[kaltura] Support playlists (#4986)

Authored by: jwoglom, pukkandan
pull/4406/head
James Woglom 2 years ago committed by GitHub
parent 62b8dac490
commit 58fb927ebd
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

@ -15,13 +15,14 @@ from ..utils import (
unsmuggle_url, unsmuggle_url,
smuggle_url, smuggle_url,
traverse_obj, traverse_obj,
remove_start
) )
class KalturaIE(InfoExtractor): class KalturaIE(InfoExtractor):
_VALID_URL = r'''(?x) _VALID_URL = r'''(?x)
(?: (?:
kaltura:(?P<partner_id>\d+):(?P<id>[0-9a-z_]+)| kaltura:(?P<partner_id>\w+):(?P<id>\w+)(?::(?P<player_type>\w+))?|
https?:// https?://
(:?(?:www|cdnapi(?:sec)?)\.)?kaltura\.com(?::\d+)?/ (:?(?:www|cdnapi(?:sec)?)\.)?kaltura\.com(?::\d+)?/
(?: (?:
@ -56,6 +57,7 @@ class KalturaIE(InfoExtractor):
'thumbnail': 're:^https?://.*/thumbnail/.*', 'thumbnail': 're:^https?://.*/thumbnail/.*',
'timestamp': int, 'timestamp': int,
}, },
'skip': 'The access to this service is forbidden since the specified partner is blocked'
}, },
{ {
'url': 'http://www.kaltura.com/index.php/kwidget/cache_st/1300318621/wid/_269692/uiconf_id/3873291/entry_id/1_1jc2y3e4', 'url': 'http://www.kaltura.com/index.php/kwidget/cache_st/1300318621/wid/_269692/uiconf_id/3873291/entry_id/1_1jc2y3e4',
@ -108,6 +110,80 @@ class KalturaIE(InfoExtractor):
# unavailable source format # unavailable source format
'url': 'kaltura:513551:1_66x4rg7o', 'url': 'kaltura:513551:1_66x4rg7o',
'only_matching': True, 'only_matching': True,
},
{
# html5lib URL using kwidget player
'url': 'https://cdnapisec.kaltura.com/html5/html5lib/v2.46/mwEmbedFrame.php/p/691292/uiconf_id/20499062/entry_id/0_c076mna6?wid=_691292&iframeembed=true&playerId=kaltura_player_1420508608&entry_id=0_c076mna6&flashvars%5BakamaiHD.loadingPolicy%5D=preInitialize&flashvars%5BakamaiHD.asyncInit%5D=true&flashvars%5BstreamerType%5D=hdnetwork',
'info_dict': {
'id': '0_c076mna6',
'ext': 'mp4',
'title': 'md5:4883e7acbcbf42583a2dddc97dee4855',
'duration': 3608,
'uploader_id': 'commons@swinburne.edu.au',
'timestamp': 1408086874,
'view_count': int,
'upload_date': '20140815',
'thumbnail': 'http://cfvod.kaltura.com/p/691292/sp/69129200/thumbnail/entry_id/0_c076mna6/version/100022',
}
},
{
# html5lib playlist URL using kwidget player
'url': 'https://cdnapisec.kaltura.com/html5/html5lib/v2.89/mwEmbedFrame.php/p/2019031/uiconf_id/40436601?wid=1_4j3m32cv&iframeembed=true&playerId=kaltura_player_&flashvars[playlistAPI.kpl0Id]=1_jovey5nu&flashvars[ks]=&&flashvars[imageDefaultDuration]=30&flashvars[localizationCode]=en&flashvars[leadWithHTML5]=true&flashvars[forceMobileHTML5]=true&flashvars[nextPrevBtn.plugin]=true&flashvars[hotspots.plugin]=true&flashvars[sideBarContainer.plugin]=true&flashvars[sideBarContainer.position]=left&flashvars[sideBarContainer.clickToClose]=true&flashvars[chapters.plugin]=true&flashvars[chapters.layout]=vertical&flashvars[chapters.thumbnailRotator]=false&flashvars[streamSelector.plugin]=true&flashvars[EmbedPlayer.SpinnerTarget]=videoHolder&flashvars[dualScreen.plugin]=true&flashvars[playlistAPI.playlistUrl]=https://canvasgatechtest.kaf.kaltura.com/playlist/details/{playlistAPI.kpl0Id}/categoryid/126428551',
'info_dict': {
