You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
yt-dlp/yt_dlp/extractor/bilibili.py

1233 lines
50 KiB
Python

import base64
import functools
import itertools
import math
import urllib.error
import urllib.parse
from .common import InfoExtractor, SearchInfoExtractor
from ..dependencies import Cryptodome
from ..utils import (
ExtractorError,
GeoRestrictedError,
InAdvancePagedList,
OnDemandPagedList,
filter_dict,
float_or_none,
format_field,
int_or_none,
make_archive_id,
merge_dicts,
mimetype2ext,
parse_count,
parse_qs,
qualities,
smuggle_url,
srt_subtitles_timecode,
str_or_none,
traverse_obj,
unsmuggle_url,
url_or_none,
urlencode_postdata,
)
class BilibiliBaseIE(InfoExtractor):
def extract_formats(self, play_info):
format_names = {
r['quality']: traverse_obj(r, 'new_description', 'display_desc')
for r in traverse_obj(play_info, ('support_formats', lambda _, v: v['quality']))
}
audios = traverse_obj(play_info, ('dash', 'audio', ...))
flac_audio = traverse_obj(play_info, ('dash', 'flac', 'audio'))
if flac_audio:
audios.append(flac_audio)
formats = [{
'url': traverse_obj(audio, 'baseUrl', 'base_url', 'url'),
'ext': mimetype2ext(traverse_obj(audio, 'mimeType', 'mime_type')),
'acodec': audio.get('codecs'),
'vcodec': 'none',
'tbr': float_or_none(audio.get('bandwidth'), scale=1000),
'filesize': int_or_none(audio.get('size'))
} for audio in audios]
formats.extend({
'url': traverse_obj(video, 'baseUrl', 'base_url', 'url'),
'ext': mimetype2ext(traverse_obj(video, 'mimeType', 'mime_type')),
'fps': float_or_none(traverse_obj(video, 'frameRate', 'frame_rate')),
'width': int_or_none(video.get('width')),
'height': int_or_none(video.get('height')),
'vcodec': video.get('codecs'),
'acodec': 'none' if audios else None,
'tbr': float_or_none(video.get('bandwidth'), scale=1000),
'filesize': int_or_none(video.get('size')),
'quality': int_or_none(video.get('id')),
'format': format_names.get(video.get('id')),
} for video in traverse_obj(play_info, ('dash', 'video', ...)))
missing_formats = format_names.keys() - set(traverse_obj(formats, (..., 'quality')))
if missing_formats:
self.to_screen(f'Format(s) {", ".join(format_names[i] for i in missing_formats)} are missing; '
f'you have to login or become premium member to download them. {self._login_hint()}')
return formats
def json2srt(self, json_data):
srt_data = ''
for idx, line in enumerate(json_data.get('body') or []):
srt_data += (f'{idx + 1}\n'
f'{srt_subtitles_timecode(line["from"])} --> {srt_subtitles_timecode(line["to"])}\n'
f'{line["content"]}\n\n')
return srt_data
def _get_subtitles(self, video_id, initial_state, cid):
subtitles = {
'danmaku': [{
'ext': 'xml',
'url': f'https://comment.bilibili.com/{cid}.xml',
}]
}
for s in traverse_obj(initial_state, ('videoData', 'subtitle', 'list')) or []:
subtitles.setdefault(s['lan'], []).append({
'ext': 'srt',
'data': self.json2srt(self._download_json(s['subtitle_url'], video_id))
})
return subtitles
def _get_chapters(self, aid, cid):
chapters = aid and cid and self._download_json(
'https://api.bilibili.com/x/player/v2', aid, query={'aid': aid, 'cid': cid},
note='Extracting chapters', fatal=False)
return traverse_obj(chapters, ('data', 'view_points', ..., {
'title': 'content',
'start_time': 'from',
'end_time': 'to',
})) or None
def _get_comments(self, aid):
for idx in itertools.count(1):
replies = traverse_obj(
self._download_json(
f'https://api.bilibili.com/x/v2/reply?pn={idx}&oid={aid}&type=1&jsonp=jsonp&sort=2&_=1567227301685',
aid, note=f'Extracting comments from page {idx}', fatal=False),
('data', 'replies'))
if not replies:
return
for children in map(self._get_all_children, replies):
yield from children
def _get_all_children(self, reply):
yield {
'author': traverse_obj(reply, ('member', 'uname')),
'author_id': traverse_obj(reply, ('member', 'mid')),
'id': reply.get('rpid'),
'text': traverse_obj(reply, ('content', 'message')),
'timestamp': reply.get('ctime'),
'parent': reply.get('parent') or 'root',
}
for children in map(self._get_all_children, traverse_obj(reply, ('replies', ...))):
yield from children
class BiliBiliIE(BilibiliBaseIE):
