0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-04 23:35:04 +00:00
yt-dlp/youtube_dl/extractor/bloomberg.py
Yen Chi Hsuan a3fa6024d6
[bloomberg] Fix test_Bloomberg
In this test case, sometimes HLS is the best format while sometimes HDS
is. To prevent occasional test failures, force HDS to be the best
format. In the past, testing against HDS formats causes the same error
as #9214, which is fixed as #9377 landed.
2016-05-12 20:08:42 +08:00

60 lines
2.2 KiB
Python

from __future__ import unicode_literals
import re
from .common import InfoExtractor
class BloombergIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?bloomberg\.com/(?:[^/]+/)*(?P<id>[^/?#]+)'
_TESTS = [{
'url': 'http://www.bloomberg.com/news/videos/b/aaeae121-5949-481e-a1ce-4562db6f5df2',
# The md5 checksum changes
'info_dict': {
'id': 'qurhIVlJSB6hzkVi229d8g',
'ext': 'flv',
'title': 'Shah\'s Presentation on Foreign-Exchange Strategies',
'description': 'md5:a8ba0302912d03d246979735c17d2761',
},
'params': {
'format': 'best[format_id^=hds]',
},
}, {
'url': 'http://www.bloomberg.com/news/articles/2015-11-12/five-strange-things-that-have-been-happening-in-financial-markets',
'only_matching': True,
}, {
'url': 'http://www.bloomberg.com/politics/videos/2015-11-25/karl-rove-on-jeb-bush-s-struggles-stopping-trump',
'only_matching': True,
}]
def _real_extract(self, url):
name = self._match_id(url)
webpage = self._download_webpage(url, name)
video_id = self._search_regex(
r'["\']bmmrId["\']\s*:\s*(["\'])(?P<url>.+?)\1',
webpage, 'id', group='url')
title = re.sub(': Video$', '', self._og_search_title(webpage))
embed_info = self._download_json(
'http://www.bloomberg.com/api/embed?id=%s' % video_id, video_id)
formats = []
for stream in embed_info['streams']:
stream_url = stream.get('url')
if not stream_url:
continue
if stream['muxing_format'] == 'TS':
formats.extend(self._extract_m3u8_formats(
stream_url, video_id, 'mp4', m3u8_id='hls', fatal=False))
else:
formats.extend(self._extract_f4m_formats(
stream_url, video_id, f4m_id='hds', fatal=False))
self._sort_formats(formats)
return {
'id': video_id,
'title': title,
'formats': formats,
'description': self._og_search_description(webpage),
'thumbnail': self._og_search_thumbnail(webpage),
}