yt-dlc/youtube_dl/extractor/vidzi.py

30 lines
984 B
Python
Raw Normal View History

2014-10-17 16:38:16 +00:00
import re
from .common import InfoExtractor
class VidziIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?vidzi\.tv/(?P<id>\w+)'
_TEST = {
'url': 'http://vidzi.tv/m1chxrwq7bx9',
'md5': '5c4c4a8ca2281a199c8eefe8f411d630',
'info_dict': {
'id': 'm1chxrwq7bx9',
'ext': 'mp4',
'title': 'Watch Cadbury Dream Factory S01E04 HDTV x264 FiHTV mp4',
},
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group('id')
webpage = self._download_webpage('http://vidzi.tv/' + video_id, video_id)
video_url = self._html_search_regex(r'{\s*file\s*:\s*"([^"]+)"\s*}', webpage, u'vidzi url')
title = self._html_search_regex(r'<Title>([^<]+)<\/Title>', webpage, u'vidzi title')
return {
'id': video_id,
'title': title,
'url': video_url,
'ext': 'mp4',
}