'id': '1_jovey5nu',
'title': '00-00 Introduction'
},
'playlist': [
{
'info_dict': {
'id': '1_b1y5hlvx',
'ext': 'mp4',
'title': 'CS7646_00-00 Introductio_Introduction',
'duration': 91,
'thumbnail': 'http://cfvod.kaltura.com/p/2019031/sp/201903100/thumbnail/entry_id/1_b1y5hlvx/version/100001',
'view_count': int,
'timestamp': 1533154447,
'upload_date': '20180801',
'uploader_id': 'djoyner3',
}
}, {
'info_dict': {
'id': '1_jfb7mdpn',
'ext': 'mp4',
'title': 'CS7646_00-00 Introductio_Three parts to the course',
'duration': 63,
'thumbnail': 'http://cfvod.kaltura.com/p/2019031/sp/201903100/thumbnail/entry_id/1_jfb7mdpn/version/100001',
'view_count': int,
'timestamp': 1533154489,
'upload_date': '20180801',
'uploader_id': 'djoyner3',
}
}, {
'info_dict': {
'id': '1_8xflxdp7',
'ext': 'mp4',
'title': 'CS7646_00-00 Introductio_Textbooks',
'duration': 37,
'thumbnail': 'http://cfvod.kaltura.com/p/2019031/sp/201903100/thumbnail/entry_id/1_8xflxdp7/version/100001',
'view_count': int,
'timestamp': 1533154512,
'upload_date': '20180801',
'uploader_id': 'djoyner3',
}
}, {
'info_dict': {
'id': '1_3hqew8kn',
'ext': 'mp4',
'title': 'CS7646_00-00 Introductio_Prerequisites',
'duration': 49,
'thumbnail': 'http://cfvod.kaltura.com/p/2019031/sp/201903100/thumbnail/entry_id/1_3hqew8kn/version/100001',
'view_count': int,
'timestamp': 1533154536,
'upload_date': '20180801',
'uploader_id': 'djoyner3',
}
}
]
} }
] ]
@ -187,7 +263,14 @@ class KalturaIE(InfoExtractor):
return data return data
def _get_video_info(self, video_id, partner_id, service_url=None): def _get_video_info(self, video_id, partner_id, service_url=None, player_type='html5'):
assert player_type in ('html5', 'kwidget')
if player_type == 'kwidget':
return self._get_video_info_kwidget(video_id, partner_id, service_url)
return self._get_video_info_html5(video_id, partner_id, service_url)
def _get_video_info_html5(self, video_id, partner_id, service_url=None):
actions = [ actions = [
{ {
'apiVersion': '3.3.0', 'apiVersion': '3.3.0',
@ -200,8 +283,9 @@ class KalturaIE(InfoExtractor):
'expiry': 86400, 'expiry': 86400,
'service': 'session', 'service': 'session',
'action': 'startWidgetSession', 'action': 'startWidgetSession',
'widgetId': '_%s' % partner_id, 'widgetId': self._build_widget_id(partner_id),
}, },
# info
{ {
'action': 'list', 'action': 'list',
'filter': {'redirectFromEntryId': video_id}, 'filter': {'redirectFromEntryId': video_id},
@ -212,12 +296,14 @@ class KalturaIE(InfoExtractor):
'fields': 'createdAt,dataUrl,duration,name,plays,thumbnailUrl,userId', 'fields': 'createdAt,dataUrl,duration,name,plays,thumbnailUrl,userId',
}, },
}, },
# flavor_assets
{ {
'action': 'getbyentryid', 'action': 'getbyentryid',
'entryId': video_id, 'entryId': video_id,
'service': 'flavorAsset', 'service': 'flavorAsset',
'ks': '{1:result:ks}', 'ks': '{1:result:ks}',
}, },
# captions
{ {
'action': 'list', 'action': 'list',
'filter:entryIdEqual': video_id, 'filter:entryIdEqual': video_id,
@ -226,17 +312,85 @@ class KalturaIE(InfoExtractor):
}, },
] ]
return self._kaltura_api_call( return self._kaltura_api_call(
video_id, actions, service_url, note='Downloading video info JSON') video_id, actions, service_url, note='Downloading video info JSON (Kaltura html5 player)')
