Fix playlist order

The "index" attribute does not seem to correspond to the playlist
order (at least, not exactly).

An OrderedDict would really help here.  If support for Python 2.6 is
ever dropped...  :)

test/test_youtube_lists.py

@@ -58,12 +58,21 @@ class TestYoutubeLists(unittest.TestCase):
         # Save generator output
         playlist = [v for v in result['entries']]
 
+        # Find videos in playlist
         for video in videos:
             matching_videos = [v for v in playlist if v['id'] == video['id']]
 
             self.assertEqual(len(matching_videos), 1)
             self.assertEqual(matching_videos[0]['title'], video['title'])
 
+        # TODO: It would be good to check that the videos are returned
+        # in the correct order (not necessarily back-to-back), which,
+        # of course, requires creating the test data in the correct
+        # order. The reason is that simple mistakes (like forgetting
+        # that dicts don't keep insertion order) can result in the
+        # order being wrong. This could be in a separate test, or it
+        # could go here.
+
     def test_youtube_playlist_noplaylist(self):
         dl = FakeYDL()
         dl.params['noplaylist'] = True

youtube_dl/extractor/youtube.py

@@ -1586,11 +1586,19 @@ class YoutubePlaylistIE(YoutubeBaseInfoExtractor):
                 matches = self._VIDEO_RE.finditer(content_html)
 
                 # Get videos from current page. Using OrderedDict to
-                # avoid duplicates would make this much simpler.
+                # avoid duplicates would make this much
+                # simpler. Lacking that, we store the order of the
+                # videos as video_num so we can sort the dict, keeping
+                # the order of the playlist. We have to avoid
+                # duplicates because it seems that every video in the
+                # playlist shows up in the HTML/JSON twice: once
+                # without a title, and once with a title. Maybe using
+                # something like bs4 instead of regexps would also be a
+                # good idea.
                 new_videos = {}
+                num = 0
                 for m in matches:
-                    video_index = m.group('index')
-                    if video_index == '0':
+                    if m.group('index') == '0':
                         # Ignore link with index 0
                         continue
 
@@ -1602,21 +1610,18 @@ class YoutubePlaylistIE(YoutubeBaseInfoExtractor):
                         video_title = None
 
                     if video_id in new_videos:
-                        # Duplicate video
-
+                        # Video is already in dict
                         if video_title and not new_videos[video_id]['title']:
                             # Set missing title
                             new_videos[video_id]['title'] = video_title
-
-                        new_videos[video_id]['index'] = video_index
-
                     else:
-                        # New video
-                        new_videos[video_id] = {'index': int(video_index),
-                                                'title': video_title}
+                        # Video not in dict
+                        new_videos[video_id] = {'num': num, 'title': video_title}
 
-                # Sort videos by index
-                new_videos = sorted(new_videos.iteritems(), key=lambda v: v[1]['index'])
+                    num += 1
+
+                # Sort videos by playlist order
+                new_videos = sorted(new_videos.iteritems(), key=lambda v: v[1]['num'])
 
                 # Yield current list of videos
                 for video in new_videos: