+from __future__ import unicode_literals
+
import os
import re
from .common import InfoExtractor
-from ..utils import (
+from ..compat import (
+ compat_urllib_parse_unquote,
compat_urllib_parse_urlparse,
- compat_urllib_request,
- compat_urllib_parse,
)
+from ..utils import sanitized_Request
+
class MofosexIE(InfoExtractor):
- _VALID_URL = r'^(?:https?://)?(?:www\.)?(?P<url>mofosex\.com/videos/(?P<videoid>[0-9]+)/.*?\.html)'
+ _VALID_URL = r'https?://(?:www\.)?(?P<url>mofosex\.com/videos/(?P<id>[0-9]+)/.*?\.html)'
_TEST = {
- u'url': u'http://www.mofosex.com/videos/5018/japanese-teen-music-video.html',
- u'file': u'5018.mp4',
- u'md5': u'1b2eb47ac33cc75d4a80e3026b613c5a',
- u'info_dict': {
- u"title": u"Japanese Teen Music Video",
- u"age_limit": 18,
+ 'url': 'http://www.mofosex.com/videos/5018/japanese-teen-music-video.html',
+ 'md5': '1b2eb47ac33cc75d4a80e3026b613c5a',
+ 'info_dict': {
+ 'id': '5018',
+ 'ext': 'mp4',
+ 'title': 'Japanese Teen Music Video',
+ 'age_limit': 18,
}
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
- video_id = mobj.group('videoid')
+ video_id = mobj.group('id')
url = 'http://www.' + mobj.group('url')
- req = compat_urllib_request.Request(url)
+ req = sanitized_Request(url)
req.add_header('Cookie', 'age_verified=1')
webpage = self._download_webpage(req, video_id)
- video_title = self._html_search_regex(r'<h1>(.+?)<', webpage, u'title')
- video_url = compat_urllib_parse.unquote(self._html_search_regex(r'flashvars.video_url = \'([^\']+)', webpage, u'video_url'))
+ video_title = self._html_search_regex(r'<h1>(.+?)<', webpage, 'title')
+ video_url = compat_urllib_parse_unquote(self._html_search_regex(r'flashvars.video_url = \'([^\']+)', webpage, 'video_url'))
path = compat_urllib_parse_urlparse(video_url).path
extension = os.path.splitext(path)[1][1:]
format = path.split('/')[5].split('_')[:2]