1
0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-12-05 20:48:04 +01:00
youtube-dl/youtube_dl/extractor/qqmusic.py

318 lines
11 KiB
Python
Raw Normal View History

2015-03-17 12:03:29 +01:00
# coding: utf-8
from __future__ import unicode_literals
2015-03-18 06:56:02 +01:00
import random
import time
2015-03-18 07:59:33 +01:00
import re
2015-03-18 06:56:02 +01:00
2015-03-17 12:03:29 +01:00
from .common import InfoExtractor
2015-03-18 18:47:07 +01:00
from ..utils import (
strip_jsonp,
unescapeHTML,
2015-06-18 07:59:37 +02:00
clean_html,
2015-03-18 18:47:07 +01:00
)
2015-03-18 07:59:33 +01:00
from ..compat import compat_urllib_request
2015-03-17 12:03:29 +01:00
class QQMusicIE(InfoExtractor):
IE_NAME = 'qqmusic'
2015-07-10 18:52:07 +02:00
IE_DESC = 'QQ音乐'
2015-03-17 12:03:29 +01:00
_VALID_URL = r'http://y.qq.com/#type=song&mid=(?P<id>[0-9A-Za-z]+)'
_TESTS = [{
'url': 'http://y.qq.com/#type=song&mid=004295Et37taLD',
'md5': '9ce1c1c8445f561506d2e3cfb0255705',
2015-03-17 12:03:29 +01:00
'info_dict': {
'id': '004295Et37taLD',
'ext': 'mp3',
2015-03-17 12:03:29 +01:00
'title': '可惜没如果',
'upload_date': '20141227',
'creator': '林俊杰',
'description': 'md5:d327722d0361576fde558f1ac68a7065',
'thumbnail': 're:^https?://.*\.jpg$',
}
}, {
'note': 'There is no mp3-320 version of this song.',
'url': 'http://y.qq.com/#type=song&mid=004MsGEo3DdNxV',
'md5': 'fa3926f0c585cda0af8fa4f796482e3e',
'info_dict': {
'id': '004MsGEo3DdNxV',
'ext': 'mp3',
'title': '如果',
'upload_date': '20050626',
'creator': '李季美',
'description': 'md5:46857d5ed62bc4ba84607a805dccf437',
'thumbnail': 're:^https?://.*\.jpg$',
2015-03-17 12:03:29 +01:00
}
}]
_FORMATS = {
2015-06-05 00:04:26 +02:00
'mp3-320': {'prefix': 'M800', 'ext': 'mp3', 'preference': 40, 'abr': 320},
'mp3-128': {'prefix': 'M500', 'ext': 'mp3', 'preference': 30, 'abr': 128},
'm4a': {'prefix': 'C200', 'ext': 'm4a', 'preference': 10}
}
2015-03-18 06:56:02 +01:00
# Reference: m_r_GetRUin() in top_player.js
# http://imgcache.gtimg.cn/music/portal_v3/y/top_player.js
@staticmethod
def m_r_get_ruin():
curMs = int(time.time() * 1000) % 1000
return int(round(random.random() * 2147483647) * curMs % 1E10)
2015-03-17 12:03:29 +01:00
def _real_extract(self, url):
mid = self._match_id(url)
detail_info_page = self._download_webpage(
'http://s.plcloud.music.qq.com/fcgi-bin/fcg_yqq_song_detail_info.fcg?songmid=%s&play=0' % mid,
2015-03-18 07:59:33 +01:00
mid, note='Download song detail info',
errnote='Unable to get song detail info', encoding='gbk')
2015-03-17 12:03:29 +01:00
song_name = self._html_search_regex(
r"songname:\s*'([^']+)'", detail_info_page, 'song name')
publish_time = self._html_search_regex(
r'发行时间:(\d{4}-\d{2}-\d{2})', detail_info_page,
'publish time', default=None)
if publish_time:
publish_time = publish_time.replace('-', '')
2015-03-17 12:03:29 +01:00
singer = self._html_search_regex(
r"singer:\s*'([^']+)", detail_info_page, 'singer', default=None)
lrc_content = self._html_search_regex(
r'<div class="content" id="lrc_content"[^<>]*>([^<>]+)</div>',
detail_info_page, 'LRC lyrics', default=None)
if lrc_content:
lrc_content = lrc_content.replace('\\n', '\n')
2015-03-17 12:03:29 +01:00
thumbnail_url = None
albummid = self._search_regex(
2015-06-18 15:13:03 +02:00
[r'albummid:\'([0-9a-zA-Z]+)\'', r'"albummid":"([0-9a-zA-Z]+)"'],
detail_info_page, 'album mid', default=None)
if albummid:
thumbnail_url = "http://i.gtimg.cn/music/photo/mid_album_500/%s/%s/%s.jpg" \
% (albummid[-2:-1], albummid[-1], albummid)
2015-03-18 06:56:02 +01:00
guid = self.m_r_get_ruin()
2015-03-17 12:03:29 +01:00
vkey = self._download_json(
