mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-13 06:47:50 +01:00
7a03f88c40
Closes #9196, Closes #10239 Authored by: hafeoz
647 lines
25 KiB
Python
647 lines
25 KiB
Python
import hashlib
|
|
import itertools
|
|
import json
|
|
import random
|
|
import re
|
|
import time
|
|
|
|
from .common import InfoExtractor
|
|
from ..aes import aes_ecb_encrypt, pkcs7_padding
|
|
from ..utils import (
|
|
ExtractorError,
|
|
int_or_none,
|
|
join_nonempty,
|
|
str_or_none,
|
|
strftime_or_none,
|
|
traverse_obj,
|
|
unified_strdate,
|
|
url_or_none,
|
|
urljoin,
|
|
variadic,
|
|
)
|
|
|
|
|
|
class NetEaseMusicBaseIE(InfoExtractor):
|
|
# XXX: _extract_formats logic depends on the order of the levels in each tier
|
|
_LEVELS = (
|
|
'standard', # free tier; 标准; 128kbps mp3 or aac
|
|
'higher', # free tier; 192kbps mp3 or aac
|
|
'exhigh', # free tier; 极高 (HQ); 320kbps mp3 or aac
|
|
'lossless', # VIP tier; 无损 (SQ); 48kHz/16bit flac
|
|
'hires', # VIP tier; 高解析度无损 (Hi-Res); 192kHz/24bit flac
|
|
'jyeffect', # VIP tier; 高清臻音 (Spatial Audio); 96kHz/24bit flac
|
|
'jymaster', # SVIP tier; 超清母带 (Master); 192kHz/24bit flac
|
|
'sky', # SVIP tier; 沉浸环绕声 (Surround Audio); flac
|
|
)
|
|
_API_BASE = 'http://music.163.com/api/'
|
|
_GEO_BYPASS = False
|
|
|
|
@staticmethod
|
|
def _kilo_or_none(value):
|
|
return int_or_none(value, scale=1000)
|
|
|
|
def _create_eapi_cipher(self, api_path, query_body, cookies):
|
|
request_text = json.dumps({**query_body, 'header': cookies}, separators=(',', ':'))
|
|
|
|
message = f'nobody{api_path}use{request_text}md5forencrypt'.encode('latin1')
|
|
msg_digest = hashlib.md5(message).hexdigest()
|
|
|
|
data = pkcs7_padding(list(str.encode(
|
|
f'{api_path}-36cd479b6b5-{request_text}-36cd479b6b5-{msg_digest}')))
|
|
encrypted = bytes(aes_ecb_encrypt(data, list(b'e82ckenh8dichen8')))
|
|
return f'params={encrypted.hex().upper()}'.encode()
|
|
|
|
def _download_eapi_json(self, path, video_id, query_body, headers={}, **kwargs):
|
|
cookies = {
|
|
'osver': 'undefined',
|
|
'deviceId': 'undefined',
|
|
'appver': '8.0.0',
|
|
'versioncode': '140',
|
|
'mobilename': 'undefined',
|
|
'buildver': '1623435496',
|
|
'resolution': '1920x1080',
|
|
'__csrf': '',
|
|
'os': 'pc',
|
|
'channel': 'undefined',
|
|
'requestId': f'{int(time.time() * 1000)}_{random.randint(0, 1000):04}',
|
|
**traverse_obj(self._get_cookies(self._API_BASE), {
|
|
'MUSIC_U': ('MUSIC_U', {lambda i: i.value}),
|
|
}),
|
|
}
|
|
return self._download_json(
|
|
urljoin('https://interface3.music.163.com/', f'/eapi{path}'), video_id,
|
|
data=self._create_eapi_cipher(f'/api{path}', query_body, cookies), headers={
|
|
'Referer': 'https://music.163.com',
|
|
'Cookie': '; '.join([f'{k}={v}' for k, v in cookies.items()]),
|
|
**headers,
|
|
}, **kwargs)
|
|
|
|
def _call_player_api(self, song_id, level):
|
|
return self._download_eapi_json(
|
|
'/song/enhance/player/url/v1', song_id,
|
|
{'ids': f'[{song_id}]', 'level': level, 'encodeType': 'flac'},
|
|
note=f'Downloading song URL info: level {level}')
|
|
|
|
def _extract_formats(self, info):
