mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-25 04:04:35 +01:00
109 lines
3.9 KiB
Python
109 lines
3.9 KiB
Python
# encoding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
from .cbs import CBSIE
|
|
from ..utils import (
|
|
parse_duration,
|
|
)
|
|
|
|
|
|
class CBSNewsIE(CBSIE):
|
|
IE_DESC = 'CBS News'
|
|
_VALID_URL = r'https?://(?:www\.)?cbsnews\.com/(?:news|videos)/(?P<id>[\da-z_-]+)'
|
|
|
|
_TESTS = [
|
|
{
|
|
'url': 'http://www.cbsnews.com/news/tesla-and-spacex-elon-musks-industrial-empire/',
|
|
'info_dict': {
|
|
'id': 'tesla-and-spacex-elon-musks-industrial-empire',
|
|
'ext': 'flv',
|
|
'title': 'Tesla and SpaceX: Elon Musk\'s industrial empire',
|
|
'thumbnail': 'http://beta.img.cbsnews.com/i/2014/03/30/60147937-2f53-4565-ad64-1bdd6eb64679/60-0330-pelley-640x360.jpg',
|
|
'duration': 791,
|
|
},
|
|
'params': {
|
|
# rtmp download
|
|
'skip_download': True,
|
|
},
|
|
'skip': 'Subscribers only',
|
|
},
|
|
{
|
|
'url': 'http://www.cbsnews.com/videos/fort-hood-shooting-army-downplays-mental-illness-as-cause-of-attack/',
|
|
'info_dict': {
|
|
'id': 'SNJBOYzXiWBOvaLsdzwH8fmtP1SCd91Y',
|
|
'ext': 'mp4',
|
|
'title': 'Fort Hood shooting: Army downplays mental illness as cause of attack',
|
|
'description': 'md5:4a6983e480542d8b333a947bfc64ddc7',
|
|
'upload_date': '20140404',
|
|
'timestamp': 1396650660,
|
|
'uploader': 'CBSI-NEW',
|
|
'thumbnail': 're:^https?://.*\.jpg$',
|
|
'duration': 205,
|
|
'subtitles': {
|
|
'en': [{
|
|
'ext': 'ttml',
|
|
}],
|
|
},
|
|
},
|
|
'params': {
|
|
# m3u8 download
|
|
'skip_download': True,
|
|
},
|
|
},
|
|
]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
video_info = self._parse_json(self._html_search_regex(
|
|
r'(?:<ul class="media-list items" id="media-related-items"><li data-video-info|<div id="cbsNewsVideoPlayer" data-video-player-options)=\'({.+?})\'',
|
|
webpage, 'video JSON info'), video_id)
|
|
|
|
item = video_info['item'] if 'item' in video_info else video_info
|
|
guid = item['mpxRefId']
|
|
return self._extract_video_info(guid)
|
|
|
|
|
|
class CBSNewsLiveVideoIE(InfoExtractor):
|
|
IE_DESC = 'CBS News Live Videos'
|
|
_VALID_URL = r'https?://(?:www\.)?cbsnews\.com/live/video/(?P<id>[\da-z_-]+)'
|
|
|
|
# Live videos get deleted soon. See http://www.cbsnews.com/live/ for the latest examples
|
|
_TEST = {
|
|
'url': 'http://www.cbsnews.com/live/video/clinton-sanders-prepare-to-face-off-in-nh/',
|
|
'info_dict': {
|
|
'id': 'clinton-sanders-prepare-to-face-off-in-nh',
|
|
'ext': 'flv',
|
|
'title': 'Clinton, Sanders Prepare To Face Off In NH',
|
|
'duration': 334,
|
|
},
|
|
'skip': 'Video gone',
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
video_info = self._parse_json(self._html_search_regex(
|
|
r'data-story-obj=\'({.+?})\'', webpage, 'video JSON info'), video_id)['story']
|
|
|
|
hdcore_sign = 'hdcore=3.3.1'
|
|
f4m_formats = self._extract_f4m_formats(video_info['url'] + '&' + hdcore_sign, video_id)
|
|
if f4m_formats:
|
|
for entry in f4m_formats:
|
|
# URLs without the extra param induce an 404 error
|
|
entry.update({'extra_param_to_segment_url': hdcore_sign})
|
|
self._sort_formats(f4m_formats)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': video_info['headline'],
|
|
'thumbnail': video_info.get('thumbnail_url_hd') or video_info.get('thumbnail_url_sd'),
|
|
'duration': parse_duration(video_info.get('segmentDur')),
|
|
'formats': f4m_formats,
|
|
}
|