[youtube] Fix feed extraction

In order to extract videos from further pages, we need to get various variables that are in an argument to the `ytcfg.set` call in a script on the feed page.

youtube_dl/extractor/youtube.py

@@ -3309,6 +3309,8 @@ class YoutubeFeedsInfoExtractor(YoutubeBaseInfoExtractor):
     Subclasses must define the _FEED_NAME and _PLAYLIST_TITLE properties.
     """
     _LOGIN_REQUIRED = True
+    _FEED_DATA = r'window\[\"ytInitialData\"\]\W?=\W?({.*?});'
+    _YTCFG_DATA =  r"ytcfg.set\(({.*?})\)"
 
     @property
     def IE_NAME(self):
@@ -3317,37 +3319,91 @@ class YoutubeFeedsInfoExtractor(YoutubeBaseInfoExtractor):
     def _real_initialize(self):
         self._login()
 
+
+    def _find_videos_in_json(self, extracted):
+        videos = []
+        continuation = None        
+
+        def _real_find(obj):
+            if obj is None or isinstance(obj, str):
+                return
+
+            if type(obj) is list:
+                for elem in obj:
+                    _real_find(elem)
+
+            if type(obj) is dict:
+                if "videoId" in obj:
+                    videos.append(obj)
+                    return
+                
+                if "nextContinuationData" in obj:
+                    nonlocal continuation
+                    continuation = obj["nextContinuationData"]
+                    return 
+                
+                for _, o in obj.items():
+                    _real_find(o)
+
+        _real_find(extracted)
+
+        return videos, continuation
+        
     def _entries(self, page):
-        # The extraction process is the same as for playlists, but the regex
-        # for the video ids doesn't contain an index
-        ids = []
-        more_widget_html = content_html = page
+        info = []
+
+        yt_conf = self._parse_json(self._search_regex(self._YTCFG_DATA, page, 'ytcfg.set'), None)
+
+        search_response = self._parse_json(self._search_regex(self._FEED_DATA, page, 'ytInitialData'), None)
+
         for page_num in itertools.count(1):
-            matches = re.findall(r'href="\s*/watch\?v=([0-9A-Za-z_-]{11})', content_html)
+            video_info, continuation = self._find_videos_in_json(search_response)
 
-            # 'recommended' feed has infinite 'load more' and each new portion spins
-            # the same videos in (sometimes) slightly different order, so we'll check
-            # for unicity and break when portion has no new videos
-            new_ids = list(filter(lambda video_id: video_id not in ids, orderedSet(matches)))
-            if not new_ids:
+            new_info = [] 
+
+            for v in video_info:
+                v_id = try_get(v, lambda x: x['videoId'])
+                if not v_id:
+                    continue
+
+                have_video = False 
+                for old in info:
+                    if old['videoId'] == v_id:
+                        have_video = True
                         break
 
-            ids.extend(new_ids)
+                if not have_video:
+                    new_info.append(v)
 
-            for entry in self._ids_to_results(new_ids):
-                yield entry
-
-            mobj = re.search(r'data-uix-load-more-href="/?(?P<more>[^"]+)"', more_widget_html)
-            if not mobj:
+            if not new_info:
                 break
 
-            more = self._download_json(
-                'https://www.youtube.com/%s' % mobj.group('more'), self._PLAYLIST_TITLE,
+            info.extend(new_info)
+
+            for video in new_info:
+                yield self.url_result(try_get(video, lambda x: x['videoId']), YoutubeIE.ie_key(), video_title=try_get(video, lambda x: x['title']['simpleText']))
+
+            if not continuation:
+                break
+
+            search_response = self._download_json(
+                'https://www.youtube.com/browse_ajax', self._PLAYLIST_TITLE,
                 'Downloading page #%s' % page_num,
                 transform_source=uppercase_escape,
-                headers=self._YOUTUBE_CLIENT_HEADERS)
-            content_html = more['content_html']
-            more_widget_html = more['load_more_widget_html']
+                query={
+                    "ctoken": try_get(continuation, lambda x: x["continuation"]),
+                    "continuation": try_get(continuation, lambda x: x["continuation"]),
+                    "itct": try_get(continuation, lambda x: x["clickTrackingParams"])
+                },
+                headers={
+                    "X-YouTube-Client-Name": try_get(yt_conf, lambda x: x["INNERTUBE_CONTEXT_CLIENT_NAME"]),
+                    "X-YouTube-Client-Version": try_get(yt_conf, lambda x: x["INNERTUBE_CONTEXT_CLIENT_VERSION"]),
+                    "X-Youtube-Identity-Token": try_get(yt_conf, lambda x: x["ID_TOKEN"]),
+                    "X-YouTube-Device": try_get(yt_conf, lambda x: x["DEVICE"]),
+                    "X-YouTube-Page-CL": try_get(yt_conf, lambda x: x["PAGE_CL"]),
+                    "X-YouTube-Page-Label": try_get(yt_conf, lambda x: x["PAGE_BUILD_LABEL"]),
+                    "X-YouTube-Variants-Checksum": try_get(yt_conf, lambda x: x["VARIANTS_CHECKSUM"]), 
+                })
 
     def _real_extract(self, url):
         page = self._download_webpage(