X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/e76f531201cd41dfc0ce00be28bcc5c575c7acc5..c64c42cb1f9487398f8b91a51266ff6e79115b90:/youtube_dl/extractor/channel9.py diff --git a/youtube_dl/extractor/channel9.py b/youtube_dl/extractor/channel9.py index c74553d..34d4e61 100644 --- a/youtube_dl/extractor/channel9.py +++ b/youtube_dl/extractor/channel9.py @@ -20,54 +20,64 @@ class Channel9IE(InfoExtractor): ''' IE_DESC = 'Channel 9' IE_NAME = 'channel9' - _VALID_URL = r'https?://(?:www\.)?channel9\.msdn\.com/(?P.+)/?' - - _TESTS = [ - { - 'url': 'http://channel9.msdn.com/Events/TechEd/Australia/2013/KOS002', - 'md5': 'bbd75296ba47916b754e73c3a4bbdf10', - 'info_dict': { - 'id': 'Events/TechEd/Australia/2013/KOS002', - 'ext': 'mp4', - 'title': 'Developer Kick-Off Session: Stuff We Love', - 'description': 'md5:c08d72240b7c87fcecafe2692f80e35f', - 'duration': 4576, - 'thumbnail': 're:http://.*\.jpg', - 'session_code': 'KOS002', - 'session_day': 'Day 1', - 'session_room': 'Arena 1A', - 'session_speakers': ['Ed Blankenship', 'Andrew Coates', 'Brady Gaster', 'Patrick Klug', 'Mads Kristensen'], - }, + _VALID_URL = r'https?://(?:www\.)?channel9\.msdn\.com/(?P.+?)(?P/RSS)?/?(?:[?#&]|$)' + + _TESTS = [{ + 'url': 'http://channel9.msdn.com/Events/TechEd/Australia/2013/KOS002', + 'md5': 'bbd75296ba47916b754e73c3a4bbdf10', + 'info_dict': { + 'id': 'Events/TechEd/Australia/2013/KOS002', + 'ext': 'mp4', + 'title': 'Developer Kick-Off Session: Stuff We Love', + 'description': 'md5:c08d72240b7c87fcecafe2692f80e35f', + 'duration': 4576, + 'thumbnail': 're:http://.*\.jpg', + 'session_code': 'KOS002', + 'session_day': 'Day 1', + 'session_room': 'Arena 1A', + 'session_speakers': ['Ed Blankenship', 'Andrew Coates', 'Brady Gaster', 'Patrick Klug', + 'Mads Kristensen'], }, - { - 'url': 'http://channel9.msdn.com/posts/Self-service-BI-with-Power-BI-nuclear-testing', - 'md5': 'b43ee4529d111bc37ba7ee4f34813e68', - 'info_dict': { - 'id': 'posts/Self-service-BI-with-Power-BI-nuclear-testing', - 'ext': 'mp4', - 'title': 'Self-service BI with Power BI - nuclear testing', - 'description': 'md5:d1e6ecaafa7fb52a2cacdf9599829f5b', - 'duration': 1540, - 'thumbnail': 're:http://.*\.jpg', - 'authors': ['Mike Wilmot'], - }, + }, { + 'url': 'http://channel9.msdn.com/posts/Self-service-BI-with-Power-BI-nuclear-testing', + 'md5': 'b43ee4529d111bc37ba7ee4f34813e68', + 'info_dict': { + 'id': 'posts/Self-service-BI-with-Power-BI-nuclear-testing', + 'ext': 'mp4', + 'title': 'Self-service BI with Power BI - nuclear testing', + 'description': 'md5:d1e6ecaafa7fb52a2cacdf9599829f5b', + 'duration': 1540, + 'thumbnail': 're:http://.*\.jpg', + 'authors': ['Mike Wilmot'], }, - { - # low quality mp4 is best - 'url': 'https://channel9.msdn.com/Events/CPP/CppCon-2015/Ranges-for-the-Standard-Library', - 'info_dict': { - 'id': 'Events/CPP/CppCon-2015/Ranges-for-the-Standard-Library', - 'ext': 'mp4', - 'title': 'Ranges for the Standard Library', - 'description': 'md5:2e6b4917677af3728c5f6d63784c4c5d', - 'duration': 5646, - 'thumbnail': 're:http://.*\.jpg', - }, - 'params': { - 'skip_download': True, - }, - } - ] + }, { + # low quality mp4 is best + 'url': 'https://channel9.msdn.com/Events/CPP/CppCon-2015/Ranges-for-the-Standard-Library', + 'info_dict': { + 'id': 'Events/CPP/CppCon-2015/Ranges-for-the-Standard-Library', + 'ext': 'mp4', + 'title': 'Ranges for the Standard Library', + 'description': 'md5:2e6b4917677af3728c5f6d63784c4c5d', + 'duration': 5646, + 'thumbnail': 're:http://.*\.jpg', + }, + 'params': { + 'skip_download': True, + }, + }, { + 'url': 'https://channel9.msdn.com/Niners/Splendid22/Queue/76acff796e8f411184b008028e0d492b/RSS', + 'info_dict': { + 'id': 'Niners/Splendid22/Queue/76acff796e8f411184b008028e0d492b', + 'title': 'Channel 9', + }, + 'playlist_count': 2, + }, { + 'url': 'https://channel9.msdn.com/Events/DEVintersection/DEVintersection-2016/RSS', + 'only_matching': True, + }, { + 'url': 'https://channel9.msdn.com/Events/Speakers/scott-hanselman/RSS?UrlSafeName=scott-hanselman', + 'only_matching': True, + }] _RSS_URL = 'http://channel9.msdn.com/%s/RSS' @@ -254,22 +264,30 @@ class Channel9IE(InfoExtractor): return self.playlist_result(contents) - def _extract_list(self, content_path): - rss = self._download_xml(self._RSS_URL % content_path, content_path, 'Downloading RSS') + def _extract_list(self, video_id, rss_url=None): + if not rss_url: + rss_url = self._RSS_URL % video_id + rss = self._download_xml(rss_url, video_id, 'Downloading RSS') entries = [self.url_result(session_url.text, 'Channel9') for session_url in rss.findall('./channel/item/link')] title_text = rss.find('./channel/title').text - return self.playlist_result(entries, content_path, title_text) + return self.playlist_result(entries, video_id, title_text) def _real_extract(self, url): mobj = re.match(self._VALID_URL, url) content_path = mobj.group('contentpath') + rss = mobj.group('rss') + + if rss: + return self._extract_list(content_path, url) - webpage = self._download_webpage(url, content_path, 'Downloading web page') + webpage = self._download_webpage( + url, content_path, 'Downloading web page') - page_type_m = re.search(r'', webpage) - if page_type_m is not None: - page_type = page_type_m.group('pagetype') + page_type = self._search_regex( + r']+name=(["\'])WT\.entryid\1[^>]+content=(["\'])(?P[^:]+).+?\2', + webpage, 'page type', default=None, group='pagetype') + if page_type: if page_type == 'Entry': # Any 'item'-like page, may contain downloadable content return self._extract_entry_item(webpage, content_path) elif page_type == 'Session': # Event session page, may contain downloadable content @@ -278,6 +296,5 @@ class Channel9IE(InfoExtractor): return self._extract_list(content_path) else: raise ExtractorError('Unexpected WT.entryid %s' % page_type, expected=True) - else: # Assuming list return self._extract_list(content_path)