0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-14 01:05:06 +00:00
yt-dlp/yt_dlp/extractor/triller.py

330 lines
13 KiB
Python
Raw Normal View History

import itertools
import json
import re
from .common import InfoExtractor
from ..networking import HEADRequest
from ..utils import (
ExtractorError,
UnsupportedError,
determine_ext,
int_or_none,
parse_resolution,
str_or_none,
traverse_obj,
unified_timestamp,
url_basename,
url_or_none,
urljoin,
)
class TrillerBaseIE(InfoExtractor):
_NETRC_MACHINE = 'triller'
_API_BASE_URL = 'https://social.triller.co/v1.5'
_API_HEADERS = {'Origin': 'https://triller.co'}
def _perform_login(self, username, password):
if self._API_HEADERS.get('Authorization'):
return
headers = {**self._API_HEADERS, 'Content-Type': 'application/json'}
user_check = traverse_obj(self._download_json(
f'{self._API_BASE_URL}/api/user/is-valid-username', None, note='Checking username',
fatal=False, expected_status=400, headers=headers,
data=json.dumps({'username': username}, separators=(',', ':')).encode()), 'status')
if user_check: # endpoint returns `"status":false` if username exists
raise ExtractorError('Unable to login: Invalid username', expected=True)
login = self._download_json(
f'{self._API_BASE_URL}/user/auth', None, note='Logging in', fatal=False,
expected_status=400, headers=headers, data=json.dumps({
'username': username,
'password': password,
}, separators=(',', ':')).encode()) or {}
if not login.get('auth_token'):
if login.get('error') == 1008:
raise ExtractorError('Unable to login: Incorrect password', expected=True)
raise ExtractorError('Unable to login')
self._API_HEADERS['Authorization'] = f'Bearer {login["auth_token"]}'
def _get_comments(self, video_id, limit=15):
comment_info = self._download_json(
f'{self._API_BASE_URL}/api/videos/{video_id}/comments_v2',
video_id, fatal=False, note='Downloading comments API JSON',
headers=self._API_HEADERS, query={'limit': limit}) or {}
if not comment_info.get('comments'):
return
yield from traverse_obj(comment_info, ('comments', ..., {
'id': ('id', {str_or_none}),
'text': 'body',
'author': ('author', 'username'),
'author_id': ('author', 'user_id'),
'timestamp': ('timestamp', {unified_timestamp}),
}))
def _parse_video_info(self, video_info, username, user_id, display_id=None):
video_id = str(video_info['id'])
display_id = display_id or video_info.get('video_uuid')
if traverse_obj(video_info, (
None, ('transcoded_url', 'video_url', 'stream_url', 'audio_url'),
{lambda x: re.search(r'/copyright/', x)}), get_all=False):
self.raise_no_formats('This video has been removed due to licensing restrictions', expected=True)
def format_info(url):
return {
'url': url,
'ext': determine_ext(url),
'format_id': url_basename(url).split('.')[0],
}
formats = []
if determine_ext(video_info.get('transcoded_url')) == 'm3u8':
formats.extend(self._extract_m3u8_formats(
video_info['transcoded_url'], video_id, 'mp4', m3u8_id='hls', fatal=False))
for video in traverse_obj(video_info, ('video_set', lambda _, v: url_or_none(v['url']))):
formats.append({
**format_info(video['url']),
**parse_resolution(video.get('resolution')),
'vcodec': video.get('codec'),
'vbr': int_or_none(video.get('bitrate'), 1000),
})
video_url = traverse_obj(video_info, 'video_url', 'stream_url', expected_type=url_or_none)
if video_url:
formats.append({
**format_info(video_url),
'vcodec': 'h264',
**traverse_obj(video_info, {
'width': 'width',
'height': 'height',
'filesize': 'filesize',
}, expected_type=int_or_none),
})
audio_url = url_or_none(video_info.get('audio_url'))
if audio_url:
formats.append(format_info(audio_url))
comment_count = traverse_obj(video_info, ('comment_count', {int_or_none}))
return {
'id': video_id,
'display_id': display_id,
'uploader': username,
