1
0
Fork 0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-11-18 16:26:23 +00:00
youtube-dl/youtube_dl/extractor/gbnews.py

140 lines
5 KiB
Python
Raw Normal View History

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
extract_attributes,
ExtractorError,
T,
traverse_obj,
txt_or_none,
url_or_none,
)
class GBNewsIE(InfoExtractor):
IE_DESC = 'GB News clips, features and live stream'
# \w+ is normally shows or news, but apparently any word redirects to the correct URL
_VALID_URL = r'https?://(?:www\.)?gbnews\.(?:uk|com)/(?:\w+/)?(?P<id>[^#?]+)'
_PLATFORM = 'safari'
_SSMP_URL = 'https://mm-v2.simplestream.com/ssmp/api.php'
_TESTS = [{
'url': 'https://www.gbnews.uk/shows/andrew-neils-message-to-companies-choosing-to-boycott-gb-news/106889',
'info_dict': {
'id': '106889',
'ext': 'mp4',
'title': "Andrew Neil's message to companies choosing to boycott GB News",
'description': 'md5:b281f5d22fd6d5eda64a4e3ba771b351',
},
'skip': '404 not found',
}, {
'url': 'https://www.gbnews.com/news/bbc-claudine-gay-harvard-university-antisemitism-row',
'info_dict': {
'id': '52264136',
'display_id': 'bbc-claudine-gay-harvard-university-antisemitism-row',
'ext': 'mp4',
'title': 'BBC deletes post after furious backlash over headline downplaying antisemitism',
'description': 'The post was criticised by former employers of the broadcaster',
},
}, {
'url': 'https://www.gbnews.uk/watchlive',
'info_dict': {
'id': '1069',
'display_id': 'watchlive',
'ext': 'mp4',
'title': 'GB News Live',
'is_live': True,
},
'params': {
'skip_download': 'm3u8',
},
}]
def _real_extract(self, url):
display_id = self._match_id(url).split('/')[-1]
webpage = self._download_webpage(url, display_id)
# extraction based on https://github.com/ytdl-org/youtube-dl/issues/29341
'''
<div id="video-106908"
class="simplestream"
data-id="GB001"
data-type="vod"
data-key="3Li3Nt2Qs8Ct3Xq9Fi5Uy0Mb2Bj0Qs"
data-token="f9c317c727dc07f515b20036c8ef14a6"
data-expiry="1624300052"
data-uvid="37900558"
data-poster="https://thumbnails.simplestreamcdn.com/gbnews/ondemand/37900558.jpg?width=700&"
data-npaw="false"
data-env="production">
'''
# exception if no match
video_data = self._search_regex(
r'(<div\s[^>]*\bclass\s*=\s*(\'|")(?!.*sidebar\b)simplestream(?:\s[\s\w$-]*)?\2[^>]*>)',
webpage, 'video data')
video_data = extract_attributes(video_data)
ss_id = video_data.get('data-id')
if not ss_id:
raise ExtractorError('Simplestream ID not found')
json_data = self._download_json(
self._SSMP_URL, display_id,
note='Downloading Simplestream JSON metadata',
errnote='Unable to download Simplestream JSON metadata',
query={
'id': ss_id,
'env': video_data.get('data-env', 'production'),
}, fatal=False)
meta_url = traverse_obj(json_data, ('response', 'api_hostname'))
if not meta_url:
raise ExtractorError('No API host found')
uvid = video_data['data-uvid']
dtype = video_data.get('data-type')
stream_data = self._download_json(
'%s/api/%s/stream/%s' % (meta_url, 'show' if dtype == 'vod' else dtype, uvid),
uvid,
query={
'key': video_data.get('data-key'),
'platform': self._PLATFORM,
},
headers={
'Token': video_data.get('data-token'),
'Token-Expiry': video_data.get('data-expiry'),
'Uvid': uvid,
}, fatal=False)
stream_url = traverse_obj(stream_data, (
'response', 'stream', T(url_or_none)))
if not stream_url:
raise ExtractorError('No stream data/URL')
# now known to be a dict
stream_data = stream_data['response']
drm = stream_data.get('drm')
if drm:
self.report_drm(uvid)
formats = self._extract_m3u8_formats(
stream_url, uvid, ext='mp4', entry_protocol='m3u8_native',
fatal=False)
# exception if no formats
self._sort_formats(formats)
return {
'id': uvid,
'display_id': display_id,
'title': (traverse_obj(stream_data, ('title', T(txt_or_none)))
or self._og_search_title(webpage, default=None)
or display_id.replace('-', ' ').capitalize()),
'description': self._og_search_description(webpage, default=None),
'thumbnail': (traverse_obj(video_data, ('data-poster', T(url_or_none)))
or self._og_search_thumbnail(webpage)),
'formats': formats,
'is_live': (dtype == 'live') or None,
}