|
|
|
@ -7,6 +7,7 @@ from .common import InfoExtractor
|
|
|
|
|
from ..utils import (
|
|
|
|
|
compat_urllib_request,
|
|
|
|
|
int_or_none,
|
|
|
|
|
ExtractorError,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
@ -94,8 +95,12 @@ class VeohIE(InfoExtractor):
|
|
|
|
|
if video_id.startswith('v'):
|
|
|
|
|
rsp = self._download_xml(
|
|
|
|
|
r'http://www.veoh.com/api/findByPermalink?permalink=%s' % video_id, video_id, 'Downloading video XML')
|
|
|
|
|
if rsp.get('stat') == 'ok':
|
|
|
|
|
stat = rsp.get('stat')
|
|
|
|
|
if stat == 'ok':
|
|
|
|
|
return self._extract_video(rsp.find('./videoList/video'))
|
|
|
|
|
elif stat == 'fail':
|
|
|
|
|
raise ExtractorError(
|
|
|
|
|
'%s said: %s' % (self.IE_NAME, rsp.find('./errorList/error').get('errorMessage')), expected=True)
|
|
|
|
|
|
|
|
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
|
age_limit = 0
|
|
|
|
|