'uploader_id': user_id or traverse_obj(video_info, ('user', 'user_id', {str_or_none})),
'webpage_url': urljoin(f'https://triller.co/@{username}/video/', display_id),
'uploader_url': f'https://triller.co/@{username}',
'extractor_key': TrillerIE.ie_key(),
'extractor': TrillerIE.IE_NAME,
'formats': formats,
'comment_count': comment_count,
'__post_extractor': self.extract_comments(video_id, comment_count),
**traverse_obj(video_info, {
'title': ('description', {lambda x: x.replace('\r\n', ' ')}),
'description': 'description',
'creator': ((('user'), ('users', lambda _, v: str(v['user_id']) == user_id)), 'name'),
'thumbnail': ('thumbnail_url', {url_or_none}),
'timestamp': ('timestamp', {unified_timestamp}),
'duration': ('duration', {int_or_none}),
'view_count': ('play_count', {int_or_none}),
'like_count': ('likes_count', {int_or_none}),
'artist': 'song_artist',
'track': 'song_title',
}, get_all=False),
}
class TrillerIE(TrillerBaseIE):
_VALID_URL = r'''(?x)
https?://(?:www\.)?triller\.co/
@(?P<username>[\w.]+)/video/(?P<id>[\da-f]{8}-(?:[\da-f]{4}-){3}[\da-f]{12})
'''
_TESTS = [{
'url': 'https://triller.co/@theestallion/video/2358fcd7-3df2-4c77-84c8-1d091610a6cf',
'md5': '228662d783923b60d78395fedddc0a20',
'info_dict': {
'id': '71595734',
'ext': 'mp4',
'title': 'md5:9a2bf9435c5c4292678996a464669416',
'thumbnail': r're:^https://uploads\.cdn\.triller\.co/.+\.jpg$',
'description': 'md5:9a2bf9435c5c4292678996a464669416',
'uploader': 'theestallion',
'uploader_id': '18992236',
'creator': 'Megan Thee Stallion',
'timestamp': 1660598222,
'upload_date': '20220815',
'duration': 47,
'view_count': int,
'like_count': int,
'artist': 'Megan Thee Stallion',
'track': 'Her',
'uploader_url': 'https://triller.co/@theestallion',
'comment_count': int,
},
'skip': 'This video has been removed due to licensing restrictions',
}, {
'url': 'https://triller.co/@charlidamelio/video/46c6fcfa-aa9e-4503-a50c-68444f44cddc',
'md5': '874055f462af5b0699b9dbb527a505a0',
'info_dict': {
'id': '71621339',
'ext': 'mp4',
'title': 'md5:4c91ea82760fe0fffb71b8c3aa7295fc',
'display_id': '46c6fcfa-aa9e-4503-a50c-68444f44cddc',
'thumbnail': r're:^https://uploads\.cdn\.triller\.co/.+\.jpg$',
'description': 'md5:4c91ea82760fe0fffb71b8c3aa7295fc',
'uploader': 'charlidamelio',
'uploader_id': '1875551',
'creator': 'charli damelio',
'timestamp': 1660773354,
'upload_date': '20220817',
'duration': 16,
'view_count': int,
'like_count': int,
'artist': 'Dixie',
'track': 'Someone to Blame',
'uploader_url': 'https://triller.co/@charlidamelio',
'comment_count': int,
},
}, {
'url': 'https://triller.co/@theestallion/video/07f35f38-1f51-48e2-8c5f-f7a8e829988f',
'md5': 'af7b3553e4b8bfca507636471ee2eb41',
'info_dict': {
'id': '71837829',
'ext': 'mp4',
'title': 'UNGRATEFUL VIDEO OUT NOW 👏🏾👏🏾👏🏾 💙💙 link my bio #womeninhiphop',
'display_id': '07f35f38-1f51-48e2-8c5f-f7a8e829988f',
'thumbnail': r're:^https://uploads\.cdn\.triller\.co/.+\.jpg$',
'description': 'UNGRATEFUL VIDEO OUT NOW 👏🏾👏🏾👏🏾 💙💙 link my bio\r\n #womeninhiphop',
'uploader': 'theestallion',
'uploader_id': '18992236',
'creator': 'Megan Thee Stallion',
'timestamp': 1662486178,
'upload_date': '20220906',
'duration': 30,
'view_count': int,
'like_count': int,
'artist': 'Unknown',
'track': 'Unknown',
'uploader_url': 'https://triller.co/@theestallion',
'comment_count': int,
},
}]
def _real_extract(self, url):
username, display_id = self._match_valid_url(url).group('username', 'id')
video_info = self._download_json(
f'{self._API_BASE_URL}/api/videos/{display_id}', display_id,
headers=self._API_HEADERS)['videos'][0]
return self._parse_video_info(video_info, username, None, display_id)
class TrillerUserIE(TrillerBaseIE):
_VALID_URL = r'https?://(?:www\.)?triller\.co/@(?P<id>[\w.]+)/?(?:$|[#?])'
