|
|
- from __future__ import unicode_literals
-
- from .common import InfoExtractor
-
-
- class MyVidsterIE(InfoExtractor):
- _VALID_URL = r'https?://(?:www\.)?myvidster\.com/video/(?P<id>\d+)/'
-
- _TEST = {
- 'url': 'http://www.myvidster.com/video/32059805/Hot_chemistry_with_raw_love_making',
- 'md5': '95296d0231c1363222c3441af62dc4ca',
- 'info_dict': {
- 'id': '3685814',
- 'title': 'md5:7d8427d6d02c4fbcef50fe269980c749',
- 'upload_date': '20141027',
- 'uploader': 'utkualp',
- 'ext': 'mp4',
- 'age_limit': 18,
- },
- 'add_ie': ['XHamster'],
- }
-
- def _real_extract(self, url):
- video_id = self._match_id(url)
- webpage = self._download_webpage(url, video_id)
-
- return self.url_result(self._html_search_regex(
- r'rel="videolink" href="(?P<real_url>.*)">',
- webpage, 'real video url'))
|