|
|
formats = []
|
|
song_id = info['id']
|
|
for level in self._LEVELS:
|
|
song = traverse_obj(
|
|
self._call_player_api(song_id, level), ('data', lambda _, v: url_or_none(v['url']), any))
|
|
if not song:
|
|
break # Media is not available due to removal or geo-restriction
|
|
actual_level = song.get('level')
|
|
if actual_level and actual_level != level:
|
|
if level in ('lossless', 'jymaster'):
|
|
break # We've already extracted the highest level of the user's account tier
|
|
continue
|
|
formats.append({
|
|
'url': song['url'],
|
|
'format_id': level,
|
|
'vcodec': 'none',
|
|
**traverse_obj(song, {
|
|
'ext': ('type', {str}),
|
|
'abr': ('br', {self._kilo_or_none}),
|
|
'filesize': ('size', {int_or_none}),
|
|
}),
|
|
})
|
|
if not actual_level:
|
|
break # Only 1 level is available if API does not return a value (netease:program)
|
|
if not formats:
|
|
self.raise_geo_restricted(
|
|
'No media links found; possibly due to geo restriction', countries=['CN'])
|
|
return formats
|
|
|
|
def _query_api(self, endpoint, video_id, note):
|
|
result = self._download_json(
|
|
f'{self._API_BASE}{endpoint}', video_id, note, headers={'Referer': self._API_BASE})
|
|
code = traverse_obj(result, ('code', {int}))
|
|
message = traverse_obj(result, ('message', {str})) or ''
|
|
if code == -462:
|
|
self.raise_login_required(f'Login required to download: {message}')
|
|
elif code != 200:
|
|
raise ExtractorError(f'Failed to get meta info: {code} {message}')
|
|
return result
|
|
|
|
def _get_entries(self, songs_data, entry_keys=None, id_key='id', name_key='name'):
|
|
for song in traverse_obj(songs_data, (
|
|
*variadic(entry_keys, (str, bytes, dict, set)),
|
|
lambda _, v: int_or_none(v[id_key]) is not None)):
|
|
song_id = str(song[id_key])
|
|
yield self.url_result(
|
|
f'http://music.163.com/#/song?id={song_id}', NetEaseMusicIE,
|
|
song_id, traverse_obj(song, (name_key, {str})))
|
|
|
|
|
|
class NetEaseMusicIE(NetEaseMusicBaseIE):
|
|
IE_NAME = 'netease:song'
|
|
IE_DESC = '网易云音乐'
|
|
_VALID_URL = r'https?://(?:y\.)?music\.163\.com/(?:[#m]/)?song\?.*?\bid=(?P<id>[0-9]+)'
|
|
_TESTS = [{
|
|
'url': 'https://music.163.com/#/song?id=550136151',
|
|
'info_dict': {
|
|
'id': '550136151',
|
|
'ext': 'mp3',
|
|
'title': 'It\'s Ok (Live)',
|
|
'creators': 'count:10',
|
|
'timestamp': 1522944000,
|
|
'upload_date': '20180405',
|
|
'description': 'md5:9fd07059c2ccee3950dc8363429a3135',
|
|
'duration': 197,
|
|
'thumbnail': r're:^http.*\.jpg',
|
|
'album': '偶像练习生 表演曲目合集',
|
|
'average_rating': int,
|
|
'album_artists': ['偶像练习生'],
|
|
},
|
|
}, {
|
|
'url': 'http://music.163.com/song?id=17241424',
|
|
'info_dict': {
|
|
'id': '17241424',
|
|
'ext': 'mp3',
|
|
'title': 'Opus 28',
|
|
'upload_date': '20080211',
|
|
'timestamp': 1202745600,
|
|
'duration': 263,
|
|
'thumbnail': r're:^http.*\.jpg',
|
|
'album': 'Piano Solos Vol. 2',
|
|
'album_artist': 'Dustin O\'Halloran',
|
|
'average_rating': int,
|
|
'description': '[00:05.00]纯音乐,请欣赏\n',
|
|
'album_artists': ['Dustin O\'Halloran'],
|
|
'creators': ['Dustin O\'Halloran'],
|
|
