[youtube] Improve video upload date handling (#3029)

* Don't prefer UTC upload date for past live streams/premieres
* Improve regex (fixes a regression)

Authored-by: coletdjnz

yt_dlp/extractor/youtube.py

@@ -730,11 +730,11 @@ def _extract_time_text(self, renderer, *path_list):
             timestamp = (
                 unified_timestamp(text) or unified_timestamp(
                     self._search_regex(
-                        (r'(?:.+|^)(?:live|premieres|ed|ing)(?:\s*on)?\s*(.+\d)', r'\w+[\s,\.-]*\w+[\s,\.-]+20\d{2}'),
+                        (r'([a-z]+\s*\d{1,2},?\s*20\d{2})', r'(?:.+|^)(?:live|premieres|ed|ing)(?:\s*(?:on|for))?\s*(.+\d)'),
                         text.lower(), 'time text', default=None)))
 
         if text and timestamp is None:
-            self.report_warning('Cannot parse localized time text' + bug_reports_message(), only_once=True)
+            self.report_warning(f"Cannot parse localized time text '{text}'" + bug_reports_message(), only_once=True)
         return timestamp, text
 
     def _extract_response(self, item_id, query, note='Downloading API JSON', headers=None,
@@ -1204,7 +1204,7 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'id': 'Tq92D6wQ1mg',
                 'title': '[MMD] Adios - EVERGLOW [+Motion DL]',
                 'ext': 'mp4',
-                'upload_date': '20191227',
+                'upload_date': '20191228',
                 'uploader_id': 'UC1yoRdFoFJaCY-AGfD9W0wQ',
                 'uploader': 'Projekt Melody',
                 'description': 'md5:17eccca93a786d51bc67646756894066',
@@ -1297,6 +1297,7 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
             },
             'expected_warnings': [
                 'DASH manifest missing',
+                'Some formats are possibly damaged'
             ]
         },
         # Olympics (https://github.com/ytdl-org/youtube-dl/issues/4431)
@@ -1569,7 +1570,7 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'title': 'md5:e41008789470fc2533a3252216f1c1d1',
                 'description': 'md5:a677553cf0840649b731a3024aeff4cc',
                 'duration': 721,
-                'upload_date': '20150127',
+                'upload_date': '20150128',
                 'uploader_id': 'BerkmanCenter',
                 'uploader_url': r're:https?://(?:www\.)?youtube\.com/user/BerkmanCenter',
                 'uploader': 'The Berkman Klein Center for Internet & Society',
@@ -1601,7 +1602,7 @@ class YoutubeIE(YoutubeBaseInfoExtractor):
                 'title': 'Democratic Socialism and Foreign Policy | Bernie Sanders',
                 'description': 'md5:13a2503d7b5904ef4b223aa101628f39',
                 'duration': 4060,
-                'upload_date': '20151119',
+                'upload_date': '20151120',
                 'uploader': 'Bernie Sanders',
                 'uploader_id': 'UCH1dpzjCEiGAt8CXkryhkZg',
                 'uploader_url': r're:https?://(?:www\.)?youtube\.com/channel/UCH1dpzjCEiGAt8CXkryhkZg',
@@ -3565,14 +3566,12 @@ def process_language(container, base_url, lang_code, sub_name, query):
                 or self._extract_chapters_from_engagement_panel(initial_data, duration)
                 or None)
 
-            contents = try_get(
-                initial_data,
-                lambda x: x['contents']['twoColumnWatchNextResults']['results']['results']['contents'],
-                list) or []
-            for content in contents:
-                vpir = content.get('videoPrimaryInfoRenderer')
+        contents = traverse_obj(
+            initial_data, ('contents', 'twoColumnWatchNextResults', 'results', 'results', 'contents'),
+            expected_type=list, default=[])
+
+        vpir = get_first(contents, 'videoPrimaryInfoRenderer')
         if vpir:
-                    info['upload_date'] = strftime_or_none(self._extract_time_text(vpir, 'dateText')[0], '%Y%m%d')
             stl = vpir.get('superTitleLink')
             if stl:
                 stl = self._get_text(stl)
@@ -3613,7 +3612,7 @@ def process_language(container, base_url, lang_code, sub_name, query):
                     'like_count': str_to_int(like_count),
                     'dislike_count': str_to_int(dislike_count),
                 })
-                vsir = content.get('videoSecondaryInfoRenderer')
+        vsir = get_first(contents, 'videoSecondaryInfoRenderer')
         if vsir:
             vor = traverse_obj(vsir, ('owner', 'videoOwnerRenderer'))
             info.update({
@@ -3652,15 +3651,15 @@ def process_language(container, base_url, lang_code, sub_name, query):
             'channel_url': 'uploader_url',
         }
 
-        # The upload date for scheduled and current live streams / premieres in microformats
-        # is generally the true upload date. Although not in UTC, we will prefer that in this case.
-        # Note this changes to the published date when the stream/premiere has finished.
+        # The upload date for scheduled, live and past live streams / premieres in microformats
+        # may be different from the stream date. Although not in UTC, we will prefer it in this case.
         # See: https://github.com/yt-dlp/yt-dlp/pull/2223#issuecomment-1008485139
-        if not info.get('upload_date') or info.get('is_live') or info.get('live_status') == 'is_upcoming':
-            info['upload_date'] = (
+        upload_date = (
             unified_strdate(get_first(microformats, 'uploadDate'))
-                or unified_strdate(search_meta('uploadDate'))
-                or info.get('upload_date'))
+            or unified_strdate(search_meta('uploadDate')))
+        if not upload_date or (not info.get('is_live') and not info.get('was_live') and info.get('live_status') != 'is_upcoming'):
+            upload_date = strftime_or_none(self._extract_time_text(vpir, 'dateText')[0], '%Y%m%d')
+        info['upload_date'] = upload_date
 
         for to, frm in fallbacks.items():
             if not info.get(to):