'http://base.music.qq.com/fcgi-bin/fcg_musicexpress.fcg?json=3&guid=%s' % guid,
mid, note='Retrieve vkey', errnote='Unable to get vkey',
transform_source=strip_jsonp)['key']
formats = []
for format_id, details in self._FORMATS.items():
formats.append({
'url': 'http://cc.stream.qqmusic.qq.com/%s%s.%s?vkey=%s&guid=%s&fromtag=0'
% (details['prefix'], mid, details['ext'], vkey, guid),
'format': format_id,
'format_id': format_id,
'preference': details['preference'],
'abr': details.get('abr'),
})
2015-06-18 17:09:04 +02:00
self._check_formats(formats, mid)
self._sort_formats(formats)
2015-03-17 12:03:29 +01:00
return {
'id': mid,
'formats': formats,
2015-03-17 12:03:29 +01:00
'title': song_name,
'upload_date': publish_time,
'creator': singer,
'description': lrc_content,
'thumbnail': thumbnail_url,
2015-03-17 12:03:29 +01:00
}
2015-03-18 07:59:33 +01:00
2015-03-18 18:47:07 +01:00
class QQPlaylistBaseIE(InfoExtractor):
@staticmethod
def qq_static_url(category, mid):
return 'http://y.qq.com/y/static/%s/%s/%s/%s.html' % (category, mid[-2], mid[-1], mid)
@classmethod
def get_entries_from_page(cls, page):
entries = []
for item in re.findall(r'class="data"[^<>]*>([^<>]+)</', page):
song_mid = unescapeHTML(item).split('|')[-5]
entries.append(cls.url_result(
'http://y.qq.com/#type=song&mid=' + song_mid, 'QQMusic',
song_mid))
2015-03-18 18:47:07 +01:00
return entries
class QQMusicSingerIE(QQPlaylistBaseIE):
IE_NAME = 'qqmusic:singer'
2015-07-10 19:21:36 +02:00
IE_DESC = 'QQ音乐 - 歌手'
2015-03-18 07:59:33 +01:00
_VALID_URL = r'http://y.qq.com/#type=singer&mid=(?P<id>[0-9A-Za-z]+)'
_TEST = {
'url': 'http://y.qq.com/#type=singer&mid=001BLpXF2DyJe2',
'info_dict': {
'id': '001BLpXF2DyJe2',
'title': '林俊杰',
'description': 'md5:2a222d89ba4455a3af19940c0481bb78',
},
'playlist_count': 12,
}
def _real_extract(self, url):
mid = self._match_id(url)
singer_page = self._download_webpage(
2015-03-18 18:47:07 +01:00
self.qq_static_url('singer', mid), mid, 'Download singer page')
2015-03-18 07:59:33 +01:00
2015-03-18 18:47:07 +01:00
entries = self.get_entries_from_page(singer_page)
2015-03-18 07:59:33 +01:00
singer_name = self._html_search_regex(
r"singername\s*:\s*'([^']+)'", singer_page, 'singer name',
default=None)
singer_id = self._html_search_regex(
r"singerid\s*:\s*'([0-9]+)'", singer_page, 'singer id',
default=None)
singer_desc = None
if singer_id:
req = compat_urllib_request.Request(
'http://s.plcloud.music.qq.com/fcgi-bin/fcg_get_singer_desc.fcg?utf8=1&outCharset=utf-8&format=xml&singerid=%s' % singer_id)
req.add_header(
'Referer', 'http://s.plcloud.music.qq.com/xhr_proxy_utf8.html')
singer_desc_page = self._download_xml(
2015-03-18 18:47:07 +01:00
req, mid, 'Donwload singer description XML')
2015-03-18 07:59:33 +01:00
singer_desc = singer_desc_page.find('./data/info/desc').text
return self.playlist_result(entries, mid, singer_name, singer_desc)
2015-03-18 18:47:07 +01:00
class QQMusicAlbumIE(QQPlaylistBaseIE):
IE_NAME = 'qqmusic:album'
2015-07-10 19:21:36 +02:00
IE_DESC = 'QQ音乐 - 专辑'
2015-03-18 18:47:07 +01:00
_VALID_URL = r'http://y.qq.com/#type=album&mid=(?P<id>[0-9A-Za-z]+)'
2015-06-18 09:52:04 +02:00
_TESTS = [{
'url': 'http://y.qq.com/#type=album&mid=000gXCTb2AhRR1',
2015-03-18 18:47:07 +01:00
'info_dict': {
'id': '000gXCTb2AhRR1',
'title': '我们都是这样长大的',
2015-07-06 11:08:32 +02:00
'description': 'md5:179c5dce203a5931970d306aa9607ea6',
2015-03-18 18:47:07 +01:00
},
'playlist_count': 4,
2015-06-18 09:52:04 +02:00
}, {
'url': 'http://y.qq.com/#type=album&mid=002Y5a3b3AlCu3',
'info_dict': {
'id': '002Y5a3b3AlCu3',
'title': '그리고...',
2015-07-06 11:08:32 +02:00
'description': 'md5:a48823755615508a95080e81b51ba729',
