|
@@ -0,0 +1,48 @@
|
|
|
+from __future__ import unicode_literals
|
|
|
+
|
|
|
+from .common import InfoExtractor
|
|
|
+
|
|
|
+
|
|
|
+class FoxgayIE(InfoExtractor):
|
|
|
+ _VALID_URL = r'http://(?:www\.)?foxgay\.com/videos/(?:\S+-)?(?P<id>\d+)\.shtml'
|
|
|
+ _TEST = {
|
|
|
+ 'url': 'http://foxgay.com/videos/fuck-turkish-style-2582.shtml',
|
|
|
+ 'md5': '80d72beab5d04e1655a56ad37afe6841',
|
|
|
+ 'info_dict': {
|
|
|
+ 'id': '2582',
|
|
|
+ 'ext': 'mp4',
|
|
|
+ 'title': 'md5:6122f7ae0fc6b21ebdf59c5e083ce25a',
|
|
|
+ 'description': 'md5:5e51dc4405f1fd315f7927daed2ce5cf',
|
|
|
+ 'age_limit': 18,
|
|
|
+ },
|
|
|
+ }
|
|
|
+
|
|
|
+ def _real_extract(self, url):
|
|
|
+ video_id = self._match_id(url)
|
|
|
+
|
|
|
+ # Download the initial webpage and extract all available data.
|
|
|
+ webpage = self._download_webpage(url, video_id)
|
|
|
+ title = self._html_search_regex(
|
|
|
+ r'<title>(?P<title>.*?)</title>',
|
|
|
+ webpage, 'title', fatal=False)
|
|
|
+ description = self._html_search_regex(
|
|
|
+ r'<div class="ico_desc"><h2>(?P<description>.*?)</h2>',
|
|
|
+ webpage, 'description', fatal=False)
|
|
|
+
|
|
|
+ # Find the URL for the iFrame which contains the actual video.
|
|
|
+ iframe = self._download_webpage(
|
|
|
+ self._html_search_regex(r'iframe src="(?P<frame>.*?)"', webpage, 'video frame'),
|
|
|
+ video_id)
|
|
|
+ video_url = self._html_search_regex(
|
|
|
+ r"v_path = '(?P<vid>http://.*?)'", iframe, 'url')
|
|
|
+ thumb_url = self._html_search_regex(
|
|
|
+ r"t_path = '(?P<thumb>http://.*?)'", iframe, 'thumbnail', fatal=False)
|
|
|
+
|
|
|
+ return {
|
|
|
+ 'id': video_id,
|
|
|
+ 'title': title,
|
|
|
+ 'url': video_url,
|
|
|
+ 'description': description,
|
|
|
+ 'thumbnail': thumb_url,
|
|
|
+ 'age_limit': 18,
|
|
|
+ }
|