2018-02-24 20:42:41 +01:00
|
|
|
from __future__ import unicode_literals
|
2018-02-11 23:09:08 +01:00
|
|
|
from .common import InfoExtractor
|
|
|
|
class UninettunoIE(InfoExtractor):
|
|
|
|
_VALID_URL = r'(?:https?://)?(?:www\.)?uninettuno\.tv/Video.aspx\?v\=(?P<id>\w+)'
|
|
|
|
def _real_extract(self, url):
|
|
|
|
mobj = re.match(self._VALID_URL, url)
|
|
|
|
video_id = mobj.group('id')
|
|
|
|
webpage_url = 'https://www.uninettuno.tv/Video.aspx?v=' + video_id
|
|
|
|
webpage = self._download_webpage(webpage_url, video_id)
|
|
|
|
self.report_extraction(video_id)
|
2018-02-24 20:50:27 +01:00
|
|
|
video_url = self._html_search_regex(r'[{sources: [{ file: "//(.+?)" ,type: "mp4" }]', webpage, 'video URL')
|
2018-02-11 23:09:08 +01:00
|
|
|
return [{
|
|
|
|
'id': video_id,
|
|
|
|
'url': video_url,
|
|
|
|
'ext': 'mp4',
|
|
|
|
'title': self._og_search_title(webpage),
|
|
|
|
}]
|