mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-01-07 17:16:08 +00:00
[vimeo] Delegate ondemand redirects to ondemand extractor (closes #10994)
This commit is contained in:
parent
9dc13a6780
commit
f16f8505b1
|
@ -322,6 +322,22 @@ class VimeoIE(VimeoBaseInfoExtractor):
|
|||
},
|
||||
'expected_warnings': ['Unable to download JSON metadata'],
|
||||
},
|
||||
{
|
||||
# redirects to ondemand extractor and should be passed throught it
|
||||
# for successful extraction
|
||||
'url': 'https://vimeo.com/73445910',
|
||||
'info_dict': {
|
||||
'id': '73445910',
|
||||
'ext': 'mp4',
|
||||
'title': 'The Reluctant Revolutionary',
|
||||
'uploader': '10Ft Films',
|
||||
'uploader_url': 're:https?://(?:www\.)?vimeo\.com/tenfootfilms',
|
||||
'uploader_id': 'tenfootfilms',
|
||||
},
|
||||
'params': {
|
||||
'skip_download': True,
|
||||
},
|
||||
},
|
||||
{
|
||||
'url': 'http://vimeo.com/moogaloop.swf?clip_id=2539741',
|
||||
'only_matching': True,
|
||||
|
@ -414,7 +430,12 @@ class VimeoIE(VimeoBaseInfoExtractor):
|
|||
# Retrieve video webpage to extract further information
|
||||
request = sanitized_Request(url, headers=headers)
|
||||
try:
|
||||
webpage = self._download_webpage(request, video_id)
|
||||
webpage, urlh = self._download_webpage_handle(request, video_id)
|
||||
# Some URLs redirect to ondemand can't be extracted with
|
||||
# this extractor right away thus should be passed through
|
||||
# ondemand extractor (e.g. https://vimeo.com/73445910)
|
||||
if VimeoOndemandIE.suitable(urlh.geturl()):
|
||||
return self.url_result(urlh.geturl(), VimeoOndemandIE.ie_key())
|
||||
except ExtractorError as ee:
|
||||
if isinstance(ee.cause, compat_HTTPError) and ee.cause.code == 403:
|
||||
errmsg = ee.cause.read()
|
||||
|
|
Loading…
Reference in a new issue