import re
+import json
from .common import InfoExtractor
from ..utils import (
compat_urllib_request,
- compat_urllib_parse,
ExtractorError,
unescapeHTML,
# Extract URL, uploader and title from webpage
self.report_extraction(video_id)
- mobj = re.search(r'\s*var flashvars = (.*)', webpage)
- if mobj is None:
- raise ExtractorError(u'Unable to extract media URL')
- flashvars = compat_urllib_parse.unquote(mobj.group(1))
- for key in ['hd1080URL', 'hd720URL', 'hqURL', 'sdURL', 'ldURL', 'video_url']:
- if key in flashvars:
- max_quality = key
- self.to_screen(u'Using %s' % key)
- break
- else:
- raise ExtractorError(u'Unable to extract video URL')
-
- mobj = re.search(r'"' + max_quality + r'":"(.+?)"', flashvars)
- if mobj is None:
- raise ExtractorError(u'Unable to extract video URL')
-
- video_url = compat_urllib_parse.unquote(mobj.group(1)).replace('\\/', '/')
-
- # TODO: support choosing qualities
mobj = re.search(r'<meta property="og:title" content="(?P<title>[^"]*)" />', webpage)
if mobj is None:
if mobj is not None:
video_upload_date = mobj.group(3) + mobj.group(2) + mobj.group(1)
+ embed_url = 'http://www.dailymotion.com/embed/video/%s' % video_id
+ embed_page = self._download_webpage(embed_url, video_id,
+ u'Downloading embed page')
+ info = self._search_regex(r'var info = ({.*?}),', embed_page, 'video info')
+ info = json.loads(info)
+
+ # TODO: support choosing qualities
+
+ for key in ['stream_h264_hd1080_url','stream_h264_hd_url',
+ 'stream_h264_hq_url','stream_h264_url',
+ 'stream_h264_ld_url']:
+ if info.get(key):#key in info and info[key]:
+ max_quality = key
+ self.to_screen(u'Using %s' % key)
+ break
+ else:
+ raise ExtractorError(u'Unable to extract video URL')
+ video_url = info[max_quality]
+
return [{
'id': video_id,
'url': video_url,