'subtitles': {'lyrics': [{'ext': 'lrc'}]},
|
|
},
|
|
}, {
|
|
'url': 'https://y.music.163.com/m/song?app_version=8.8.45&id=95670&uct2=sKnvS4+0YStsWkqsPhFijw%3D%3D&dlt=0846',
|
|
'md5': 'b896be78d8d34bd7bb665b26710913ff',
|
|
'info_dict': {
|
|
'id': '95670',
|
|
'ext': 'mp3',
|
|
'title': '国际歌',
|
|
'upload_date': '19911130',
|
|
'timestamp': 691516800,
|
|
'description': 'md5:1ba2f911a2b0aa398479f595224f2141',
|
|
'subtitles': {'lyrics': [{'ext': 'lrc'}]},
|
|
'duration': 268,
|
|
'alt_title': '伴唱:现代人乐队 合唱:总政歌舞团',
|
|
'thumbnail': r're:^http.*\.jpg',
|
|
'average_rating': int,
|
|
'album': '红色摇滚',
|
|
'album_artist': '侯牧人',
|
|
'creators': ['马备'],
|
|
'album_artists': ['侯牧人'],
|
|
},
|
|
}, {
|
|
'url': 'http://music.163.com/#/song?id=32102397',
|
|
'md5': '3e909614ce09b1ccef4a3eb205441190',
|
|
'info_dict': {
|
|
'id': '32102397',
|
|
'ext': 'mp3',
|
|
'title': 'Bad Blood',
|
|
'creators': ['Taylor Swift', 'Kendrick Lamar'],
|
|
'upload_date': '20150516',
|
|
'timestamp': 1431792000,
|
|
'description': 'md5:21535156efb73d6d1c355f95616e285a',
|
|
'subtitles': {'lyrics': [{'ext': 'lrc'}]},
|
|
'duration': 199,
|
|
'thumbnail': r're:^http.*\.jpg',
|
|
'album': 'Bad Blood',
|
|
'average_rating': int,
|
|
'album_artist': 'Taylor Swift',
|
|
},
|
|
'skip': 'Blocked outside Mainland China',
|
|
}, {
|
|
'note': 'Has translated name.',
|
|
'url': 'http://music.163.com/#/song?id=22735043',
|
|
'info_dict': {
|
|
'id': '22735043',
|
|
'ext': 'mp3',
|
|
'title': '소원을 말해봐 (Genie)',
|
|
'creators': ['少女时代'],
|
|
'upload_date': '20100127',
|
|
'timestamp': 1264608000,
|
|
'description': 'md5:03d1ffebec3139aa4bafe302369269c5',
|
|
'subtitles': {'lyrics': [{'ext': 'lrc'}]},
|
|
'duration': 229,
|
|
'alt_title': '说出愿望吧(Genie)',
|
|
'thumbnail': r're:^http.*\.jpg',
|
|
'average_rating': int,
|
|
'album': 'Oh!',
|
|
'album_artist': '少女时代',
|
|
},
|
|
'skip': 'Blocked outside Mainland China',
|
|
}]
|
|
|
|
def _process_lyrics(self, lyrics_info):
|
|
original = traverse_obj(lyrics_info, ('lrc', 'lyric', {str}))
|
|
translated = traverse_obj(lyrics_info, ('tlyric', 'lyric', {str}))
|
|
|
|
if not original or original == '[99:00.00]纯音乐,请欣赏\n':
|
|
return None
|
|
|
|
if not translated:
|
|
return {
|
|
'lyrics': [{'data': original, 'ext': 'lrc'}],
|
|
}
|
|
|
|
lyrics_expr = r'(\[[0-9]{2}:[0-9]{2}\.[0-9]{2,}\])([^\n]+)'
|
|
original_ts_texts = re.findall(lyrics_expr, original)
|
|
translation_ts_dict = dict(re.findall(lyrics_expr, translated))
|
|
|
|
merged = '\n'.join(
|
|
join_nonempty(f'{timestamp}{text}', translation_ts_dict.get(timestamp, ''), delim=' / ')
|
|
for timestamp, text in original_ts_texts)
|
|
|
|
return {
|
|
'lyrics_merged': [{'data': merged, 'ext': 'lrc'}],
|
|
'lyrics': [{'data': original, 'ext': 'lrc'}],
|
|
'lyrics_translated': [{'data': translated, 'ext': 'lrc'}],
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
song_id = self._match_id(url)
|
|
|
|
info = self._query_api(
|
|
f'song/detail?id={song_id}&ids=%5B{song_id}%5D', song_id, 'Downloading song info')['songs'][0]
|
|
|
|
formats = self._extract_formats(info)
|
