from __future__ import unicode_literals
-import re
-
from .common import InfoExtractor
'info_dict': {
'id': 'inyouchuu-etsu-bonus',
'ext': 'mp4',
- "title": "Inyouchuu Etsu Bonus",
- "age_limit": 18,
+ 'title': 'Inyouchuu Etsu Bonus',
+ 'age_limit': 18,
}
}
def _real_extract(self, url):
- mobj = re.match(self._VALID_URL, url)
- video_id = mobj.group('id')
+ video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
title = self._html_search_regex(
- r'<h2 class="posttitle"><a[^>]*>([^<]+)</a>',
+ r'<h2[^>]+class="posttitle"[^>]*><a[^>]*>([^<]+)</a>',
webpage, 'title')
wrap_url = self._html_search_regex(
- r'<iframe src="([^"]+mp4)"', webpage, 'wrapper url')
+ r'<iframe[^>]+src="([^"]+mp4)"', webpage, 'wrapper url')
wrap_webpage = self._download_webpage(wrap_url, video_id)
video_url = self._html_search_regex(
- r'clip:\s*{\s*url: "([^"]*)"', wrap_webpage, 'video url')
+ r'file\s*:\s*"([^"]+)"', wrap_webpage, 'video url')
return {
'id': video_id,