aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl/extractor/qqmusic.py
diff options
context:
space:
mode:
authorforDream <gam2046@gmail.com>2017-08-14 08:28:41 +0800
committerforDream <gam2046@gmail.com>2017-08-15 13:14:35 +0800
commit134d85a7bda9189dc3b07dc10a69699084529a6a (patch)
tree8b0a87eb47a2e5d1d5511435810614c7f5035f1e /youtube_dl/extractor/qqmusic.py
parent5c037c0d1f155e951050533690d6e990654cfcc9 (diff)
[qqmusic] review
Diffstat (limited to 'youtube_dl/extractor/qqmusic.py')
-rw-r--r--youtube_dl/extractor/qqmusic.py21
1 files changed, 10 insertions, 11 deletions
diff --git a/youtube_dl/extractor/qqmusic.py b/youtube_dl/extractor/qqmusic.py
index 42be6bc7b..38f4b2cab 100644
--- a/youtube_dl/extractor/qqmusic.py
+++ b/youtube_dl/extractor/qqmusic.py
@@ -166,15 +166,15 @@ class QQPlaylistBaseIE(InfoExtractor):
default_num = 1
json_text = self.get_singer_all_songs(singmid, default_num)
- json_obj = self._parse_json(json_text, singmid)
+ json_obj_all_songs = self._parse_json(json_text, singmid)
- if json_obj['code'] == 0:
- total = json_obj['data']['total']
+ if json_obj_all_songs['code'] == 0:
+ total = json_obj_all_songs['data']['total']
json_text = self.get_singer_all_songs(singmid, total)
- json_obj = self._parse_json(json_text, singmid)
+ json_obj_all_songs = self._parse_json(json_text, singmid)
- for item in json_obj['data']['list']:
- if not (item['musicData'].get('songmid') is None):
+ for item in json_obj_all_songs['data']['list']:
+ if item['musicData'].get('songmid') is not None:
songmid = item['musicData']['songmid']
entries.append(self.url_result(r'https://y.qq.com/n/yqq/song/%s.html' % songmid, 'QQMusic', songmid))
@@ -248,7 +248,7 @@ class QQMusicAlbumIE(QQPlaylistBaseIE):
entries = [
self.url_result(
- 'https://y.qq.com/n/yqq/song/' + song['songmid'] + ".html", 'QQMusic', song['songmid']
+ 'https://y.qq.com/n/yqq/song/' + song['songmid'] + '.html', 'QQMusic', song['songmid']
) for song in album['list']
]
album_name = album.get('name')
@@ -294,8 +294,7 @@ class QQMusicToplistIE(QQPlaylistBaseIE):
def _real_extract(self, url):
list_id = self._match_id(url)
- # list_type, num_id = list_id.split("_")
- list_type = "toplist"
+ list_type = 'toplist'
num_id = list_id
toplist_json = self._download_json(
@@ -305,7 +304,7 @@ class QQMusicToplistIE(QQPlaylistBaseIE):
entries = [
self.url_result(
- 'https://y.qq.com/n/yqq/song/' + song['data']['songmid'] + ".html", 'QQMusic',
+ 'https://y.qq.com/n/yqq/song/' + song['data']['songmid'] + '.html', 'QQMusic',
song['data']['songmid']
) for song in toplist_json['songlist']
]
@@ -357,7 +356,7 @@ class QQMusicPlaylistIE(QQPlaylistBaseIE):
cdlist = list_json['cdlist'][0]
entries = [
self.url_result(
- 'https://y.qq.com/n/yqq/song/' + song['songmid'] + ".html", 'QQMusic', song['songmid']
+ 'https://y.qq.com/n/yqq/song/' + song['songmid'] + '.html', 'QQMusic', song['songmid']
) for song in cdlist['songlist']
]