|
|
|
lyrics = self._process_lyrics(self._query_api(
|
|
f'song/lyric?id={song_id}&lv=-1&tv=-1', song_id, 'Downloading lyrics data'))
|
|
lyric_data = {
|
|
'description': traverse_obj(lyrics, (('lyrics_merged', 'lyrics'), 0, 'data'), get_all=False),
|
|
'subtitles': lyrics,
|
|
} if lyrics else {}
|
|
|
|
return {
|
|
'id': song_id,
|
|
'formats': formats,
|
|
'alt_title': '/'.join(traverse_obj(info, (('transNames', 'alias'), ...))) or None,
|
|
'creators': traverse_obj(info, ('artists', ..., 'name')) or None,
|
|
'album_artists': traverse_obj(info, ('album', 'artists', ..., 'name')) or None,
|
|
**lyric_data,
|
|
**traverse_obj(info, {
|
|
'title': ('name', {str}),
|
|
'timestamp': ('album', 'publishTime', {self._kilo_or_none}),
|
|
'thumbnail': ('album', 'picUrl', {url_or_none}),
|
|
'duration': ('duration', {self._kilo_or_none}),
|
|
'album': ('album', 'name', {str}),
|
|
'average_rating': ('score', {int_or_none}),
|
|
}),
|
|
}
|
|
|
|
|
|
class NetEaseMusicAlbumIE(NetEaseMusicBaseIE):
|
|
IE_NAME = 'netease:album'
|
|
IE_DESC = '网易云音乐 - 专辑'
|
|
_VALID_URL = r'https?://music\.163\.com/(?:#/)?album\?id=(?P<id>[0-9]+)'
|
|
_TESTS = [{
|
|
'url': 'https://music.163.com/#/album?id=133153666',
|
|
'info_dict': {
|
|
'id': '133153666',
|
|
'title': '桃几的翻唱',
|
|
'upload_date': '20210913',
|
|
'description': '桃几2021年翻唱合集',
|
|
'thumbnail': r're:^http.*\.jpg',
|
|
},
|
|
'playlist_mincount': 12,
|
|
}, {
|
|
'url': 'http://music.163.com/#/album?id=220780',
|
|
'info_dict': {
|
|
'id': '220780',
|
|
'title': 'B\'Day',
|
|
'upload_date': '20060904',
|
|
'description': 'md5:71a74e1d8f392d88cf1bbe48879ad0b0',
|
|
'thumbnail': r're:^http.*\.jpg',
|
|
},
|
|
'playlist_count': 23,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
album_id = self._match_id(url)
|
|
webpage = self._download_webpage(f'https://music.163.com/album?id={album_id}', album_id)
|
|
|
|
songs = self._search_json(
|
|
r'<textarea[^>]+\bid="song-list-pre-data"[^>]*>', webpage, 'metainfo', album_id,
|
|
end_pattern=r'</textarea>', contains_pattern=r'\[(?s:.+)\]')
|
|
metainfo = {
|
|
'title': self._og_search_property('title', webpage, 'title', fatal=False),
|
|
'description': self._html_search_regex(
|
|
(rf'<div[^>]+\bid="album-desc-{suffix}"[^>]*>(.*?)</div>' for suffix in ('more', 'dot')),
|
|
webpage, 'description', flags=re.S, fatal=False),
|
|
'thumbnail': self._og_search_property('image', webpage, 'thumbnail', fatal=False),
|
|
'upload_date': unified_strdate(self._html_search_meta('music:release_date', webpage, 'date', fatal=False)),
|
|
}
|
|
return self.playlist_result(self._get_entries(songs), album_id, **metainfo)
|
|
|
|
|
|
class NetEaseMusicSingerIE(NetEaseMusicBaseIE):
|
|
IE_NAME = 'netease:singer'
|
|
IE_DESC = '网易云音乐 - 歌手'
|
|
_VALID_URL = r'https?://music\.163\.com/(?:#/)?artist\?id=(?P<id>[0-9]+)'
|
|
_TESTS = [{
|
|
'note': 'Singer has aliases.',
|
|
'url': 'http://music.163.com/#/artist?id=10559',
|
|
'info_dict': {
|
|
'id': '10559',
|
|
'title': '张惠妹 - aMEI;阿妹;阿密特',
|
|
},
|
|
'playlist_count': 50,
|
|
}, {
|
|
'note': 'Singer has translated name.',