2015-06-18 09:52:04 +02:00
},
'playlist_count': 8,
}]
2015-03-18 18:47:07 +01:00
def _real_extract(self, url):
mid = self._match_id(url)
2015-06-18 09:52:04 +02:00
album = self._download_json(
'http://i.y.qq.com/v8/fcg-bin/fcg_v8_album_info_cp.fcg?albummid=%s&format=json' % mid,
mid, 'Download album page')['data']
entries = [
self.url_result(
'http://y.qq.com/#type=song&mid=' + song['songmid'], 'QQMusic', song['songmid']
) for song in album['list']
]
album_name = album.get('name')
2015-06-18 09:52:04 +02:00
album_detail = album.get('desc')
2015-07-06 11:08:32 +02:00
if album_detail is not None:
album_detail = album_detail.strip()
2015-03-18 18:47:07 +01:00
return self.playlist_result(entries, mid, album_name, album_detail)
class QQMusicToplistIE(QQPlaylistBaseIE):
IE_NAME = 'qqmusic:toplist'
2015-07-10 19:21:36 +02:00
IE_DESC = 'QQ音乐 - 排行榜'
_VALID_URL = r'http://y\.qq\.com/#type=toplist&p=(?P<id>(top|global)_[0-9]+)'
2015-05-14 17:25:43 +02:00
_TESTS = [{
2015-06-04 05:27:18 +02:00
'url': 'http://y.qq.com/#type=toplist&p=global_123',
'info_dict': {
2015-06-04 05:27:18 +02:00
'id': 'global_123',
'title': '美国iTunes榜',
},
'playlist_count': 10,
}, {
2015-06-04 05:27:18 +02:00
'url': 'http://y.qq.com/#type=toplist&p=top_3',
'info_dict': {
2015-06-04 05:27:18 +02:00
'id': 'top_3',
'title': 'QQ音乐巅峰榜·欧美',
2015-06-04 05:27:18 +02:00
'description': 'QQ音乐巅峰榜·欧美根据用户收听行为自动生成集结当下最流行的欧美新歌:更新时间每周四22点|统'
'计周期:一周(上周四至本周三)|统计对象:三个月内发行的欧美歌曲|统计数量100首|统计算法:根据'
'歌曲在一周内的有效播放次数由高到低取前100名同一歌手最多允许5首歌曲同时上榜|有效播放次数:'
'登录用户完整播放一首歌曲记为一次有效播放同一用户收听同一首歌曲每天记录为1次有效播放'
},
'playlist_count': 100,
}, {
2015-06-04 05:27:18 +02:00
'url': 'http://y.qq.com/#type=toplist&p=global_106',
'info_dict': {
2015-06-04 05:27:18 +02:00
'id': 'global_106',
'title': '韩国Mnet榜',
},
'playlist_count': 50,
}]
def _real_extract(self, url):
list_id = self._match_id(url)
2015-05-14 17:28:42 +02:00
list_type, num_id = list_id.split("_")
2015-05-14 17:28:42 +02:00
toplist_json = self._download_json(
2015-06-04 05:27:18 +02:00
'http://i.y.qq.com/v8/fcg-bin/fcg_v8_toplist_cp.fcg?type=%s&topid=%s&format=json'
% (list_type, num_id),
list_id, 'Download toplist page')
2015-06-04 05:27:18 +02:00
entries = [
self.url_result(
'http://y.qq.com/#type=song&mid=' + song['data']['songmid'], 'QQMusic', song['data']['songmid']
) for song in toplist_json['songlist']
]
topinfo = toplist_json.get('topinfo', {})
list_name = topinfo.get('ListName')
list_description = topinfo.get('info')
2015-06-04 05:27:18 +02:00
return self.playlist_result(entries, list_id, list_name, list_description)
2015-06-18 07:59:37 +02:00
class QQMusicPlaylistIE(QQPlaylistBaseIE):
IE_NAME = 'qqmusic:playlist'
2015-07-10 19:21:36 +02:00
IE_DESC = 'QQ音乐 - 歌单'
2015-06-18 07:59:37 +02:00
_VALID_URL = r'http://y\.qq\.com/#type=taoge&id=(?P<id>[0-9]+)'
_TEST = {
'url': 'http://y.qq.com/#type=taoge&id=3462654915',
'info_dict': {
'id': '3462654915',
'title': '韩国5月新歌精选下旬',
'description': 'md5:d2c9d758a96b9888cf4fe82f603121d4',
},
'playlist_count': 40,
}
def _real_extract(self, url):
list_id = self._match_id(url)
list_json = self._download_json(
'http://i.y.qq.com/qzone-music/fcg-bin/fcg_ucc_getcdinfo_byids_cp.fcg?type=1&json=1&utf8=1&onlysong=0&disstid=%s'
% list_id, list_id, 'Download list page',
transform_source=strip_jsonp)['cdlist'][0]
entries = [
self.url_result(
'http://y.qq.com/#type=song&mid=' + song['songmid'], 'QQMusic', song['songmid']
) for song in list_json['songlist']
]
list_name = list_json.get('dissname')
2015-06-18 07:59:37 +02:00
list_description = clean_html(unescapeHTML(list_json.get('desc')))
return self.playlist_result(entries, list_id, list_name, list_description)