X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/5920ef2b4969021b7f83d154b325036d9b598877..89bb9e23b34f43c6563524917f7dffb9841a972e:/youtube_dl/extractor/ciscolive.py diff --git a/youtube_dl/extractor/ciscolive.py b/youtube_dl/extractor/ciscolive.py new file mode 100644 index 0000000..c99b6ee --- /dev/null +++ b/youtube_dl/extractor/ciscolive.py @@ -0,0 +1,142 @@ +# coding: utf-8 +from __future__ import unicode_literals + +import itertools + +from .common import InfoExtractor +from ..compat import ( + compat_parse_qs, + compat_urllib_parse_urlparse, +) +from ..utils import ( + clean_html, + float_or_none, + int_or_none, + try_get, + urlencode_postdata, +) + + +class CiscoLiveBaseIE(InfoExtractor): + # These appear to be constant across all Cisco Live presentations + # and are not tied to any user session or event + RAINFOCUS_API_URL = 'https://events.rainfocus.com/api/%s' + RAINFOCUS_API_PROFILE_ID = 'Na3vqYdAlJFSxhYTYQGuMbpafMqftalz' + RAINFOCUS_WIDGET_ID = 'n6l4Lo05R8fiy3RpUBm447dZN8uNWoye' + BRIGHTCOVE_URL_TEMPLATE = 'http://players.brightcove.net/5647924234001/SyK2FdqjM_default/index.html?videoId=%s' + + HEADERS = { + 'Origin': 'https://ciscolive.cisco.com', + 'rfApiProfileId': RAINFOCUS_API_PROFILE_ID, + 'rfWidgetId': RAINFOCUS_WIDGET_ID, + } + + def _call_api(self, ep, rf_id, query, referrer, note=None): + headers = self.HEADERS.copy() + headers['Referer'] = referrer + return self._download_json( + self.RAINFOCUS_API_URL % ep, rf_id, note=note, + data=urlencode_postdata(query), headers=headers) + + def _parse_rf_item(self, rf_item): + event_name = rf_item.get('eventName') + title = rf_item['title'] + description = clean_html(rf_item.get('abstract')) + presenter_name = try_get(rf_item, lambda x: x['participants'][0]['fullName']) + bc_id = rf_item['videos'][0]['url'] + bc_url = self.BRIGHTCOVE_URL_TEMPLATE % bc_id + duration = float_or_none(try_get(rf_item, lambda x: x['times'][0]['length'])) + location = try_get(rf_item, lambda x: x['times'][0]['room']) + + if duration: + duration = duration * 60 + + return { + '_type': 'url_transparent', + 'url': bc_url, + 'ie_key': 'BrightcoveNew', + 'title': title, + 'description': description, + 'duration': duration, + 'creator': presenter_name, + 'location': location, + 'series': event_name, + } + + +class CiscoLiveSessionIE(CiscoLiveBaseIE): + _VALID_URL = r'https?://ciscolive\.cisco\.com/on-demand-library/\??[^#]*#/session/(?P[^/?&]+)' + _TEST = { + 'url': 'https://ciscolive.cisco.com/on-demand-library/?#/session/1423353499155001FoSs', + 'md5': 'c98acf395ed9c9f766941c70f5352e22', + 'info_dict': { + 'id': '5803694304001', + 'ext': 'mp4', + 'title': '13 Smart Automations to Monitor Your Cisco IOS Network', + 'description': 'md5:ec4a436019e09a918dec17714803f7cc', + 'timestamp': 1530305395, + 'upload_date': '20180629', + 'uploader_id': '5647924234001', + 'location': '16B Mezz.', + }, + } + + def _real_extract(self, url): + rf_id = self._match_id(url) + rf_result = self._call_api('session', rf_id, {'id': rf_id}, url) + return self._parse_rf_item(rf_result['items'][0]) + + +class CiscoLiveSearchIE(CiscoLiveBaseIE): + _VALID_URL = r'https?://ciscolive\.cisco\.com/on-demand-library/' + _TESTS = [{ + 'url': 'https://ciscolive.cisco.com/on-demand-library/?search.event=ciscoliveus2018&search.technicallevel=scpsSkillLevel_aintroductory&search.focus=scpsSessionFocus_designAndDeployment#/', + 'info_dict': { + 'title': 'Search query', + }, + 'playlist_count': 5, + }, { + 'url': 'https://ciscolive.cisco.com/on-demand-library/?search.technology=scpsTechnology_applicationDevelopment&search.technology=scpsTechnology_ipv6&search.focus=scpsSessionFocus_troubleshootingTroubleshooting#/', + 'only_matching': True, + }] + + @classmethod + def suitable(cls, url): + return False if CiscoLiveSessionIE.suitable(url) else super(CiscoLiveSearchIE, cls).suitable(url) + + @staticmethod + def _check_bc_id_exists(rf_item): + return int_or_none(try_get(rf_item, lambda x: x['videos'][0]['url'])) is not None + + def _entries(self, query, url): + query['size'] = 50 + query['from'] = 0 + for page_num in itertools.count(1): + results = self._call_api( + 'search', None, query, url, + 'Downloading search JSON page %d' % page_num) + sl = try_get(results, lambda x: x['sectionList'][0], dict) + if sl: + results = sl + items = results.get('items') + if not items or not isinstance(items, list): + break + for item in items: + if not isinstance(item, dict): + continue + if not self._check_bc_id_exists(item): + continue + yield self._parse_rf_item(item) + size = int_or_none(results.get('size')) + if size is not None: + query['size'] = size + total = int_or_none(results.get('total')) + if total is not None and query['from'] + query['size'] > total: + break + query['from'] += query['size'] + + def _real_extract(self, url): + query = compat_parse_qs(compat_urllib_parse_urlparse(url).query) + query['type'] = 'session' + return self.playlist_result( + self._entries(query, url), playlist_title='Search query')