diff options
-rw-r--r-- | test/test_youtube_lists.py | 9 | ||||
-rw-r--r-- | youtube_dl/extractor/youtube.py | 24 |
2 files changed, 31 insertions, 2 deletions
diff --git a/test/test_youtube_lists.py b/test/test_youtube_lists.py index 8fd073f31..95f07d129 100644 --- a/test/test_youtube_lists.py +++ b/test/test_youtube_lists.py @@ -107,5 +107,14 @@ class TestYoutubeLists(unittest.TestCase): result = ie.extract('http://www.youtube.com/show/airdisasters') self.assertTrue(len(result) >= 3) + def test_youtube_mix(self): + dl = FakeYDL() + ie = YoutubePlaylistIE(dl) + result = ie.extract('http://www.youtube.com/watch?v=lLJf9qJHR3E&list=RDrjFaenf1T-Y') + entries = result['entries'] + self.assertTrue(len(entries) >= 20) + original_video = entries[0] + self.assertEqual(original_video['id'], 'rjFaenf1T-Y') + if __name__ == '__main__': unittest.main() diff --git a/youtube_dl/extractor/youtube.py b/youtube_dl/extractor/youtube.py index a76a9071a..9ef5fecce 100644 --- a/youtube_dl/extractor/youtube.py +++ b/youtube_dl/extractor/youtube.py @@ -28,6 +28,7 @@ from ..utils import ( clean_html, get_cachedir, get_element_by_id, + get_element_by_attribute, ExtractorError, unescapeHTML, unified_strdate, @@ -1537,6 +1538,22 @@ class YoutubePlaylistIE(YoutubeBaseInfoExtractor): def _real_initialize(self): self._login() + def _ids_to_results(self, ids): + return [self.url_result(vid_id, 'Youtube', video_id=vid_id) + for vid_id in ids] + + def _extract_mix(self, playlist_id): + # The mixes are generated from a a single video + # the id of the playlist is just 'RD' + video_id + url = 'https://youtube.com/watch?v=%s&list=%s' % (playlist_id[2:], playlist_id) + webpage = self._download_webpage(url, playlist_id, u'Downloading Youtube mix') + title = clean_html(get_element_by_attribute('class', 'title long-title', webpage)) + video_re = r'data-index="\d+".*?href="/watch\?v=([0-9A-Za-z_-]{11})&[^"]*?list=%s' % re.escape(playlist_id) + ids = orderedSet(re.findall(video_re, webpage)) + url_results = self._ids_to_results(ids) + + return self.playlist_result(url_results, playlist_id, title) + def _real_extract(self, url): # Extract playlist id mobj = re.match(self._VALID_URL, url, re.VERBOSE) @@ -1554,6 +1571,10 @@ class YoutubePlaylistIE(YoutubeBaseInfoExtractor): else: self.to_screen(u'Downloading playlist PL%s - add --no-playlist to just download video %s' % (playlist_id, video_id)) + if len(playlist_id) == 13: # 'RD' + 11 characters for the video id + # Mixes require a custom extraction process + return self._extract_mix(playlist_id) + # Extract the video ids from the playlist pages ids = [] @@ -1571,8 +1592,7 @@ class YoutubePlaylistIE(YoutubeBaseInfoExtractor): playlist_title = self._og_search_title(page) - url_results = [self.url_result(vid_id, 'Youtube', video_id=vid_id) - for vid_id in ids] + url_results = self._ids_to_results(ids) return self.playlist_result(url_results, playlist_id, playlist_title) |