aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl
diff options
context:
space:
mode:
Diffstat (limited to 'youtube_dl')
-rw-r--r--youtube_dl/extractor/mtv.py7
1 files changed, 3 insertions, 4 deletions
diff --git a/youtube_dl/extractor/mtv.py b/youtube_dl/extractor/mtv.py
index c23033d44..a801c8123 100644
--- a/youtube_dl/extractor/mtv.py
+++ b/youtube_dl/extractor/mtv.py
@@ -15,6 +15,7 @@ from ..utils import (
class MTVIE(InfoExtractor):
_VALID_URL = r'^(?P<proto>https?://)?(?:www\.)?mtv\.com/videos/[^/]+/(?P<videoid>[0-9]+)/[^/]+$'
+ _WORKING = False
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
@@ -26,8 +27,8 @@ class MTVIE(InfoExtractor):
webpage = self._download_webpage(url, video_id)
- song_name = self._html_search_regex(r'<meta name="mtv_vt" content="([^"]+)"/>',
- webpage, u'song name', fatal=False)
+ #song_name = self._html_search_regex(r'<meta name="mtv_vt" content="([^"]+)"/>',
+ # webpage, u'song name', fatal=False)
video_title = self._html_search_regex(r'<meta name="mtv_an" content="([^"]+)"/>',
webpage, u'title')
@@ -47,7 +48,6 @@ class MTVIE(InfoExtractor):
raise ExtractorError(u'Unable to download video metadata: %s' % compat_str(err))
mdoc = xml.etree.ElementTree.fromstring(metadataXml)
- print(metadataXml)
renditions = mdoc.findall('.//rendition')
# For now, always pick the highest quality.
@@ -63,7 +63,6 @@ class MTVIE(InfoExtractor):
info = {
'id': video_id,
'url': video_url,
- 'uploader': performer,
'upload_date': None,
'title': video_title,
'ext': ext,