From f4c8bbcfc2c28d434a5696c88fe001ded68155ca Mon Sep 17 00:00:00 2001 From: =?utf8?q?Jaime=20Marqui=CC=81nez=20Ferra=CC=81ndiz?= Date: Thu, 20 Jun 2013 20:51:20 +0200 Subject: [PATCH] TEDIE: download the best quality video and use the new _search_regex functions Also extracts the description. --- test/tests.json | 4 ++-- youtube_dl/InfoExtractors.py | 44 ++++++++++++++++-------------------- 2 files changed, 22 insertions(+), 26 deletions(-) diff --git a/test/tests.json b/test/tests.json index 418897731..3e0db297d 100644 --- a/test/tests.json +++ b/test/tests.json @@ -363,10 +363,10 @@ "name": "TED", "url": "http://www.ted.com/talks/dan_dennett_on_our_consciousness.html", "file": "102.mp4", - "md5": "7bc087e71d16f18f9b8ab9fa62a8a031", + "md5": "8cd9dfa41ee000ce658fd48fb5d89a61", "info_dict": { "title": "Dan Dennett: The illusion of consciousness", - "thumbnail": "http://images.ted.com/images/ted/488_389x292.jpg" + "description": "md5:c6fa72e6eedbd938c9caf6b2702f5922" } }, { diff --git a/youtube_dl/InfoExtractors.py b/youtube_dl/InfoExtractors.py index d9c555643..17e0f8323 100755 --- a/youtube_dl/InfoExtractors.py +++ b/youtube_dl/InfoExtractors.py @@ -3792,10 +3792,6 @@ class TEDIE(InfoExtractor): self.to_screen(u'Getting info of playlist %s: "%s"' % (playlist_id,name)) return [self._playlist_videos_info(url,name,playlist_id)] - def _talk_video_link(self,mediaSlug): - '''Returns the video link for that mediaSlug''' - return 'http://download.ted.com/talks/%s.mp4' % mediaSlug - def _playlist_videos_info(self,url,name,playlist_id=0): '''Returns the videos of the playlist''' video_RE=r''' @@ -3808,9 +3804,8 @@ class TEDIE(InfoExtractor): m_videos=re.finditer(video_RE,webpage,re.VERBOSE) m_names=re.finditer(video_name_RE,webpage) - playlist_RE = r'div class="headline">(\s*?)

(\s*?)(?P.*?)' - m_playlist = re.search(playlist_RE, webpage) - playlist_title = m_playlist.group('playlist_title') + playlist_title = self._html_search_regex(r'div class="headline">\s*?

\s*?(.*?)', + webpage, 'playlist title') playlist_entries = [] for m_video, m_name in zip(m_videos,m_names): @@ -3821,27 +3816,28 @@ class TEDIE(InfoExtractor): def _talk_info(self, url, video_id=0): """Return the video for the talk in the url""" - m=re.match(self._VALID_URL, url,re.VERBOSE) - videoName=m.group('name') - webpage=self._download_webpage(url, video_id, 'Downloading \"%s\" page' % videoName) + m = re.match(self._VALID_URL, url,re.VERBOSE) + video_name = m.group('name') + webpage = self._download_webpage(url, video_id, 'Downloading \"%s\" page' % video_name) + self.report_extraction(video_name) # If the url includes the language we get the title translated - title_RE=r'(?P.*)</span>' - title=re.search(title_RE, webpage).group('title') - info_RE=r'''<script\ type="text/javascript">var\ talkDetails\ =(.*?) - "id":(?P<videoID>[\d]+).*? - "mediaSlug":"(?P<mediaSlug>[\w\d]+?)"''' - thumb_RE=r'</span>[\s.]*</div>[\s.]*<img src="(?P<thumbnail>.*?)"' - thumb_match=re.search(thumb_RE,webpage) - info_match=re.search(info_RE,webpage,re.VERBOSE) - video_id=info_match.group('videoID') - mediaSlug=info_match.group('mediaSlug') - video_url=self._talk_video_link(mediaSlug) + title = self._html_search_regex(r'<span id="altHeadline" >(?P<title>.*)</span>', + webpage, 'title') + json_data = self._search_regex(r'<script.*?>var talkDetails = ({.*?})</script>', + webpage, 'json data') + info = json.loads(json_data) + desc = self._html_search_regex(r'<div class="talk-intro">.*?<p.*?>(.*?)</p>', + webpage, 'description', flags = re.DOTALL) + + thumbnail = self._search_regex(r'</span>[\s.]*</div>[\s.]*<img src="(.*?)"', + webpage, 'thumbnail') info = { - 'id': video_id, - 'url': video_url, + 'id': info['id'], + 'url': info['htmlStreams'][-1]['file'], 'ext': 'mp4', 'title': title, - 'thumbnail': thumb_match.group('thumbnail') + 'thumbnail': thumbnail, + 'description': desc, } return info -- 2.39.5