mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-03 17:00:51 +01:00
Fixed DailymotionIE, now downloads high-def mp4s, which might be too much (?)
This commit is contained in:
parent
551fa9dfbf
commit
349e2e3e21
BIN
youtube-dl
BIN
youtube-dl
Binary file not shown.
|
@ -617,7 +617,7 @@ def _real_extract(self, url):
|
|||
|
||||
video_id = mobj.group(1)
|
||||
|
||||
video_extension = 'flv'
|
||||
video_extension = 'mp4'
|
||||
|
||||
# Retrieve video webpage to extract further information
|
||||
request = urllib2.Request(url)
|
||||
|
@ -631,20 +631,18 @@ def _real_extract(self, url):
|
|||
|
||||
# Extract URL, uploader and title from webpage
|
||||
self.report_extraction(video_id)
|
||||
mobj = re.search(r'(?i)addVariable\(\"sequence\"\s*,\s*\"([^\"]+?)\"\)', webpage)
|
||||
mobj = re.search(r'\s*var flashvars = (.*)', webpage)
|
||||
if mobj is None:
|
||||
self._downloader.trouble(u'ERROR: unable to extract media URL')
|
||||
return
|
||||
sequence = urllib.unquote(mobj.group(1))
|
||||
mobj = re.search(r',\"sdURL\"\:\"([^\"]+?)\",', sequence)
|
||||
flashvars = urllib.unquote(mobj.group(1))
|
||||
mobj = re.search(r'"hqURL":"(.+?)"', flashvars)
|
||||
if mobj is None:
|
||||
self._downloader.trouble(u'ERROR: unable to extract media URL')
|
||||
return
|
||||
mediaURL = urllib.unquote(mobj.group(1)).replace('\\', '')
|
||||
hqURL = mobj.group(1).replace('\\/', '/')
|
||||
|
||||
# if needed add http://www.dailymotion.com/ if relative URL
|
||||
|
||||
video_url = mediaURL
|
||||
# TODO: support ldurl and sdurl qualities
|
||||
|
||||
mobj = re.search(r'<meta property="og:title" content="(?P<title>[^"]*)" />', webpage)
|
||||
if mobj is None:
|
||||
|
@ -660,7 +658,7 @@ def _real_extract(self, url):
|
|||
|
||||
return [{
|
||||
'id': video_id.decode('utf-8'),
|
||||
'url': video_url.decode('utf-8'),
|
||||
'url': hqURL.decode('utf-8'),
|
||||
'uploader': video_uploader.decode('utf-8'),
|
||||
'upload_date': u'NA',
|
||||
'title': video_title,
|
||||
|
|
Loading…
Reference in New Issue
Block a user