551578322',
'ext': 'mp4',
'title': 'Dusty & The Mayor',
'uploader': 'Michael Chau',
'uploader_id': 'user29061007',
'uploader_url': 'https://vimeo.com/user29061007',
'duration': 478,
'thumbnail': 'https://i.vimeocdn.com/video/1139658575-0dfdce6e9a2401fe09feb24bf0d14e6f24a53c12f447ff688ace61009ad4c1ba-d_1280',
},
}, {
# Playlist result only with auth
'url': 'https://twitter.com/Srirachachau/status/1395079556562706435',
'playlist_mincount': 2,
'info_dict': {
@ -842,6 +810,7 @@ class TwitterIE(TwitterBaseIE):
'uploader_url': 'https://twitter.com/Srirachachau',
'timestamp': 1621447860,
},
'skip': 'Requires authentication',
}, {
'url': 'https://twitter.com/DavidToons_/status/1578353380363501568',
'playlist_mincount': 2,
@ -860,6 +829,7 @@ class TwitterIE(TwitterBaseIE):
'upload_date': '20221007',
'age_limit': 0,
},
'skip': 'Requires authentication',
}, {
'url': 'https://twitter.com/primevideouk/status/1578401165338976258',
'playlist_count': 2,
@ -873,8 +843,6 @@ class TwitterIE(TwitterBaseIE):
'upload_date': '20221007',
'age_limit': 0,
'uploader_url': 'https://twitter.com/primevideouk',
'comment_count': int,
'repost_count': int,
'like_count': int,
'tags': ['TheRingsOfPower'],
},
@ -896,6 +864,7 @@ class TwitterIE(TwitterBaseIE):
},
'add_ie': ['TwitterSpaces'],
'params': {'skip_download': 'm3u8'},
'skip': 'Requires authentication',
}, {
# URL specifies video number but --yes-playlist
'url': 'https://twitter.com/CTVJLaidlaw/status/1600649710662213632/video/1',
@ -905,9 +874,7 @@ class TwitterIE(TwitterBaseIE):
'title': 'md5:be05989b0722e114103ed3851a0ffae2',
'timestamp': 1670459604.0,
'description': 'md5:591c19ce66fadc2359725d5cd0d1052c',
'comment_count': int,
'uploader_id': 'CTVJLaidlaw',
'repost_count': int,
'tags': ['colorectalcancer', 'cancerjourney', 'imnotaquitter'],
'upload_date': '20221208',
'age_limit': 0,
@ -926,14 +893,11 @@ class TwitterIE(TwitterBaseIE):
'timestamp': 1670459604.0,
'uploader_id': 'CTVJLaidlaw',
'uploader': 'Jocelyn Laidlaw',
'repost_count': int,
'comment_count': int,
'tags': ['colorectalcancer', 'cancerjourney', 'imnotaquitter'],
'duration': 102.226,
'uploader_url': 'https://twitter.com/CTVJLaidlaw',
'display_id': '1600649710662213632',
'like_count': int,
'view_count': int,
'description': 'md5:591c19ce66fadc2359725d5cd0d1052c',
'upload_date': '20221208',
'age_limit': 0,
@ -959,9 +923,6 @@ class TwitterIE(TwitterBaseIE):
'age_limit': 18,
'tags': [],
'like_count': int,
'repost_count': int,
'comment_count': int,
'view_count': int,
},
}, {
'url': 'https://twitter.com/hlo_again/status/1599108751385972737/video/2',
@ -974,10 +935,7 @@ class TwitterIE(TwitterBaseIE):
'like_count': int,
'uploader_id': 'hlo_again',
'thumbnail': 'https://pbs.twimg.com/ext_tw_video_thumb/1599108643743473680/pu/img/UG3xjov4rgg5sbYM.jpg?name=orig',
'repost_count': int,
'duration': 9.531,
'comment_count': int,
'view_count': int,
'upload_date': '20221203',
'age_limit': 0,
'timestamp': 1670092210.0,
@ -994,14 +952,11 @@ class TwitterIE(TwitterBaseIE):
'ext': 'mp4',
'uploader_url': 'https://twitter.com/MunTheShinobi',
'description': 'This is a genius ad by Apple. \U0001f525\U0001f525\U0001f525\U0001f525\U0001f525 https://t.co/cNsA0MoOml',
'view_count': int,
'thumbnail': 'https://pbs.twimg.com/ext_tw_video_thumb/1600009362759733248/pu/img/XVhFQivj75H_YxxV.jpg?name=orig',
'age_limit': 0,
'uploader': 'Mün The Shinobi',
'repost_count': int,
'upload_date': '20221206',
'title': 'Mün The Shinobi - This is a genius ad by Apple. \U0001f525\U0001f525\U0001f525\U0001f525\U0001f525',
'comment_count': int,
'like_count': int,
'tags': [],
'uploader_id': 'MunTheShinobi',
@ -1009,14 +964,14 @@ class TwitterIE(TwitterBaseIE):
'timestamp': 1670306984.0,
},
}, {
# url to retweet id, legacy API
# url to retweet id
'url': 'https://twitter.com/liberdalau/status/1623739803874349067',
'info_dict': {
'id': '1623274794488659969',
'display_id': '1623739803874349067',
'ext': 'mp4',
'title': 'Johnny Bullets - Me after going viral to over 30million people: Whoopsie-daisy',
'description': 'md5:e873616a4a8fe0f93e71872678a672f3',
'description': 'md5:224d62f54b0cdef8e33d4c56c41ac503',
'uploader': 'Johnny Bullets',
'uploader_id': 'Johnnybull3ts',
'uploader_url': 'https://twitter.com/Johnnybull3ts',
@ -1027,10 +982,7 @@ class TwitterIE(TwitterBaseIE):
'upload_date': '20230208',
'thumbnail': r're:https://pbs\.twimg\.com/ext_tw_video_thumb/.+',