_VALID_URL = r'https?://www\.bilibili\.com/video/[aAbB][vV](?P<id>[^/?#&]+)'
_TESTS = [{
'url': 'https://www.bilibili.com/video/BV13x41117TL',
'info_dict': {
'id': 'BV13x41117TL',
'title': '阿滴英文|英文歌分享#6 "Closer',
'ext': 'mp4',
'description': '滴妹今天唱Closer給你聽! 有史以来,被推最多次也是最久的歌曲,其实歌词跟我原本想像差蛮多的,不过还是好听! 微博@阿滴英文',
'uploader_id': '65880958',
'uploader': '阿滴英文',
'thumbnail': r're:^https?://.*\.(jpg|jpeg|png)$',
'duration': 554.117,
'tags': list,
'comment_count': int,
'upload_date': '20170301',
'timestamp': 1488353834,
'like_count': int,
'view_count': int,
},
}, {
# old av URL version
'url': 'http://www.bilibili.com/video/av1074402/',
'info_dict': {
'thumbnail': r're:^https?://.*\.(jpg|jpeg)$',
'ext': 'mp4',
'uploader': '菊子桑',
'uploader_id': '156160',
'id': 'BV11x411K7CN',
'title': '【金坷垃】金泡沫',
'duration': 308.36,
'upload_date': '20140420',
'timestamp': 1397983878,
'description': 'md5:ce18c2a2d2193f0df2917d270f2e5923',
'like_count': int,
'comment_count': int,
'view_count': int,
'tags': list,
},
'params': {'skip_download': True},
}, {
'note': 'Anthology',
'url': 'https://www.bilibili.com/video/BV1bK411W797',
'info_dict': {
'id': 'BV1bK411W797',
'title': '物语中的人物是如何吐槽自己的OP的'
},
'playlist_count': 18,
'playlist': [{
'info_dict': {
'id': 'BV1bK411W797_p1',
'ext': 'mp4',
'title': '物语中的人物是如何吐槽自己的OP的 p01 Staple Stable/战场原+羽川',
'tags': 'count:11',
'timestamp': 1589601697,
'thumbnail': r're:^https?://.*\.(jpg|jpeg|png)$',
'uploader': '打牌还是打桩',
'uploader_id': '150259984',
'like_count': int,
'comment_count': int,
'upload_date': '20200516',
'view_count': int,
'description': 'md5:e3c401cf7bc363118d1783dd74068a68',
'duration': 90.314,
}
}]
}, {
'note': 'Specific page of Anthology',
'url': 'https://www.bilibili.com/video/BV1bK411W797?p=1',
'info_dict': {
'id': 'BV1bK411W797_p1',
'ext': 'mp4',
'title': '物语中的人物是如何吐槽自己的OP的 p01 Staple Stable/战场原+羽川',
'tags': 'count:11',
'timestamp': 1589601697,
'thumbnail': r're:^https?://.*\.(jpg|jpeg|png)$',
'uploader': '打牌还是打桩',
'uploader_id': '150259984',
'like_count': int,
'comment_count': int,
'upload_date': '20200516',
'view_count': int,
'description': 'md5:e3c401cf7bc363118d1783dd74068a68',
'duration': 90.314,
}
}, {
'note': 'video has subtitles',
'url': 'https://www.bilibili.com/video/BV12N4y1M7rh',
'info_dict': {
'id': 'BV12N4y1M7rh',
'ext': 'mp4',
'title': 'md5:96e8bb42c2b432c0d4ce3434a61479c1',
'tags': list,
'description': 'md5:afde2b7ba9025c01d9e3dde10de221e4',
'duration': 313.557,
'upload_date': '20220709',
'uploader': '小夫Tech',
'timestamp': 1657347907,
'uploader_id': '1326814124',
'comment_count': int,
'view_count': int,
'like_count': int,
'thumbnail': r're:^https?://.*\.(jpg|jpeg|png)$',
'subtitles': 'count:2'
},
'params': {'listsubtitles': True},
}, {
'url': 'https://www.bilibili.com/video/av8903802/',
'info_dict': {
'id': 'BV13x41117TL',
'ext': 'mp4',
'title': '阿滴英文|英文歌分享#6 "Closer',
'upload_date': '20170301',
'description': 'md5:3b1b9e25b78da4ef87e9b548b88ee76a',
'timestamp': 1488353834,
'uploader_id': '65880958',
'uploader': '阿滴英文',
'thumbnail': r're:^https?://.*\.(jpg|jpeg|png)$',
'duration': 554.117,
'tags': list,
'comment_count': int,
'view_count': int,
'like_count': int,
},
'params': {
'skip_download': True,
},
}, {
'note': 'video has chapter',
'url': 'https://www.bilibili.com/video/BV1vL411G7N7/',
'info_dict': {
'id': 'BV1vL411G7N7',
'ext': 'mp4',
'title': '如何为你的B站视频添加进度条分段',
'timestamp': 1634554558,
'upload_date': '20211018',
'description': 'md5:a9a3d6702b3a94518d419b2e9c320a6d',
'tags': list,
'uploader': '爱喝咖啡的当麻',
'duration': 669.482,
'uploader_id': '1680903',
'chapters': 'count:6',
'comment_count': int,
'view_count': int,
'like_count': int,
'thumbnail': r're:^https?://.*\.(jpg|jpeg|png)$',
},
'params': {'skip_download': True},
}]
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
initial_state = self._search_json(r'window\.__INITIAL_STATE__\s*=', webpage, 'initial state', video_id)
play_info = self._search_json(r'window\.__playinfo__\s*=', webpage, 'play info', video_id)['data']
video_data = initial_state['videoData']
video_id, title = video_data['bvid'], video_data.get('title')