|
|
'url': 'http://music.163.com/#/artist?id=124098',
|
|
'info_dict': {
|
|
'id': '124098',
|
|
'title': '李昇基 - 이승기',
|
|
},
|
|
'playlist_count': 50,
|
|
}, {
|
|
'note': 'Singer with both translated and alias',
|
|
'url': 'https://music.163.com/#/artist?id=159692',
|
|
'info_dict': {
|
|
'id': '159692',
|
|
'title': '初音ミク - 初音未来;Hatsune Miku',
|
|
},
|
|
'playlist_count': 50,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
singer_id = self._match_id(url)
|
|
|
|
info = self._query_api(
|
|
f'artist/{singer_id}?id={singer_id}', singer_id, note='Downloading singer data')
|
|
|
|
name = join_nonempty(
|
|
traverse_obj(info, ('artist', 'name', {str})),
|
|
join_nonempty(*traverse_obj(info, ('artist', ('trans', ('alias', ...)), {str})), delim=';'),
|
|
delim=' - ')
|
|
|
|
return self.playlist_result(self._get_entries(info, 'hotSongs'), singer_id, name)
|
|
|
|
|
|
class NetEaseMusicListIE(NetEaseMusicBaseIE):
|
|
IE_NAME = 'netease:playlist'
|
|
IE_DESC = '网易云音乐 - 歌单'
|
|
_VALID_URL = r'https?://music\.163\.com/(?:#/)?(?:playlist|discover/toplist)\?id=(?P<id>[0-9]+)'
|
|
_TESTS = [{
|
|
'url': 'http://music.163.com/#/playlist?id=79177352',
|
|
'info_dict': {
|
|
'id': '79177352',
|
|
'title': 'Billboard 2007 Top 100',
|
|
'description': 'md5:12fd0819cab2965b9583ace0f8b7b022',
|
|
'tags': ['欧美'],
|
|
'uploader': '浑然破灭',
|
|
'uploader_id': '67549805',
|
|
'timestamp': int,
|
|
'upload_date': r're:\d{8}',
|
|
},
|
|
'playlist_mincount': 95,
|
|
}, {
|
|
'note': 'Toplist/Charts sample',
|
|
'url': 'https://music.163.com/#/discover/toplist?id=60198',
|
|
'info_dict': {
|
|
'id': '60198',
|
|
'title': 're:美国Billboard榜 [0-9]{4}-[0-9]{2}-[0-9]{2}',
|
|
'description': '美国Billboard排行榜',
|
|
'tags': ['流行', '欧美', '榜单'],
|
|
'uploader': 'Billboard公告牌',
|
|
'uploader_id': '48171',
|
|
'timestamp': int,
|
|
'upload_date': r're:\d{8}',
|
|
},
|
|
'playlist_count': 100,
|
|
}, {
|
|
'note': 'Toplist/Charts sample',
|
|
'url': 'http://music.163.com/#/discover/toplist?id=3733003',
|
|
'info_dict': {
|
|
'id': '3733003',
|
|
'title': 're:韩国Melon排行榜周榜(?: [0-9]{4}-[0-9]{2}-[0-9]{2})?',
|
|
'description': 'md5:73ec782a612711cadc7872d9c1e134fc',
|
|
'upload_date': '20200109',
|
|
'uploader_id': '2937386',
|
|
'tags': ['韩语', '榜单'],
|
|
'uploader': 'Melon榜单',
|
|
'timestamp': 1578569373,
|
|
},
|
|
'playlist_count': 50,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
list_id = self._match_id(url)
|
|
|
|
info = self._download_eapi_json(
|
|
'/v3/playlist/detail', list_id,
|
|
{'id': list_id, 't': '-1', 'n': '500', 's': '0'},
|
|
note='Downloading playlist info')
|
|
|
|
metainfo = traverse_obj(info, ('playlist', {
|
|
'title': ('name', {str}),
|
|
'description': ('description', {str}),
|
|
'tags': ('tags', ..., {str}),
|
|
'uploader': ('creator', 'nickname', {str}),
|
|
'uploader_id': ('creator', 'userId', {str_or_none}),
|
|
'timestamp': ('updateTime', {self._kilo_or_none}),
|
|
}))
|
|
if traverse_obj(info, ('playlist', 'specialType')) == 10:
|
|
metainfo['title'] = f'{metainfo.get("title")} {strftime_or_none(metainfo.get("timestamp"), "%Y-%m-%d")}'
