mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-18 15:11:45 +00:00
vidzi Add new extractor
This commit is contained in:
parent
2acfe95f58
commit
95ee84421e
@ -493,7 +493,7 @@ from .youtube import (
|
||||
)
|
||||
|
||||
from .zdf import ZDFIE
|
||||
|
||||
from .vidzi import VidziIE
|
||||
|
||||
_ALL_CLASSES = [
|
||||
klass
|
||||
|
30
youtube_dl/extractor/vidzi.py
Normal file
30
youtube_dl/extractor/vidzi.py
Normal file
@ -0,0 +1,30 @@
|
||||
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',
|
||||
}
|
||||
|
Loading…
Reference in New Issue
Block a user