mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-20 12:24:57 +00:00
41 lines
1.4 KiB
Python
41 lines
1.4 KiB
Python
import re
|
|
# -*- coding: utf-8 -*-
|
|
# needed for the title french ê! coding utf-8- -*-
|
|
# based on the vine.co and lots of help from https://filippo.io/add-support-for-a-new-video-site-to-youtube-dl/
|
|
from .common import InfoExtractor
|
|
|
|
|
|
class TelembIE(InfoExtractor):
|
|
|
|
_VALID_URL = r'https?://www\.telemb\.be/(?P<id>.*)'
|
|
|
|
_TEST = {
|
|
u'url': u'http://www.telemb.be/mons-cook-with-danielle-des-cours-de-cuisine-en-anglais-_d_13466.html',
|
|
u'file': u'mons-cook-with-danielle-des-cours-de-cuisine-en-anglais-_d_13466.html.mp4',
|
|
u'md5': u'f45ea69878516ba039835794e0f8f783',
|
|
u'info_dict': {
|
|
u"title": u'TéléMB : Mons - Cook with Danielle : des cours de cuisine en anglais ! - Les reportages'
|
|
}
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
mobj = re.match(self._VALID_URL, url)
|
|
|
|
video_id = mobj.group('id')
|
|
webpage_url = 'http://www.telemb.be/' + video_id
|
|
webpage = self._download_webpage(webpage_url, video_id)
|
|
|
|
|
|
self.report_extraction(video_id)
|
|
|
|
video_url = self._html_search_regex(r'"(http://wowza\.imust\.org/srv/vod/.*\.mp4)"',
|
|
webpage, u'video URL')
|
|
|
|
return [{
|
|
'id': video_id,
|
|
'url': video_url,
|
|
'ext': 'mp4',
|
|
'title': self._og_search_title(webpage),
|
|
'thumbnail': self._og_search_thumbnail(webpage),
|
|
}]
|