# Bilibili anthologies are similar to playlists but all videos share the same video ID as the anthology itself.
page_list_json = traverse_obj(
self._download_json(
'https://api.bilibili.com/x/player/pagelist', video_id,
fatal=False, query={'bvid': video_id, 'jsonp': 'jsonp'},
note='Extracting videos in anthology'),
'data', expected_type=list) or []
is_anthology = len(page_list_json) > 1
part_id = int_or_none(parse_qs(url).get('p', [None])[-1])
if is_anthology and not part_id and self._yes_playlist(video_id, video_id):
return self.playlist_from_matches(
page_list_json, video_id, title, ie=BiliBiliIE,
getter=lambda entry: f'https://www.bilibili.com/video/{video_id}?p={entry["page"]}')
if is_anthology:
part_id = part_id or 1
title += f' p{part_id:02d} {traverse_obj(page_list_json, (part_id - 1, "part")) or ""}'
aid = video_data.get('aid')
old_video_id = format_field(aid, None, f'%s_part{part_id or 1}')
cid = traverse_obj(video_data, ('pages', part_id - 1, 'cid')) if part_id else video_data.get('cid')
return {
'id': f'{video_id}{format_field(part_id, None, "_p%d")}',
'formats': self.extract_formats(play_info),
'_old_archive_ids': [make_archive_id(self, old_video_id)] if old_video_id else None,
'title': title,
'description': traverse_obj(initial_state, ('videoData', 'desc')),
'view_count': traverse_obj(initial_state, ('videoData', 'stat', 'view')),
'uploader': traverse_obj(initial_state, ('upData', 'name')),
'uploader_id': traverse_obj(initial_state, ('upData', 'mid')),
'like_count': traverse_obj(initial_state, ('videoData', 'stat', 'like')),
'comment_count': traverse_obj(initial_state, ('videoData', 'stat', 'reply')),
'tags': traverse_obj(initial_state, ('tags', ..., 'tag_name')),
'thumbnail': traverse_obj(initial_state, ('videoData', 'pic')),
'timestamp': traverse_obj(initial_state, ('videoData', 'pubdate')),
'duration': float_or_none(play_info.get('timelength'), scale=1000),
'chapters': self._get_chapters(aid, cid),
'subtitles': self.extract_subtitles(video_id, initial_state, cid),
'__post_extractor': self.extract_comments(aid),
'http_headers': {'Referer': url},
}
class BiliBiliBangumiIE(BilibiliBaseIE):
_VALID_URL = r'(?x)https?://www\.bilibili\.com/bangumi/play/(?P<id>(?:ss|ep)\d+)'
_TESTS = [{
'url': 'https://www.bilibili.com/bangumi/play/ss897',
'info_dict': {
'id': 'ss897',
'ext': 'mp4',
'series': '神的记事本',
'season': '神的记事本',
'season_id': 897,
'season_number': 1,
'episode': '你与旅行包',
'episode_number': 2,
'title': '神的记事本第2话 你与旅行包',
'duration': 1428.487,
'timestamp': 1310809380,
'upload_date': '20110716',
'thumbnail': r're:^https?://.*\.(jpg|jpeg|png)$',
},
}, {
'url': 'https://www.bilibili.com/bangumi/play/ep508406',
'only_matching': True,
}]
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
if '您所在的地区无法观看本片' in webpage:
raise GeoRestrictedError('This video is restricted')
elif ('开通大会员观看' in webpage and '__playinfo__' not in webpage
or '正在观看预览,大会员免费看全片' in webpage):
self.raise_login_required('This video is for premium members only')
play_info = self._search_json(r'window\.__playinfo__\s*=', webpage, 'play info', video_id)['data']
formats = self.extract_formats(play_info)
if (not formats and '成为大会员抢先看' in webpage
and play_info.get('durl') and not play_info.get('dash')):
self.raise_login_required('This video is for premium members only')
initial_state = self._search_json(r'window\.__INITIAL_STATE__\s*=', webpage, 'initial state', video_id)
season_id = traverse_obj(initial_state, ('mediaInfo', 'season_id'))
season_number = season_id and next((
idx + 1 for idx, e in enumerate(
traverse_obj(initial_state, ('mediaInfo', 'seasons', ...)))
if e.get('season_id') == season_id
), None)
return {
'id': video_id,
'formats': formats,
'title': traverse_obj(initial_state, 'h1Title'),
'episode': traverse_obj(initial_state, ('epInfo', 'long_title')),
'episode_number': int_or_none(traverse_obj(initial_state, ('epInfo', 'title'))),
'series': traverse_obj(initial_state, ('mediaInfo', 'series')),
'season': traverse_obj(initial_state, ('mediaInfo', 'season_title')),
'season_id': season_id,
'season_number': season_number,
'thumbnail': traverse_obj(initial_state, ('epInfo', 'cover')),
'timestamp': traverse_obj(initial_state, ('epInfo', 'pub_time')),
'duration': float_or_none(play_info.get('timelength'), scale=1000),
'subtitles': self.extract_subtitles(
video_id, initial_state, traverse_obj(initial_state, ('epInfo', 'cid'))),
'__post_extractor': self.extract_comments(traverse_obj(initial_state, ('epInfo', 'aid'))),
'http_headers': {'Referer': url, **self.geo_verification_headers()},
}
class BiliBiliBangumiMediaIE(InfoExtractor):
_VALID_URL = r'https?://www\.bilibili\.com/bangumi/media/md(?P<id>\d+)'
_TESTS = [{
'url': 'https://www.bilibili.com/bangumi/media/md24097891',
'info_dict': {
'id': '24097891',
},
'playlist_mincount': 25,
}]
def _real_extract(self, url):
media_id = self._match_id(url)
webpage = self._download_webpage(url, media_id)
initial_state = self._search_json(r'window\.__INITIAL_STATE__\s*=', webpage, 'initial_state', media_id)
episode_list = self._download_json(
'https://api.bilibili.com/pgc/web/season/section', media_id,
query={'season_id': initial_state['mediaInfo']['season_id']},
note='Downloading season info')['result']['main_section']['episodes']
return self.playlist_result((
self.url_result(entry['share_url'], BiliBiliBangumiIE, entry['aid'])
for entry in episode_list), media_id)
class BilibiliSpaceBaseIE(InfoExtractor):
