webpage = self._download_webpage(url, video_id, 'Downloading page')
mobj = re.search(r'(?m)fo\.addVariable\("file",\s"(?P<fileid>[\da-z]+)"\);\n'
- '\s+fo\.addVariable\("s",\s"(?P<serverid>\d+)"\);', webpage)
+ r'\s+fo\.addVariable\("s",\s"(?P<serverid>\d+)"\);', webpage)
if mobj is None:
raise ExtractorError('Video %s does not exist' % video_id, expected=True)