def _get_video_info_kwidget(self, video_id, partner_id, service_url=None):
actions = [
{
'service': 'multirequest',
'apiVersion': '3.1',
'expiry': 86400,
'clientTag': 'kwidget:v2.89',
'format': 1, # JSON, 2 = XML, 3 = PHP
'ignoreNull': 1,
'action': 'null',
},
# header
{
'expiry': 86400,
'service': 'session',
'action': 'startWidgetSession',
'widgetId': self._build_widget_id(partner_id),
},
# (empty)
{
'expiry': 86400,
'service': 'session',
'action': 'startwidgetsession',
'widgetId': self._build_widget_id(partner_id),
'format': 9,
'apiVersion': '3.1',
'clientTag': 'kwidget:v2.89',
'ignoreNull': 1,
'ks': '{1:result:ks}'
},
# info
{
'action': 'list',
'filter': {'redirectFromEntryId': video_id},
'service': 'baseentry',
'ks': '{1:result:ks}',
'responseProfile': {
'type': 1,
'fields': 'createdAt,dataUrl,duration,name,plays,thumbnailUrl,userId',
},
},
# flavor_assets
{
'action': 'getbyentryid',
'entryId': video_id,
'service': 'flavorAsset',
'ks': '{1:result:ks}',
},
# captions
{
'action': 'list',
'filter:entryIdEqual': video_id,
'service': 'caption_captionasset',
'ks': '{1:result:ks}',
},
]
# second object (representing the second start widget session) is None
header, _, _info, flavor_assets, captions = self._kaltura_api_call(
video_id, actions, service_url, note='Downloading video info JSON (Kaltura kwidget player)')
info = _info['objects'][0]
return header, info, flavor_assets, captions
def _build_widget_id(self, partner_id):
return partner_id if '_' in partner_id else f'_{partner_id}'
IFRAME_PACKAGE_DATA_REGEX = r'window\.kalturaIframePackageData\s*='
def _real_extract(self, url): def _real_extract(self, url):
url, smuggled_data = unsmuggle_url(url, {}) url, smuggled_data = unsmuggle_url(url, {})
mobj = self._match_valid_url(url) mobj = self._match_valid_url(url)
partner_id, entry_id = mobj.group('partner_id', 'id') partner_id, entry_id, player_type = mobj.group('partner_id', 'id', 'player_type')
ks = None ks, captions = None, None
captions = None if not player_type:
player_type = 'kwidget' if 'html5lib/v2' in url else 'html5'
if partner_id and entry_id: if partner_id and entry_id:
_, info, flavor_assets, captions = self._get_video_info(entry_id, partner_id, smuggled_data.get('service_url')) _, info, flavor_assets, captions = self._get_video_info(entry_id, partner_id, smuggled_data.get('service_url'), player_type=player_type)
else: else:
path, query = mobj.group('path', 'query') path, query = mobj.group('path', 'query')
if not path and not query: if not path and not query:
@ -248,7 +402,7 @@ class KalturaIE(InfoExtractor):
splitted_path = path.split('/') splitted_path = path.split('/')
params.update(dict((zip(splitted_path[::2], [[v] for v in splitted_path[1::2]])))) params.update(dict((zip(splitted_path[::2], [[v] for v in splitted_path[1::2]]))))
if 'wid' in params: if 'wid' in params:
partner_id = params['wid'][0][1:] partner_id = remove_start(params['wid'][0], '_')
elif 'p' in params: elif 'p' in params:
partner_id = params['p'][0] partner_id = params['p'][0]
elif 'partner_id' in params: elif 'partner_id' in params:
@ -257,14 +411,13 @@ class KalturaIE(InfoExtractor):
