X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/9dc487f48b50767cf540fa36c3de2c386fd74c04..4e090bc3ceacc4e3cd464d12ea97700e3acad37d:/youtube_dl/extractor/periscope.py diff --git a/youtube_dl/extractor/periscope.py b/youtube_dl/extractor/periscope.py index c23b314..0e36230 100644 --- a/youtube_dl/extractor/periscope.py +++ b/youtube_dl/extractor/periscope.py @@ -1,6 +1,8 @@ # coding: utf-8 from __future__ import unicode_literals +import re + from .common import InfoExtractor from ..utils import ( parse_iso8601, @@ -8,7 +10,14 @@ from ..utils import ( ) -class PeriscopeIE(InfoExtractor): +class PeriscopeBaseIE(InfoExtractor): + def _call_api(self, method, query, item_id): + return self._download_json( + 'https://api.periscope.tv/api/v2/%s' % method, + item_id, query=query) + + +class PeriscopeIE(PeriscopeBaseIE): IE_DESC = 'Periscope' IE_NAME = 'periscope' _VALID_URL = r'https?://(?:www\.)?periscope\.tv/[^/]+/(?P[^/?#]+)' @@ -34,14 +43,18 @@ class PeriscopeIE(InfoExtractor): 'only_matching': True, }] - def _call_api(self, method, value): - return self._download_json( - 'https://api.periscope.tv/api/v2/%s?broadcast_id=%s' % (method, value), value) + @staticmethod + def _extract_url(webpage): + mobj = re.search( + r']+src=([\'"])(?P(?:https?:)?//(?:www\.)?periscope\.tv/(?:(?!\1).)+)\1', webpage) + if mobj: + return mobj.group('url') def _real_extract(self, url): token = self._match_id(url) - broadcast_data = self._call_api('getBroadcastPublic', token) + broadcast_data = self._call_api( + 'getBroadcastPublic', {'broadcast_id': token}, token) broadcast = broadcast_data['broadcast'] status = broadcast['status'] @@ -61,7 +74,8 @@ class PeriscopeIE(InfoExtractor): 'url': broadcast[image], } for image in ('image_url', 'image_url_small') if broadcast.get(image)] - stream = self._call_api('getAccessPublic', token) + stream = self._call_api( + 'getAccessPublic', {'broadcast_id': token}, token) formats = [] for format_id in ('replay', 'rtmp', 'hls', 'https_hls'): @@ -73,7 +87,7 @@ class PeriscopeIE(InfoExtractor): 'ext': 'flv' if format_id == 'rtmp' else 'mp4', } if format_id != 'rtmp': - f['protocol'] = 'm3u8_native' if state == 'ended' else 'm3u8' + f['protocol'] = 'm3u8_native' if state in ('ended', 'timed_out') else 'm3u8' formats.append(f) self._sort_formats(formats) @@ -88,8 +102,8 @@ class PeriscopeIE(InfoExtractor): } -class PeriscopeUserIE(InfoExtractor): - _VALID_URL = r'https?://www\.periscope\.tv/(?P[^/]+)/?$' +class PeriscopeUserIE(PeriscopeBaseIE): + _VALID_URL = r'https?://(?:www\.)?periscope\.tv/(?P[^/]+)/?$' IE_DESC = 'Periscope user videos' IE_NAME = 'periscope:user' @@ -106,23 +120,34 @@ class PeriscopeUserIE(InfoExtractor): } def _real_extract(self, url): - user_id = self._match_id(url) + user_name = self._match_id(url) - webpage = self._download_webpage(url, user_id) + webpage = self._download_webpage(url, user_name) data_store = self._parse_json( unescapeHTML(self._search_regex( r'data-store=(["\'])(?P.+?)\1', webpage, 'data store', default='{}', group='data')), - user_id) + user_name) + + user = list(data_store['UserCache']['users'].values())[0]['user'] + user_id = user['id'] + session_id = data_store['SessionToken']['public']['broadcastHistory']['token']['session_id'] + + broadcasts = self._call_api( + 'getUserBroadcastsPublic', + {'user_id': user_id, 'session_id': session_id}, + user_name)['broadcasts'] + + broadcast_ids = [ + broadcast['id'] for broadcast in broadcasts if broadcast.get('id')] - user = data_store.get('User', {}).get('user', {}) - title = user.get('display_name') or user.get('username') + title = user.get('display_name') or user.get('username') or user_name description = user.get('description') entries = [ self.url_result( - 'https://www.periscope.tv/%s/%s' % (user_id, broadcast['id'])) - for broadcast in data_store.get('UserBroadcastHistory', {}).get('broadcasts', [])] + 'https://www.periscope.tv/%s/%s' % (user_name, broadcast_id)) + for broadcast_id in broadcast_ids] return self.playlist_result(entries, user_id, title, description)