[extractor/youtube] Improve description parsing performance (#7315)

* The parsing is skipped when not needed
* The regex is improved by simulating atomic groups with lookaheads

Authored by: pukkandan, berkanteber

diff --git a/yt_dlp/extractor/youtube.py b/yt_dlp/extractor/youtube.py
index a90118680ffbc26998d930b28d76468ed8911d10..ef9f1f11c2f518616aa162a72c3c0c6f6c568c53 100644 (file)
--- a/yt_dlp/extractor/youtube.py
+++ b/yt_dlp/extractor/youtube.py
@@ -4346,15 +4346,21 @@ def process_language(container, base_url, lang_code, sub_name, query):
                         info[d_k] = parse_duration(query[k][0])
 
         # Youtube Music Auto-generated description
-        if video_description:
+        if (video_description or '').strip().endswith('\nAuto-generated by YouTube.'):
+            # XXX: Causes catastrophic backtracking if description has "·"
+            # E.g. https://www.youtube.com/watch?v=DoPaAxMQoiI
+            # Simulating atomic groups:  (?P<a>[^xy]+)x  =>  (?=(?P<a>[^xy]+))(?P=a)x
+            # reduces it, but does not fully fix it. https://regex101.com/r/8Ssf2h/2
             mobj = re.search(
                 r'''(?xs)
-                    (?P<track>[^·\n]+)·(?P<artist>[^\n]+)\n+
-                    (?P<album>[^\n]+)
+                    (?=(?P<track>[^\n·]+))(?P=track)·
+                    (?=(?P<artist>[^\n]+))(?P=artist)\n+
+                    (?=(?P<album>[^\n]+))(?P=album)\n
                     (?:.+?℗\s*(?P<release_year>\d{4})(?!\d))?
                     (?:.+?Released on\s*:\s*(?P<release_date>\d{4}-\d{2}-\d{2}))?
-                    (.+?\nArtist\s*:\s*(?P<clean_artist>[^\n]+))?
-                    .+\nAuto-generated\ by\ YouTube\.\s*$
+                    (.+?\nArtist\s*:\s*
+                        (?=(?P<clean_artist>[^\n]+))(?P=clean_artist)\n
+                    )?.+\nAuto-generated\ by\ YouTube\.\s*$
                 ''', video_description)
             if mobj:
                 release_year = mobj.group('release_year')