def _extract_playlist(self, fetch_page, get_metadata, get_entries):
first_page = fetch_page(0)
metadata = get_metadata(first_page)
paged_list = InAdvancePagedList(
lambda idx: get_entries(fetch_page(idx) if idx else first_page),
metadata['page_count'], metadata['page_size'])
return metadata, paged_list
class BilibiliSpaceVideoIE(BilibiliSpaceBaseIE):
_VALID_URL = r'https?://space\.bilibili\.com/(?P<id>\d+)(?P<video>/video)?/?(?:[?#]|$)'
_TESTS = [{
'url': 'https://space.bilibili.com/3985676/video',
'info_dict': {
'id': '3985676',
},
'playlist_mincount': 178,
}]
def _real_extract(self, url):
playlist_id, is_video_url = self._match_valid_url(url).group('id', 'video')
if not is_video_url:
self.to_screen('A channel URL was given. Only the channel\'s videos will be downloaded. '
'To download audios, add a "/audio" to the URL')
def fetch_page(page_idx):
try:
response = self._download_json('https://api.bilibili.com/x/space/arc/search',
playlist_id, note=f'Downloading page {page_idx}',
query={'mid': playlist_id, 'pn': page_idx + 1, 'jsonp': 'jsonp'})
except ExtractorError as e:
if isinstance(e.cause, urllib.error.HTTPError) and e.cause.code == 412:
raise ExtractorError(
'Request is blocked by server (412), please add cookies, wait and try later.', expected=True)
raise
if response['code'] == -401:
raise ExtractorError(
'Request is blocked by server (401), please add cookies, wait and try later.', expected=True)
return response['data']
def get_metadata(page_data):
page_size = page_data['page']['ps']
entry_count = page_data['page']['count']
return {
'page_count': math.ceil(entry_count / page_size),
'page_size': page_size,
}
def get_entries(page_data):
for entry in traverse_obj(page_data, ('list', 'vlist')) or []:
yield self.url_result(f'https://www.bilibili.com/video/{entry["bvid"]}', BiliBiliIE, entry['bvid'])
metadata, paged_list = self._extract_playlist(fetch_page, get_metadata, get_entries)
return self.playlist_result(paged_list, playlist_id)
class BilibiliSpaceAudioIE(BilibiliSpaceBaseIE):
_VALID_URL = r'https?://space\.bilibili\.com/(?P<id>\d+)/audio'
_TESTS = [{
'url': 'https://space.bilibili.com/3985676/audio',
'info_dict': {
'id': '3985676',
},
'playlist_mincount': 1,
}]
def _real_extract(self, url):
playlist_id = self._match_id(url)
def fetch_page(page_idx):
return self._download_json(
'https://api.bilibili.com/audio/music-service/web/song/upper', playlist_id,
note=f'Downloading page {page_idx}',
query={'uid': playlist_id, 'pn': page_idx + 1, 'ps': 30, 'order': 1, 'jsonp': 'jsonp'})['data']
def get_metadata(page_data):
return {
'page_count': page_data['pageCount'],
'page_size': page_data['pageSize'],
}
def get_entries(page_data):
for entry in page_data.get('data', []):
yield self.url_result(f'https://www.bilibili.com/audio/au{entry["id"]}', BilibiliAudioIE, entry['id'])
metadata, paged_list = self._extract_playlist(fetch_page, get_metadata, get_entries)
return self.playlist_result(paged_list, playlist_id)
class BilibiliSpacePlaylistIE(BilibiliSpaceBaseIE):
_VALID_URL = r'https?://space.bilibili\.com/(?P<mid>\d+)/channel/collectiondetail\?sid=(?P<sid>\d+)'
_TESTS = [{
'url': 'https://space.bilibili.com/2142762/channel/collectiondetail?sid=57445',
'info_dict': {
'id': '2142762_57445',
'title': '《底特律 变人》'
},
'playlist_mincount': 31,
}]
def _real_extract(self, url):
mid, sid = self._match_valid_url(url).group('mid', 'sid')
playlist_id = f'{mid}_{sid}'
def fetch_page(page_idx):
return self._download_json(
'https://api.bilibili.com/x/polymer/space/seasons_archives_list',
playlist_id, note=f'Downloading page {page_idx}',
query={'mid': mid, 'season_id': sid, 'page_num': page_idx + 1, 'page_size': 30})['data']
def get_metadata(page_data):
page_size = page_data['page']['page_size']
entry_count = page_data['page']['total']
return {
'page_count': math.ceil(entry_count / page_size),
'page_size': page_size,
'title': traverse_obj(page_data, ('meta', 'name'))
}
def get_entries(page_data):
for entry in page_data.get('archives', []):
yield self.url_result(f'https://www.bilibili.com/video/{entry["bvid"]}',
BiliBiliIE, entry['bvid'])
metadata, paged_list = self._extract_playlist(fetch_page, get_metadata, get_entries)
return self.playlist_result(paged_list, playlist_id, metadata['title'])
class BilibiliCategoryIE(InfoExtractor):
IE_NAME = 'Bilibili category extractor'
_MAX_RESULTS = 1000000
_VALID_URL = r'https?://www\.bilibili\.com/v/[a-zA-Z]+\/[a-zA-Z]+'
_TESTS = [{
'url': 'https://www.bilibili.com/v/kichiku/mad',
'info_dict': {
'id': 'kichiku: mad',
'title': 'kichiku: mad'
},
'playlist_mincount': 45,
'params': {
'playlistend': 45
}
}]
def _fetch_page(self, api_url, num_pages, query, page_num):
parsed_json = self._download_json(
api_url, query, query={'Search_key': query, 'pn': page_num},
note='Extracting results from page %s of %s' % (page_num, num_pages))
video_list = traverse_obj(parsed_json, ('data', 'archives'), expected_type=list)
if not video_list:
raise ExtractorError('Failed to retrieve video list for page %d' % page_num)
for video in video_list:
yield self.url_result(
'https://www.bilibili.com/video/%s' % video['bvid'], 'BiliBili', video['bvid'])
def _entries(self, category, subcategory, query):
# map of categories : subcategories : RIDs
rid_map = {
'kichiku': {
'mad': 26,
'manual_vocaloid': 126,
'guide': 22,
'theatre': 216,
'course': 127
},
}
if category not in rid_map:
raise ExtractorError(