|
|
|
|
return self.playlist_result(self._get_entries(info, ('playlist', 'tracks')), list_id, **metainfo)
|
|
|
|
|
|
class NetEaseMusicMvIE(NetEaseMusicBaseIE):
|
|
IE_NAME = 'netease:mv'
|
|
IE_DESC = '网易云音乐 - MV'
|
|
_VALID_URL = r'https?://music\.163\.com/(?:#/)?mv\?id=(?P<id>[0-9]+)'
|
|
_TESTS = [{
|
|
'url': 'https://music.163.com/#/mv?id=10958064',
|
|
'info_dict': {
|
|
'id': '10958064',
|
|
'ext': 'mp4',
|
|
'title': '交换余生',
|
|
'description': 'md5:e845872cff28820642a2b02eda428fea',
|
|
'creators': ['林俊杰'],
|
|
'upload_date': '20200916',
|
|
'thumbnail': r're:http.*\.jpg',
|
|
'duration': 364,
|
|
'view_count': int,
|
|
'like_count': int,
|
|
'comment_count': int,
|
|
},
|
|
}, {
|
|
'url': 'http://music.163.com/#/mv?id=415350',
|
|
'info_dict': {
|
|
'id': '415350',
|
|
'ext': 'mp4',
|
|
'title': '이럴거면 그러지말지',
|
|
'description': '白雅言自作曲唱甜蜜爱情',
|
|
'creators': ['白娥娟'],
|
|
'upload_date': '20150520',
|
|
'thumbnail': r're:http.*\.jpg',
|
|
'duration': 216,
|
|
'view_count': int,
|
|
'like_count': int,
|
|
'comment_count': int,
|
|
},
|
|
'skip': 'Blocked outside Mainland China',
|
|
}, {
|
|
'note': 'This MV has multiple creators.',
|
|
'url': 'https://music.163.com/#/mv?id=22593543',
|
|
'info_dict': {
|
|
'id': '22593543',
|
|
'ext': 'mp4',
|
|
'title': '老北京杀器',
|
|
'creators': ['秃子2z', '辉子', 'Saber梁维嘉'],
|
|
'duration': 206,
|
|
'upload_date': '20240618',
|
|
'like_count': int,
|
|
'comment_count': int,
|
|
'thumbnail': r're:http.*\.jpg',
|
|
'view_count': int,
|
|
},
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
mv_id = self._match_id(url)
|
|
|
|
info = self._query_api(
|
|
f'mv/detail?id={mv_id}&type=mp4', mv_id, 'Downloading mv info')['data']
|
|
|
|
formats = [
|
|
{'url': mv_url, 'ext': 'mp4', 'format_id': f'{brs}p', 'height': int_or_none(brs)}
|
|
for brs, mv_url in info['brs'].items()
|
|
]
|
|
|
|
return {
|
|
'id': mv_id,
|
|
'formats': formats,
|
|
'creators': traverse_obj(info, ('artists', ..., 'name')) or [info.get('artistName')],
|
|
**traverse_obj(info, {
|
|
'title': ('name', {str}),
|
|
'description': (('desc', 'briefDesc'), {str}, {lambda x: x or None}),
|
|
'upload_date': ('publishTime', {unified_strdate}),
|
|
'thumbnail': ('cover', {url_or_none}),
|
|
'duration': ('duration', {self._kilo_or_none}),
|
|
'view_count': ('playCount', {int_or_none}),
|
|
'like_count': ('likeCount', {int_or_none}),
|
|
'comment_count': ('commentCount', {int_or_none}),
|
|
}, get_all=False),
|
|
}
|
|
|
|
|
|
class NetEaseMusicProgramIE(NetEaseMusicBaseIE):
|
|
IE_NAME = 'netease:program'
|
|
IE_DESC = '网易云音乐 - 电台节目'
|
|
_VALID_URL = r'https?://music\.163\.com/(?:#/)?program\?id=(?P<id>[0-9]+)'
|
|
_TESTS = [{
|
|
'url': 'http://music.163.com/#/program?id=10109055',
|
|
'info_dict': {
|
|
'id': '32593346',
|
|
'ext': 'mp3',
|
|
'title': '不丹足球背后的故事',
|
|
'description': '喜马拉雅人的足球梦 ...',
|
|
'creators': ['大话西藏'],
|
|
'timestamp': 1434179287,
|
|
'upload_date': '20150613',
|
|
'thumbnail': r're:http.*\.jpg',
|
|
'duration': 900,
|
|
},
|
|
}, {
|
|
'note': 'This program has accompanying songs.',