_TESTS = [{
'url': 'https://triller.co/@theestallion',
'playlist_mincount': 12,
'info_dict': {
'id': '18992236',
'title': 'theestallion',
'thumbnail': r're:^https://uploads\.cdn\.triller\.co/.+\.jpg$',
},
}, {
'url': 'https://triller.co/@charlidamelio',
'playlist_mincount': 150,
'info_dict': {
'id': '1875551',
'title': 'charlidamelio',
'thumbnail': r're:^https://uploads\.cdn\.triller\.co/.+\.jpg$',
},
}]
def _real_initialize(self):
if not self._API_HEADERS.get('Authorization'):
guest = self._download_json(
f'{self._API_BASE_URL}/user/create_guest', None,
note='Creating guest session', data=b'', headers=self._API_HEADERS, query={
'platform': 'Web',
'app_version': '',
})
if not guest.get('auth_token'):
raise ExtractorError('Unable to fetch required auth token for user extraction')
self._API_HEADERS['Authorization'] = f'Bearer {guest["auth_token"]}'
def _entries(self, username, user_id, limit=6):
query = {'limit': limit}
for page in itertools.count(1):
videos = self._download_json(
f'{self._API_BASE_URL}/api/users/{user_id}/videos',
username, note=f'Downloading user video list page {page}',
headers=self._API_HEADERS, query=query)
for video in traverse_obj(videos, ('videos', ...)):
yield self._parse_video_info(video, username, user_id)
query['before_time'] = traverse_obj(videos, ('videos', -1, 'timestamp'))
if not query['before_time']:
break
def _real_extract(self, url):
username = self._match_id(url)
user_info = traverse_obj(self._download_json(
f'{self._API_BASE_URL}/api/users/by_username/{username}',
username, note='Downloading user info', headers=self._API_HEADERS), ('user', {dict})) or {}
if user_info.get('private') and user_info.get('followed_by_me') not in (True, 'true'):
raise ExtractorError('This user profile is private', expected=True)
elif traverse_obj(user_info, (('blocked_by_user', 'blocking_user'), {bool}), get_all=False):
raise ExtractorError('The author of the video is blocked', expected=True)
user_id = str_or_none(user_info.get('user_id'))
if not user_id:
raise ExtractorError('Unable to extract user ID')
return self.playlist_result(
self._entries(username, user_id), user_id, username, thumbnail=user_info.get('avatar_url'))
class TrillerShortIE(InfoExtractor):
_VALID_URL = r'https?://v\.triller\.co/(?P<id>\w+)'
_TESTS = [{
'url': 'https://v.triller.co/WWZNWk',
'md5': '5eb8dc2c971bd8cd794ec9e8d5e9d101',
'info_dict': {
'id': '66210052',
'ext': 'mp4',
'title': 'md5:2dfc89d154cd91a4a18cd9582ba03e16',
'display_id': 'f4480e1f-fb4e-45b9-a44c-9e6c679ce7eb',
'thumbnail': r're:^https://uploads\.cdn\.triller\.co/.+\.jpg$',
'description': 'md5:2dfc89d154cd91a4a18cd9582ba03e16',
'uploader': 'statefairent',
'uploader_id': '487545193',
'creator': 'Official Summer Fair of LA',
'timestamp': 1629655457,
'upload_date': '20210822',
'duration': 19,
'view_count': int,
'like_count': int,
'artist': 'Unknown',
'track': 'Unknown',
'uploader_url': 'https://triller.co/@statefairent',
'comment_count': int,
},
}]
def _real_extract(self, url):
real_url = self._request_webpage(HEADRequest(url), self._match_id(url)).url
if self.suitable(real_url): # Prevent infinite loop in case redirect fails
raise UnsupportedError(real_url)
return self.url_result(real_url)