mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-12-27 15:20:29 +00:00
38 lines
1.4 KiB
Python
38 lines
1.4 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import unescapeHTML
|
|
|
|
|
|
class VODPlatformIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?vod-platform\.net/[eE]mbed/(?P<id>[^/?#]+)'
|
|
_TEST = {
|
|
# from http://www.lbcgroup.tv/watch/chapter/29143/52844/%D8%A7%D9%84%D9%86%D8%B5%D8%B1%D8%A9-%D9%81%D9%8A-%D8%B6%D9%8A%D8%A7%D9%81%D8%A9-%D8%A7%D9%84%D9%80-cnn/ar
|
|
'url': 'http://vod-platform.net/embed/RufMcytHDolTH1MuKHY9Fw',
|
|
'md5': '1db2b7249ce383d6be96499006e951fc',
|
|
'info_dict': {
|
|
'id': 'RufMcytHDolTH1MuKHY9Fw',
|
|
'ext': 'mp4',
|
|
'title': 'LBCi News_ النصرة في ضيافة الـ "سي.أن.أن"',
|
|
}
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
title = unescapeHTML(self._og_search_title(webpage))
|
|
hidden_inputs = self._hidden_inputs(webpage)
|
|
|
|
formats = self._extract_wowza_formats(
|
|
hidden_inputs.get('HiddenmyhHlsLink') or hidden_inputs['HiddenmyDashLink'], video_id, skip_protocols=['f4m', 'smil'])
|
|
self._sort_formats(formats)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': title,
|
|
'thumbnail': hidden_inputs.get('HiddenThumbnail') or self._og_search_thumbnail(webpage),
|
|
'formats': formats,
|
|
}
|