mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-12-22 16:57:40 +00:00
BrightcoveIE: add support for playlists
This commit is contained in:
parent
a431154706
commit
abb285fb1b
1 changed files with 30 additions and 5 deletions
|
@ -8,8 +8,9 @@ from ..utils import (
|
|||
)
|
||||
|
||||
class BrightcoveIE(InfoExtractor):
|
||||
_VALID_URL = r'http://.*brightcove\.com/.*\?(?P<query>.*videoPlayer=(?P<id>\d*).*)'
|
||||
_VALID_URL = r'https?://.*brightcove\.com/(services|viewer).*\?(?P<query>.*)'
|
||||
_FEDERATED_URL_TEMPLATE = 'http://c.brightcove.com/services/viewer/htmlFederated?%s'
|
||||
_PLAYLIST_URL_TEMPLATE = 'http://c.brightcove.com/services/json/experience/runtime/?command=get_programming_for_experience&playerKey=%s'
|
||||
|
||||
# There is a test for Brigtcove in GenericIE, that way we test both the download
|
||||
# and the detection of videos, and we don't have to find an URL that is always valid
|
||||
|
@ -24,20 +25,30 @@ class BrightcoveIE(InfoExtractor):
|
|||
assert object_doc.attrib['class'] == u'BrightcoveExperience'
|
||||
params = {'flashID': object_doc.attrib['id'],
|
||||
'playerID': object_doc.find('./param[@name="playerID"]').attrib['value'],
|
||||
'@videoPlayer': object_doc.find('./param[@name="@videoPlayer"]').attrib['value'],
|
||||
}
|
||||
playerKey = object_doc.find('./param[@name="playerKey"]')
|
||||
# Not all pages define this value
|
||||
if playerKey is not None:
|
||||
params['playerKey'] = playerKey.attrib['value']
|
||||
videoPlayer = object_doc.find('./param[@name="@videoPlayer"]')
|
||||
if videoPlayer is not None:
|
||||
params['@videoPlayer'] = videoPlayer.attrib['value']
|
||||
data = compat_urllib_parse.urlencode(params)
|
||||
return cls._FEDERATED_URL_TEMPLATE % data
|
||||
|
||||
def _real_extract(self, url):
|
||||
mobj = re.match(self._VALID_URL, url)
|
||||
query = mobj.group('query')
|
||||
video_id = mobj.group('id')
|
||||
|
||||
m_video_id = re.search(r'videoPlayer=(\d+)', query)
|
||||
if m_video_id is not None:
|
||||
video_id = m_video_id.group(1)
|
||||
return self._get_video_info(video_id, query)
|
||||
else:
|
||||
player_key = self._search_regex(r'playerKey=(.+?)(&|$)', query, 'playlist_id')
|
||||
return self._get_playlist_info(player_key)
|
||||
|
||||
def _get_video_info(self, video_id, query):
|
||||
request_url = self._FEDERATED_URL_TEMPLATE % query
|
||||
webpage = self._download_webpage(request_url, video_id)
|
||||
|
||||
|
@ -45,11 +56,25 @@ class BrightcoveIE(InfoExtractor):
|
|||
info = self._search_regex(r'var experienceJSON = ({.*?});', webpage, 'json')
|
||||
info = json.loads(info)['data']
|
||||
video_info = info['programmedContent']['videoPlayer']['mediaDTO']
|
||||
|
||||
return self._extract_video_info(video_info)
|
||||
|
||||
def _get_playlist_info(self, player_key):
|
||||
playlist_info = self._download_webpage(self._PLAYLIST_URL_TEMPLATE % player_key,
|
||||
player_key, u'Downloading playlist information')
|
||||
|
||||
playlist_info = json.loads(playlist_info)['videoList']
|
||||
videos = [self._extract_video_info(video_info) for video_info in playlist_info['mediaCollectionDTO']['videoDTOs']]
|
||||
|
||||
return self.playlist_result(videos, playlist_id=playlist_info['id'],
|
||||
playlist_title=playlist_info['mediaCollectionDTO']['displayName'])
|
||||
|
||||
def _extract_video_info(self, video_info):
|
||||
renditions = video_info['renditions']
|
||||
renditions = sorted(renditions, key=lambda r: r['size'])
|
||||
best_format = renditions[-1]
|
||||
|
||||
return {'id': video_id,
|
||||
return {'id': video_info['id'],
|
||||
'title': video_info['displayName'],
|
||||
'url': best_format['defaultURL'],
|
||||
'ext': 'mp4',
|
||||
|
|
Loading…
Reference in a new issue