f'The category {category} isn\'t supported. Supported categories: {list(rid_map.keys())}')
if subcategory not in rid_map[category]:
raise ExtractorError(
f'The subcategory {subcategory} isn\'t supported for this category. Supported subcategories: {list(rid_map[category].keys())}')
rid_value = rid_map[category][subcategory]
api_url = 'https://api.bilibili.com/x/web-interface/newlist?rid=%d&type=1&ps=20&jsonp=jsonp' % rid_value
page_json = self._download_json(api_url, query, query={'Search_key': query, 'pn': '1'})
page_data = traverse_obj(page_json, ('data', 'page'), expected_type=dict)
count, size = int_or_none(page_data.get('count')), int_or_none(page_data.get('size'))
if count is None or not size:
raise ExtractorError('Failed to calculate either page count or size')
num_pages = math.ceil(count / size)
return OnDemandPagedList(functools.partial(
self._fetch_page, api_url, num_pages, query), size)
def _real_extract(self, url):
category, subcategory = urllib.parse.urlparse(url).path.split('/')[2:4]
query = '%s: %s' % (category, subcategory)
return self.playlist_result(self._entries(category, subcategory, query), query, query)
class BiliBiliSearchIE(SearchInfoExtractor):
IE_DESC = 'Bilibili video search'
_MAX_RESULTS = 100000
_SEARCH_KEY = 'bilisearch'
def _search_results(self, query):
for page_num in itertools.count(1):
videos = self._download_json(
'https://api.bilibili.com/x/web-interface/search/type', query,
note=f'Extracting results from page {page_num}', query={
'Search_key': query,
'keyword': query,
'page': page_num,
'context': '',
'duration': 0,
'tids_2': '',
'__refresh__': 'true',
'search_type': 'video',
'tids': 0,
'highlight': 1,
})['data'].get('result')
if not videos:
break
for video in videos:
yield self.url_result(video['arcurl'], 'BiliBili', str(video['aid']))
class BilibiliAudioBaseIE(InfoExtractor):
def _call_api(self, path, sid, query=None):
if not query:
query = {'sid': sid}
return self._download_json(
'https://www.bilibili.com/audio/music-service-c/web/' + path,
sid, query=query)['data']
class BilibiliAudioIE(BilibiliAudioBaseIE):
_VALID_URL = r'https?://(?:www\.)?bilibili\.com/audio/au(?P<id>\d+)'
_TEST = {
'url': 'https://www.bilibili.com/audio/au1003142',
'md5': 'fec4987014ec94ef9e666d4d158ad03b',
'info_dict': {
'id': '1003142',
'ext': 'm4a',
'title': '【tsukimi】YELLOW / 神山羊',
'artist': 'tsukimi',
'comment_count': int,
'description': 'YELLOW的mp3版',
'duration': 183,
'subtitles': {
'origin': [{
'ext': 'lrc',
}],
},
'thumbnail': r're:^https?://.+\.jpg',
'timestamp': 1564836614,
'upload_date': '20190803',
'uploader': 'tsukimi-つきみぐー',
'view_count': int,
},
}
def _real_extract(self, url):
au_id = self._match_id(url)
play_data = self._call_api('url', au_id)
formats = [{
'url': play_data['cdns'][0],
'filesize': int_or_none(play_data.get('size')),
'vcodec': 'none'
}]
for a_format in formats:
a_format.setdefault('http_headers', {}).update({
'Referer': url,
})
song = self._call_api('song/info', au_id)
title = song['title']
statistic = song.get('statistic') or {}
subtitles = None
lyric = song.get('lyric')
if lyric:
subtitles = {
'origin': [{
'url': lyric,
}]
}
return {
'id': au_id,
'title': title,
'formats': formats,
'artist': song.get('author'),
'comment_count': int_or_none(statistic.get('comment')),
'description': song.get('intro'),
'duration': int_or_none(song.get('duration')),
'subtitles': subtitles,
'thumbnail': song.get('cover'),
'timestamp': int_or_none(song.get('passtime')),
'uploader': song.get('uname'),
'view_count': int_or_none(statistic.get('play')),
}
class BilibiliAudioAlbumIE(BilibiliAudioBaseIE):
_VALID_URL = r'https?://(?:www\.)?bilibili\.com/audio/am(?P<id>\d+)'
_TEST = {
'url': 'https://www.bilibili.com/audio/am10624',
'info_dict': {
'id': '10624',
'title': '每日新曲推荐每日11:00更新',
'description': '每天11:00更新为你推送最新音乐',
},
'playlist_count': 19,
}
def _real_extract(self, url):
am_id = self._match_id(url)
songs = self._call_api(
'song/of-menu', am_id, {'sid': am_id, 'pn': 1, 'ps': 100})['data']
entries = []
for song in songs:
sid = str_or_none(song.get('id'))
if not sid:
continue
entries.append(self.url_result(
'https://www.bilibili.com/audio/au' + sid,
BilibiliAudioIE.ie_key(), sid))
if entries:
album_data = self._call_api('menu/info', am_id) or {}
album_title = album_data.get('title')
if album_title:
for entry in entries:
entry['album'] = album_title
return self.playlist_result(
entries, am_id, album_title, album_data.get('intro'))
return self.playlist_result(entries, am_id)
class BiliBiliPlayerIE(InfoExtractor):
_VALID_URL = r'https?://player\.bilibili\.com/player\.html\?.*?\baid=(?P<id>\d+)'
_TEST = {
'url': 'http://player.bilibili.com/player.html?aid=92494333&cid=157926707&page=1',
'only_matching': True,
}
def _real_extract(self, url):
video_id = self._match_id(url)
return self.url_result(
'http://www.bilibili.tv/video/av%s/' % video_id,
ie=BiliBiliIE.ie_key(), video_id=video_id)
class BiliIntlBaseIE(InfoExtractor):
_API_URL = 'https://api.bilibili.tv/intl/gateway'
_NETRC_MACHINE = 'biliintl'
def _call_api(self, endpoint, *args, **kwargs):
json = self._download_json(self._API_URL + endpoint, *args, **kwargs)
if json.get('code'):
if json['code'] in (10004004, 10004005, 10023006):
self.raise_login_required()
elif json['code'] == 10004001:
self.raise_geo_restricted()
else:
if json.get('message') and str(json['code']) != json['message']:
errmsg = f'{kwargs.get("errnote", "Unable to download JSON metadata")}: {self.IE_NAME} said: {json["message"]}'