raise ExtractorError('Invalid URL', expected=True) raise ExtractorError('Invalid URL', expected=True)
if 'entry_id' in params: if 'entry_id' in params:
entry_id = params['entry_id'][0] entry_id = params['entry_id'][0]
_, info, flavor_assets, captions = self._get_video_info(entry_id, partner_id) _, info, flavor_assets, captions = self._get_video_info(entry_id, partner_id, player_type=player_type)
elif 'uiconf_id' in params and 'flashvars[referenceId]' in params: elif 'uiconf_id' in params and 'flashvars[referenceId]' in params:
reference_id = params['flashvars[referenceId]'][0] reference_id = params['flashvars[referenceId]'][0]
webpage = self._download_webpage(url, reference_id) webpage = self._download_webpage(url, reference_id)
entry_data = self._parse_json(self._search_regex( entry_data = self._search_json(
r'window\.kalturaIframePackageData\s*=\s*({.*});', self.IFRAME_PACKAGE_DATA_REGEX, webpage,
webpage, 'kalturaIframePackageData'), 'kalturaIframePackageData', reference_id)['entryResult']
reference_id)['entryResult']
info, flavor_assets = entry_data['meta'], entry_data['contextData']['flavorAssets'] info, flavor_assets = entry_data['meta'], entry_data['contextData']['flavorAssets']
entry_id = info['id'] entry_id = info['id']
# Unfortunately, data returned in kalturaIframePackageData lacks # Unfortunately, data returned in kalturaIframePackageData lacks
@ -272,16 +425,29 @@ class KalturaIE(InfoExtractor):
# regular approach since we now know the entry_id # regular approach since we now know the entry_id
try: try:
_, info, flavor_assets, captions = self._get_video_info( _, info, flavor_assets, captions = self._get_video_info(
entry_id, partner_id) entry_id, partner_id, player_type=player_type)
except ExtractorError: except ExtractorError:
# Regular scenario failed but we already have everything # Regular scenario failed but we already have everything
# extracted apart from captions and can process at least # extracted apart from captions and can process at least
# with this # with this
pass pass
elif 'uiconf_id' in params and 'flashvars[playlistAPI.kpl0Id]' in params:
playlist_id = params['flashvars[playlistAPI.kpl0Id]'][0]
webpage = self._download_webpage(url, playlist_id)
playlist_data = self._search_json(
self.IFRAME_PACKAGE_DATA_REGEX, webpage,
'kalturaIframePackageData', playlist_id)['playlistResult']
return self.playlist_from_matches(
traverse_obj(playlist_data, (playlist_id, 'items', ..., 'id')),
playlist_id, traverse_obj(playlist_data, (playlist_id, 'name')),
ie=KalturaIE, getter=lambda x: f'kaltura:{partner_id}:{x}:{player_type}')
else: else:
raise ExtractorError('Invalid URL', expected=True) raise ExtractorError('Invalid URL', expected=True)
ks = params.get('flashvars[ks]', [None])[0] ks = params.get('flashvars[ks]', [None])[0]
return self._per_video_extract(smuggled_data, entry_id, info, ks, flavor_assets, captions)
def _per_video_extract(self, smuggled_data, entry_id, info, ks, flavor_assets, captions):
source_url = smuggled_data.get('source_url') source_url = smuggled_data.get('source_url')
if source_url: if source_url:
referrer = base64.b64encode( referrer = base64.b64encode(

Loading…
Cancel
Save