'like_count': int,
'repost_count': int,
'comment_count': int,
},
'params': {'extractor_args': {'twitter': {'legacy_api': ['']}}},
}, {
# onion route
'url': 'https://twitter3e4tixl4xyajtrzo62zg5vztmjuricljdp2c5kshju4avyoid.onion/TwitterBlue/status/1484226494708662273',
@ -1081,8 +1033,6 @@ def _graphql_to_legacy(self, data, twid):
if 'tombstone' in result:
cause = remove_end(traverse_obj(result, ('tombstone', 'text', 'text', {str})), '. Learn more')
if cause and 'adult content' in cause:
self.raise_login_required(cause)
raise ExtractorError(f'Twitter API says: {cause or "Unknown error"}', expected=True)
status = result.get('legacy', {})
@ -1138,19 +1088,22 @@ def _build_graphql_query(self, media_id):
def _real_extract(self, url):
twid, selected_index = self._match_valid_url(url).group('id', 'index')
if self._configuration_arg('legacy_api') and not self.is_logged_in:
status = traverse_obj(self._call_api(f'statuses/show/{twid}.json', twid, {
'cards_platform': 'Web-12',
'include_cards': 1,
'include_reply_count': 1,
'include_user_entities': 0,
'tweet_mode': 'extended',
}), 'retweeted_status', None)
if not self.is_logged_in:
try:
status = self._download_json(
'https://cdn.syndication.twimg.com/tweet-result', twid, 'Downloading syndication JSON',
headers={'User-Agent': 'Googlebot'}, query={'id': twid})
self.to_screen(f'Some metadata is missing without authentication. {self._login_hint()}')
except ExtractorError as e:
if isinstance(e.cause, urllib.error.HTTPError) and e.cause.code == 404:
self.raise_login_required('Requested tweet may only be available when logged in')
raise
else:
result = self._call_graphql_api('zZXycP0V6H7m-2r0mOnFcA/TweetDetail', twid)
status = self._graphql_to_legacy(result, twid)
status = self._graphql_to_legacy(
self._call_graphql_api('zZXycP0V6H7m-2r0mOnFcA/TweetDetail', twid), twid)
title = description = status['full_text'].replace('\n', ' ')
title = description = traverse_obj(
status, (('full_text', 'text'), {lambda x: x.replace('\n', ' ')}), get_all=False) or ''
# strip 'https -_t.co_BJYgOjSeGA' junk from filenames
title = re.sub(r'\s+(https?://[^ ]+)', '', title)
user = status.get('user') or {}
@ -1176,12 +1129,16 @@ def _real_extract(self, url):
def extract_from_video_info(media):
media_id = traverse_obj(media, 'id_str', 'id', expected_type=str_or_none)
if not media_id:
# workaround for non-authenticated responses
media_id = traverse_obj(media, (
'video_info', 'variants', ..., 'url',
{lambda x: re.search(r'_video/(\d+)/', x)[1]}), get_all=False)
self.write_debug(f'Extracting from video info: {media_id}')
video_info = media.get('video_info') or {}
formats = []
subtitles = {}
for variant in video_info.get('variants', []):
for variant in traverse_obj(media, ('video_info', 'variants', ...)):
fmts, subs = self._extract_variant_formats(variant, twid)
subtitles = self._merge_subtitles(subtitles, subs)
formats.extend(fmts)
@ -1201,12 +1158,12 @@ def add_thumbnail(name, size):
add_thumbnail('orig', media.get('original_info') or {})
return {
'id': media_id,
'id': media_id or twid,
'formats': formats,
'subtitles': subtitles,
'thumbnails': thumbnails,
'view_count': traverse_obj(media, ('mediaStats', 'viewCount', {int_or_none})),
'duration': float_or_none(video_info.get('duration_millis'), 1000),
'duration': float_or_none(traverse_obj(media, ('video_info', 'duration_millis')), 1000),
# The codec of http formats are unknown
'_format_sort_fields': ('res', 'br', 'size', 'proto'),
}
@ -1286,12 +1243,15 @@ def get_binding_value(k):
}
videos = traverse_obj(status, (
(None, 'quoted_status'), 'extended_entities', 'media', lambda _, m: m['type'] != 'photo', {dict}))
('mediaDetails', ((None, 'quoted_status'), 'extended_entities', 'media')),
lambda _, m: m['type'] != 'photo', {dict}))
if self._yes_playlist(twid, selected_index, video_label='URL-specified video number'):
selected_entries = (*map(extract_from_video_info, videos), *extract_from_card_info(status.get('card')))
else:
desired_obj = traverse_obj(status, ('extended_entities', 'media', int(selected_index) - 1, {dict}))
desired_obj = traverse_obj(status, (
('mediaDetails', ((None, 'quoted_status'), 'extended_entities', 'media')),
int(selected_index) - 1, {dict}), get_all=False)
if not desired_obj:
raise ExtractorError(f'Video #{selected_index} is unavailable', expected=True)
elif desired_obj.get('type') != 'video':