else:
errmsg = kwargs.get('errnote', 'Unable to download JSON metadata')
if kwargs.get('fatal'):
raise ExtractorError(errmsg)
else:
self.report_warning(errmsg)
return json.get('data')
def json2srt(self, json):
data = '\n\n'.join(
f'{i + 1}\n{srt_subtitles_timecode(line["from"])} --> {srt_subtitles_timecode(line["to"])}\n{line["content"]}'
for i, line in enumerate(traverse_obj(json, (
'body', lambda _, l: l['content'] and l['from'] and l['to']))))
return data
def _get_subtitles(self, *, ep_id=None, aid=None):
sub_json = self._call_api(
'/web/v2/subtitle', ep_id or aid, fatal=False,
note='Downloading subtitles list', errnote='Unable to download subtitles list',
query=filter_dict({
'platform': 'web',
's_locale': 'en_US',
'episode_id': ep_id,
'aid': aid,
})) or {}
subtitles = {}
for sub in sub_json.get('subtitles') or []:
sub_url = sub.get('url')
if not sub_url:
continue
sub_data = self._download_json(
sub_url, ep_id or aid, errnote='Unable to download subtitles', fatal=False,
note='Downloading subtitles%s' % f' for {sub["lang"]}' if sub.get('lang') else '')
if not sub_data:
continue
subtitles.setdefault(sub.get('lang_key', 'en'), []).append({
'ext': 'srt',
'data': self.json2srt(sub_data)
})
return subtitles
def _get_formats(self, *, ep_id=None, aid=None):
video_json = self._call_api(
'/web/playurl', ep_id or aid, note='Downloading video formats',
errnote='Unable to download video formats', query=filter_dict({
'platform': 'web',
'ep_id': ep_id,
'aid': aid,
}))
video_json = video_json['playurl']
formats = []
for vid in video_json.get('video') or []:
video_res = vid.get('video_resource') or {}
video_info = vid.get('stream_info') or {}
if not video_res.get('url'):
continue
formats.append({
'url': video_res['url'],
'ext': 'mp4',
'format_note': video_info.get('desc_words'),
'width': video_res.get('width'),
'height': video_res.get('height'),
'vbr': video_res.get('bandwidth'),
'acodec': 'none',
'vcodec': video_res.get('codecs'),
'filesize': video_res.get('size'),
})
for aud in video_json.get('audio_resource') or []:
if not aud.get('url'):
continue
formats.append({
'url': aud['url'],
'ext': 'mp4',
'abr': aud.get('bandwidth'),
'acodec': aud.get('codecs'),
'vcodec': 'none',
'filesize': aud.get('size'),
})
return formats
def _parse_video_metadata(self, video_data):
return {
'title': video_data.get('title_display') or video_data.get('title'),
'thumbnail': video_data.get('cover'),
'episode_number': int_or_none(self._search_regex(
r'^E(\d+)(?:$| - )', video_data.get('title_display') or '', 'episode number', default=None)),
}
def _perform_login(self, username, password):
if not Cryptodome:
raise ExtractorError('pycryptodomex not found. Please install', expected=True)
key_data = self._download_json(
'https://passport.bilibili.tv/x/intl/passport-login/web/key?lang=en-US', None,
note='Downloading login key', errnote='Unable to download login key')['data']
public_key = Cryptodome.PublicKey.RSA.importKey(key_data['key'])
password_hash = Cryptodome.Cipher.PKCS1_v1_5.new(public_key).encrypt((key_data['hash'] + password).encode('utf-8'))
login_post = self._download_json(
'https://passport.bilibili.tv/x/intl/passport-login/web/login/password?lang=en-US', None, data=urlencode_postdata({
'username': username,
'password': base64.b64encode(password_hash).decode('ascii'),
'keep_me': 'true',
's_locale': 'en_US',
'isTrusted': 'true'
}), note='Logging in', errnote='Unable to log in')
if login_post.get('code'):
if login_post.get('message'):
raise ExtractorError(f'Unable to log in: {self.IE_NAME} said: {login_post["message"]}', expected=True)
else:
raise ExtractorError('Unable to log in')
class BiliIntlIE(BiliIntlBaseIE):
_VALID_URL = r'https?://(?:www\.)?bili(?:bili\.tv|intl\.com)/(?:[a-zA-Z]{2}/)?(play/(?P<season_id>\d+)/(?P<ep_id>\d+)|video/(?P<aid>\d+))'
_TESTS = [{
# Bstation page
'url': 'https://www.bilibili.tv/en/play/34613/341736',
'info_dict': {
'id': '341736',
'ext': 'mp4',
'title': 'E2 - The First Night',
'thumbnail': r're:^https://pic\.bstarstatic\.com/ogv/.+\.png$',
'episode_number': 2,
'upload_date': '20201009',
'episode': 'Episode 2',
'timestamp': 1602259500,
'description': 'md5:297b5a17155eb645e14a14b385ab547e',
'chapters': [{
'start_time': 0,
'end_time': 76.242,
'title': '<Untitled Chapter 1>'
}, {
'start_time': 76.242,
'end_time': 161.161,
'title': 'Intro'
}, {
'start_time': 1325.742,
'end_time': 1403.903,
'title': 'Outro'
}],
}
}, {
# Non-Bstation page
'url': 'https://www.bilibili.tv/en/play/1033760/11005006',
'info_dict': {
'id': '11005006',
'ext': 'mp4',
'title': 'E3 - Who?',
'thumbnail': r're:^https://pic\.bstarstatic\.com/ogv/.+\.png$',
'episode_number': 3,
'description': 'md5:e1a775e71a35c43f141484715470ad09',
'episode': 'Episode 3',
'upload_date': '20211219',
'timestamp': 1639928700,
'chapters': [{
'start_time': 0,
'end_time': 88.0,
'title': '<Untitled Chapter 1>'
}, {
'start_time': 88.0,
'end_time': 156.0,
'title': 'Intro'
}, {
'start_time': 1173.0,
'end_time': 1259.535,
'title': 'Outro'
}],
}
}, {
# Subtitle with empty content
'url': 'https://www.bilibili.tv/en/play/1005144/10131790',
'info_dict': {
'id': '10131790',
'ext': 'mp4',
'title': 'E140 - Two Heartbeats: Kabuto\'s Trap',
'thumbnail': r're:^https://pic\.bstarstatic\.com/ogv/.+\.png$',
'episode_number': 140,
},
'skip': 'According to the copyright owner\'s request, you may only watch the video after you log in.'