|
|
'url': 'http://music.163.com/#/program?id=10141022',
|
|
'info_dict': {
|
|
'id': '10141022',
|
|
'title': '滚滚电台的有声节目',
|
|
'description': 'md5:8d594db46cc3e6509107ede70a4aaa3b',
|
|
'creators': ['滚滚电台ORZ'],
|
|
'timestamp': 1434450733,
|
|
'upload_date': '20150616',
|
|
'thumbnail': r're:http.*\.jpg',
|
|
},
|
|
'playlist_count': 4,
|
|
}, {
|
|
'note': 'This program has accompanying songs.',
|
|
'url': 'http://music.163.com/#/program?id=10141022',
|
|
'info_dict': {
|
|
'id': '32647209',
|
|
'ext': 'mp3',
|
|
'title': '滚滚电台的有声节目',
|
|
'description': 'md5:8d594db46cc3e6509107ede70a4aaa3b',
|
|
'creators': ['滚滚电台ORZ'],
|
|
'timestamp': 1434450733,
|
|
'upload_date': '20150616',
|
|
'thumbnail': r're:http.*\.jpg',
|
|
'duration': 1104,
|
|
},
|
|
'params': {
|
|
'noplaylist': True,
|
|
},
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
program_id = self._match_id(url)
|
|
|
|
info = self._query_api(
|
|
f'dj/program/detail?id={program_id}', program_id, note='Downloading program info')['program']
|
|
|
|
metainfo = traverse_obj(info, {
|
|
'title': ('name', {str}),
|
|
'description': ('description', {str}),
|
|
'creator': ('dj', 'brand', {str}),
|
|
'thumbnail': ('coverUrl', {url_or_none}),
|
|
'timestamp': ('createTime', {self._kilo_or_none}),
|
|
})
|
|
|
|
if not self._yes_playlist(
|
|
info['songs'] and program_id, info['mainSong']['id'], playlist_label='program', video_label='song'):
|
|
formats = self._extract_formats(info['mainSong'])
|
|
|
|
return {
|
|
'id': str(info['mainSong']['id']),
|
|
'formats': formats,
|
|
'duration': traverse_obj(info, ('mainSong', 'duration', {self._kilo_or_none})),
|
|
**metainfo,
|
|
}
|
|
|
|
songs = traverse_obj(info, (('mainSong', ('songs', ...)),))
|
|
return self.playlist_result(self._get_entries(songs), program_id, **metainfo)
|
|
|
|
|
|
class NetEaseMusicDjRadioIE(NetEaseMusicBaseIE):
|
|
IE_NAME = 'netease:djradio'
|
|
IE_DESC = '网易云音乐 - 电台'
|
|
_VALID_URL = r'https?://music\.163\.com/(?:#/)?djradio\?id=(?P<id>[0-9]+)'
|
|
_TEST = {
|
|
'url': 'http://music.163.com/#/djradio?id=42',
|
|
'info_dict': {
|
|
'id': '42',
|
|
'title': '声音蔓延',
|
|
'description': 'md5:c7381ebd7989f9f367668a5aee7d5f08',
|
|
},
|
|
'playlist_mincount': 40,
|
|
}
|
|
_PAGE_SIZE = 1000
|
|
|
|
def _real_extract(self, url):
|
|
dj_id = self._match_id(url)
|
|
|
|
metainfo = {}
|
|
entries = []
|
|
for offset in itertools.count(start=0, step=self._PAGE_SIZE):
|
|
info = self._query_api(
|
|
f'dj/program/byradio?asc=false&limit={self._PAGE_SIZE}&radioId={dj_id}&offset={offset}',
|
|
dj_id, note=f'Downloading dj programs - {offset}')
|
|
|
|
entries.extend(self.url_result(
|
|
f'http://music.163.com/#/program?id={program["id"]}', NetEaseMusicProgramIE,
|
|
program['id'], program.get('name')) for program in info['programs'])
|
|
if not metainfo:
|
|
metainfo = traverse_obj(info, ('programs', 0, 'radio', {
|
|
'title': ('name', {str}),
|
|
'description': ('desc', {str}),
|
|
}))
|
|
|
|
if not info['more']:
|
|
break
|
|
|
|
return self.playlist_result(entries, dj_id, **metainfo)
|