mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-22 18:22:21 +00:00
82 lines
2.9 KiB
Python
82 lines
2.9 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import extract_attributes
|
|
|
|
|
|
class TV2DKIE(InfoExtractor):
|
|
_VALID_URL = r'''(?x)
|
|
https?://
|
|
(?:www\.)?
|
|
(?:
|
|
tvsyd|
|
|
tv2ostjylland|
|
|
tvmidtvest|
|
|
tv2fyn|
|
|
tv2east|
|
|
tv2lorry|
|
|
tv2nord
|
|
)\.dk/
|
|
(:[^/]+/)*
|
|
(?P<id>[^/?\#&]+)
|
|
'''
|
|
_TESTS = [{
|
|
'url': 'https://www.tvsyd.dk/nyheder/28-10-2019/1930/1930-28-okt-2019?autoplay=1#player',
|
|
'info_dict': {
|
|
'id': '0_52jmwa0p',
|
|
'ext': 'mp4',
|
|
'title': '19:30 - 28. okt. 2019',
|
|
'timestamp': 1572290248,
|
|
'upload_date': '20191028',
|
|
'uploader_id': 'tvsyd',
|
|
'duration': 1347,
|
|
'view_count': int,
|
|
},
|
|
'params': {
|
|
'skip_download': True,
|
|
},
|
|
'add_ie': ['Kaltura'],
|
|
}, {
|
|
'url': 'https://www.tv2ostjylland.dk/artikel/minister-gaar-ind-i-sag-om-diabetes-teknologi',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://www.tv2ostjylland.dk/nyheder/28-10-2019/22/2200-nyhederne-mandag-d-28-oktober-2019?autoplay=1#player',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://www.tvmidtvest.dk/nyheder/27-10-2019/1930/1930-27-okt-2019',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://www.tv2fyn.dk/artikel/fyn-kan-faa-landets-foerste-fabrik-til-groent-jetbraendstof',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://www.tv2east.dk/artikel/gods-faar-indleveret-tonsvis-af-aebler-100-kilo-aebler-gaar-til-en-aeblebrandy',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://www.tv2lorry.dk/koebenhavn/rasmus-paludan-evakueret-til-egen-demonstration#player',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://www.tv2nord.dk/artikel/dybt-uacceptabelt',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
entries = []
|
|
for video_el in re.findall(r'(?s)<[^>]+\bdata-entryid\s*=[^>]*>', webpage):
|
|
video = extract_attributes(video_el)
|
|
kaltura_id = video.get('data-entryid')
|
|
if not kaltura_id:
|
|
continue
|
|
partner_id = video.get('data-partnerid')
|
|
if not partner_id:
|
|
continue
|
|
entries.append(self.url_result(
|
|
'kaltura:%s:%s' % (partner_id, kaltura_id), 'Kaltura',
|
|
video_id=kaltura_id))
|
|
return self.playlist_result(entries)
|