1
0
mirror of https://codeberg.org/polarisfm/youtube-dl synced 2024-11-05 01:44:33 +01:00
youtube-dl/youtube_dl/extractor/xvideos.py

54 lines
1.5 KiB
Python
Raw Normal View History

import re
from .common import InfoExtractor
from ..utils import (
compat_urllib_parse,
)
class XVideosIE(InfoExtractor):
_VALID_URL = r'^(?:https?://)?(?:www\.)?xvideos\.com/video([0-9]+)(?:.*)'
2013-06-27 20:46:46 +02:00
_TEST = {
u'url': u'http://www.xvideos.com/video939581/funny_porns_by_s_-1',
u'file': u'939581.flv',
u'md5': u'1d0c835822f0a71a7bf011855db929d0',
u'info_dict': {
u"title": u"Funny Porns By >>>>S<<<<<< -1",
u"age_limit": 18,
2013-06-27 20:46:46 +02:00
}
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group(1)
webpage = self._download_webpage(url, video_id)
self.report_extraction(video_id)
# Extract video URL
video_url = compat_urllib_parse.unquote(self._search_regex(r'flv_url=(.+?)&',
webpage, u'video URL'))
# Extract title
video_title = self._html_search_regex(r'<title>(.*?)\s+-\s+XVID',
webpage, u'title')
# Extract video thumbnail
video_thumbnail = self._search_regex(r'http://(?:img.*?\.)xvideos.com/videos/thumbs/[a-fA-F0-9]+/[a-fA-F0-9]+/[a-fA-F0-9]+/[a-fA-F0-9]+/([a-fA-F0-9.]+jpg)',
webpage, u'thumbnail', fatal=False)
info = {
'id': video_id,
'url': video_url,
'uploader': None,
'upload_date': None,
'title': video_title,
'ext': 'flv',
'thumbnail': video_thumbnail,
'description': None,
'age_limit': 18,
}
return [info]