aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl
diff options
context:
space:
mode:
authorYen Chi Hsuan <yan12125@gmail.com>2015-07-10 23:48:48 +0800
committerYen Chi Hsuan <yan12125@gmail.com>2015-07-10 23:48:48 +0800
commitcf2c5fda4f3e753cc64098e6a751cf1a220efae7 (patch)
tree9ff82a7d6951205b25cd4c1bf7f7f2c57355cd6e /youtube_dl
parenta9684c0dbf3879478fd223ce7594d58be7dffa4f (diff)
[kuwo] Use single quotes
Diffstat (limited to 'youtube_dl')
-rw-r--r--youtube_dl/extractor/kuwo.py22
1 files changed, 11 insertions, 11 deletions
diff --git a/youtube_dl/extractor/kuwo.py b/youtube_dl/extractor/kuwo.py
index 928f7f62d..2b5321cc2 100644
--- a/youtube_dl/extractor/kuwo.py
+++ b/youtube_dl/extractor/kuwo.py
@@ -27,9 +27,9 @@ class KuwoBaseIE(InfoExtractor):
formats = []
for file_format in self._FORMATS:
song_url = self._download_webpage(
- "http://antiserver.kuwo.cn/anti.s?format=%s&br=%s&rid=MUSIC_%s&type=convert_url&response=url" %
+ 'http://antiserver.kuwo.cn/anti.s?format=%s&br=%s&rid=MUSIC_%s&type=convert_url&response=url' %
(file_format['ext'], file_format.get('br', ''), song_id),
- song_id, note="Download %s url info" % file_format["format"],
+ song_id, note='Download %s url info' % file_format['format'],
)
if song_url.startswith('http://') or song_url.startswith('https://'):
formats.append({
@@ -81,7 +81,7 @@ class KuwoIE(KuwoBaseIE):
singer_name = self._html_search_regex(
r'<div[^>]+class="s_img">\s*<a[^>]+title="([^>]+)"',
webpage, 'singer name', default=None)
- lrc_content = clean_html(get_element_by_id("lrcContent", webpage))
+ lrc_content = clean_html(get_element_by_id('lrcContent', webpage))
if lrc_content == '暂无': # indicates no lyrics
lrc_content = None
@@ -94,7 +94,7 @@ class KuwoIE(KuwoBaseIE):
publish_time = None
if album_id is not None:
album_info_page = self._download_webpage(
- "http://www.kuwo.cn/album/%s/" % album_id, song_id,
+ 'http://www.kuwo.cn/album/%s/' % album_id, song_id,
note='Download album detail info',
errnote='Unable to get album detail info')
@@ -138,11 +138,11 @@ class KuwoAlbumIE(InfoExtractor):
r'<div[^>]+class="comm"[^<]+<h1[^>]+title="([^"]+)"', webpage,
'album name')
album_intro = remove_start(
- clean_html(get_element_by_id("intro", webpage)),
+ clean_html(get_element_by_id('intro', webpage)),
'%s简介:' % album_name)
entries = [
- self.url_result("http://www.kuwo.cn/yinyue/%s/" % song_id, 'Kuwo', song_id)
+ self.url_result('http://www.kuwo.cn/yinyue/%s/' % song_id, 'Kuwo', song_id)
for song_id in re.findall(
r'<p[^>]+class="listen"><a[^>]+href="http://www\.kuwo\.cn/yinyue/(\d+)/"',
webpage)
@@ -176,7 +176,7 @@ class KuwoChartIE(InfoExtractor):
r'<p[^>]+class="tabDef">(\d{4}第\d{2}期)</p>', webpage, 'chart desc')
entries = [
- self.url_result("http://www.kuwo.cn/yinyue/%s/" % song_id, 'Kuwo', song_id)
+ self.url_result('http://www.kuwo.cn/yinyue/%s/' % song_id, 'Kuwo', song_id)
for song_id in re.findall(
r'<a[^>]+href="http://www\.kuwo\.cn/yinyue/(\d+)/"', webpage)
]
@@ -221,7 +221,7 @@ class KuwoSingerIE(InfoExtractor):
errnote='Unable to get song list page #%d' % page_num)
entries.extend([
- self.url_result("http://www.kuwo.cn/yinyue/%s/" % song_id, 'Kuwo', song_id)
+ self.url_result('http://www.kuwo.cn/yinyue/%s/' % song_id, 'Kuwo', song_id)
for song_id in re.findall(
r'<p[^>]+class="m_name"><a[^>]+href="http://www\.kuwo\.cn/yinyue/([0-9]+)/',
webpage)
@@ -256,7 +256,7 @@ class KuwoCategoryIE(InfoExtractor):
r'<h1[^>]+title="([^<>]+?)">[^<>]+?</h1>', webpage, 'category name')
category_desc = remove_start(
- get_element_by_id("intro", webpage).strip(),
+ get_element_by_id('intro', webpage).strip(),
'%s简介:' % category_name)
jsonm = self._parse_json(self._html_search_regex(
@@ -264,7 +264,7 @@ class KuwoCategoryIE(InfoExtractor):
entries = [
self.url_result(
- "http://www.kuwo.cn/yinyue/%s/" % song['musicrid'],
+ 'http://www.kuwo.cn/yinyue/%s/' % song['musicrid'],
'Kuwo', song['musicrid'])
for song in jsonm['musiclist']
]
@@ -301,7 +301,7 @@ class KuwoMvIE(KuwoBaseIE):
song_name = mobj.group('song')
singer_name = mobj.group('singer')
else:
- raise ExtractorError("Unable to find song or singer names")
+ raise ExtractorError('Unable to find song or singer names')
formats = self._get_formats(song_id)