}, {
'url': 'https://www.bilibili.tv/en/video/2041863208',
'info_dict': {
'id': '2041863208',
'ext': 'mp4',
'timestamp': 1670874843,
'description': 'Scheduled for April 2023.\nStudio: ufotable',
'thumbnail': r're:https?://pic[-\.]bstarstatic.+/ugc/.+\.jpg$',
'upload_date': '20221212',
'title': 'Kimetsu no Yaiba Season 3 Official Trailer - Bstation',
}
}, {
# episode id without intro and outro
'url': 'https://www.bilibili.tv/en/play/1048837/11246489',
'info_dict': {
'id': '11246489',
'ext': 'mp4',
'title': 'E1 - Operation \'Strix\' <Owl>',
'description': 'md5:b4434eb1a9a97ad2bccb779514b89f17',
'timestamp': 1649516400,
'thumbnail': 'https://pic.bstarstatic.com/ogv/62cb1de23ada17fb70fbe7bdd6ff29c29da02a64.png',
'episode': 'Episode 1',
'episode_number': 1,
'upload_date': '20220409',
},
}, {
'url': 'https://www.biliintl.com/en/play/34613/341736',
'only_matching': True,
}, {
# User-generated content (as opposed to a series licensed from a studio)
'url': 'https://bilibili.tv/en/video/2019955076',
'only_matching': True,
}, {
# No language in URL
'url': 'https://www.bilibili.tv/video/2019955076',
'only_matching': True,
}, {
# Uppercase language in URL
'url': 'https://www.bilibili.tv/EN/video/2019955076',
'only_matching': True,
}]
def _make_url(video_id, series_id=None):
if series_id:
return f'https://www.bilibili.tv/en/play/{series_id}/{video_id}'
return f'https://www.bilibili.tv/en/video/{video_id}'
def _extract_video_metadata(self, url, video_id, season_id):
url, smuggled_data = unsmuggle_url(url, {})
if smuggled_data.get('title'):
return smuggled_data
webpage = self._download_webpage(url, video_id)
# Bstation layout
initial_data = (
self._search_json(r'window\.__INITIAL_(?:DATA|STATE)__\s*=', webpage, 'preload state', video_id, default={})
or self._search_nuxt_data(webpage, video_id, '__initialState', fatal=False, traverse=None))
video_data = traverse_obj(
initial_data, ('OgvVideo', 'epDetail'), ('UgcVideo', 'videoData'), ('ugc', 'archive'), expected_type=dict) or {}
if season_id and not video_data:
# Non-Bstation layout, read through episode list
season_json = self._call_api(f'/web/v2/ogv/play/episodes?season_id={season_id}&platform=web', video_id)
video_data = traverse_obj(season_json, (
'sections', ..., 'episodes', lambda _, v: str(v['episode_id']) == video_id
), expected_type=dict, get_all=False)
# XXX: webpage metadata may not accurate, it just used to not crash when video_data not found
return merge_dicts(
self._parse_video_metadata(video_data), self._search_json_ld(webpage, video_id), {
'title': self._html_search_meta('og:title', webpage),
'description': self._html_search_meta('og:description', webpage)
})
def _real_extract(self, url):
season_id, ep_id, aid = self._match_valid_url(url).group('season_id', 'ep_id', 'aid')
video_id = ep_id or aid
chapters = None
if ep_id:
intro_ending_json = self._call_api(
f'/web/v2/ogv/play/episode?episode_id={ep_id}&platform=web',
video_id, fatal=False) or {}
if intro_ending_json.get('skip'):
# FIXME: start time and end time seems a bit off a few second even it corrext based on ogv.*.js
# ref: https://p.bstarstatic.com/fe-static/bstar-web-new/assets/ogv.2b147442.js
chapters = [{
'start_time': float_or_none(traverse_obj(intro_ending_json, ('skip', 'opening_start_time')), 1000),
'end_time': float_or_none(traverse_obj(intro_ending_json, ('skip', 'opening_end_time')), 1000),
'title': 'Intro'
}, {
'start_time': float_or_none(traverse_obj(intro_ending_json, ('skip', 'ending_start_time')), 1000),
'end_time': float_or_none(traverse_obj(intro_ending_json, ('skip', 'ending_end_time')), 1000),
'title': 'Outro'
}]
return {
'id': video_id,
**self._extract_video_metadata(url, video_id, season_id),
'formats': self._get_formats(ep_id=ep_id, aid=aid),
'subtitles': self.extract_subtitles(ep_id=ep_id, aid=aid),
'chapters': chapters
}
class BiliIntlSeriesIE(BiliIntlBaseIE):
IE_NAME = 'biliIntl:series'
_VALID_URL = r'https?://(?:www\.)?bili(?:bili\.tv|intl\.com)/(?:[a-zA-Z]{2}/)?(?:play|media)/(?P<id>\d+)/?(?:[?#]|$)'
_TESTS = [{
'url': 'https://www.bilibili.tv/en/play/34613',
'playlist_mincount': 15,
'info_dict': {
'id': '34613',
'title': 'TONIKAWA: Over the Moon For You',
'description': 'md5:297b5a17155eb645e14a14b385ab547e',
'categories': ['Slice of life', 'Comedy', 'Romance'],
'thumbnail': r're:^https://pic\.bstarstatic\.com/ogv/.+\.png$',
'view_count': int,
},
'params': {
'skip_download': True,
},
}, {
'url': 'https://www.bilibili.tv/en/media/1048837',
'info_dict': {
'id': '1048837',
'title': 'SPY×FAMILY',
'description': 'md5:b4434eb1a9a97ad2bccb779514b89f17',
'categories': ['Adventure', 'Action', 'Comedy'],
'thumbnail': r're:^https://pic\.bstarstatic\.com/ogv/.+\.jpg$',
'view_count': int,
},
'playlist_mincount': 25,
}, {
'url': 'https://www.biliintl.com/en/play/34613',
'only_matching': True,
}, {
'url': 'https://www.biliintl.com/EN/play/34613',
'only_matching': True,
}]
def _entries(self, series_id):
series_json = self._call_api(f'/web/v2/ogv/play/episodes?season_id={series_id}&platform=web', series_id)
for episode in traverse_obj(series_json, ('sections', ..., 'episodes', ...), expected_type=dict):
episode_id = str(episode['episode_id'])
yield self.url_result(smuggle_url(
BiliIntlIE._make_url(episode_id, series_id),
self._parse_video_metadata(episode)
), BiliIntlIE, episode_id)
def _real_extract(self, url):
series_id = self._match_id(url)
series_info = self._call_api(f'/web/v2/ogv/play/season_info?season_id={series_id}&platform=web', series_id).get('season') or {}
return self.playlist_result(
self._entries(series_id), series_id, series_info.get('title'), series_info.get('description'),
categories=traverse_obj(series_info, ('styles', ..., 'title'), expected_type=str_or_none),
thumbnail=url_or_none(series_info.get('horizontal_cover')), view_count=parse_count(series_info.get('view')))
class BiliLiveIE(InfoExtractor):
_VALID_URL = r'https?://live.bilibili.com/(?:blanc/)?(?P<id>\d+)'
_TESTS = [{
'url': 'https://live.bilibili.com/196',
'info_dict': {
'id': '33989',
'description': "周六杂谈回,其他时候随机游戏。 | \n录播:@下播型泛式录播组。 | \n直播通知群全员禁言66690667090209258459971⑧481 (功能一样,别多加)",
'ext': 'flv',
'title': "太空狼人杀联动,不被爆杀就算赢",
'thumbnail': "https://i0.hdslb.com/bfs/live/new_room_cover/e607bc1529057ef4b332e1026e62cf46984c314d.jpg",
'timestamp': 1650802769,
},
'skip': 'not live'
}, {
'url': 'https://live.bilibili.com/196?broadcast_type=0&is_room_feed=1?spm_id_from=333.999.space_home.strengthen_live_card.click',
'only_matching': True
}, {
'url': 'https://live.bilibili.com/blanc/196',
'only_matching': True
}]
_FORMATS = {
80: {'format_id': 'low', 'format_note': '流畅'},
150: {'format_id': 'high_res', 'format_note': '高清'},
250: {'format_id': 'ultra_high_res', 'format_note': '超清'},
400: {'format_id': 'blue_ray', 'format_note': '蓝光'},
10000: {'format_id': 'source', 'format_note': '原画'},
20000: {'format_id': '4K', 'format_note': '4K'},
30000: {'format_id': 'dolby', 'format_note': '杜比'},
}
_quality = staticmethod(qualities(list(_FORMATS)))
def _call_api(self, path, room_id, query):
api_result = self._download_json(f'https://api.live.bilibili.com/{path}', room_id, query=query)
if api_result.get('code') != 0:
raise ExtractorError(api_result.get('message') or 'Unable to download JSON metadata')
return api_result.get('data') or {}
def _parse_formats(self, qn, fmt):
for codec in fmt.get('codec') or []:
if codec.get('current_qn') != qn:
continue
for url_info in codec['url_info']:
yield {
'url': f'{url_info["host"]}{codec["base_url"]}{url_info["extra"]}',
'ext': fmt.get('format_name'),
'vcodec': codec.get('codec_name'),
'quality': self._quality(qn),
**self._FORMATS[qn],
}
def _real_extract(self, url):
room_id = self._match_id(url)
room_data = self._call_api('room/v1/Room/get_info', room_id, {'id': room_id})
if room_data.get('live_status') == 0:
raise ExtractorError('Streamer is not live', expected=True)
formats = []
for qn in self._FORMATS.keys():
stream_data = self._call_api('xlive/web-room/v2/index/getRoomPlayInfo', room_id, {
'room_id': room_id,
'qn': qn,
'codec': '0,1',
'format': '0,2',
'mask': '0',
'no_playurl': '0',
'platform': 'web',
'protocol': '0,1',
})
for fmt in traverse_obj(stream_data, ('playurl_info', 'playurl', 'stream', ..., 'format', ...)) or []:
formats.extend(self._parse_formats(qn, fmt))
return {
'id': room_id,
'title': room_data.get('title'),
'description': room_data.get('description'),
'thumbnail': room_data.get('user_cover'),
'timestamp': stream_data.get('live_time'),
'formats': formats,
'is_live': True,
'http_headers': {
'Referer': url,
},
}