From 901ec3f08cc8c8ef4365ec2f49330087f8d93b73 Mon Sep 17 00:00:00 2001
From: Dev Kumar <devpatna2016@gmail.com>
Date: Tue, 21 Jul 2026 07:49:15 -0500
Subject: [PATCH] Make redirect dupefilter bypass chain-aware

Extend the self-redirect dupefilter fix to cover the whole redirect
chain, not just the immediate predecessor request. A redirect chain
that loops back to an earlier request in the same chain (A -> B -> C
-> A) previously still got silently dropped, because only the
immediate hop's fingerprint was compared. Track the set of
fingerprints seen so far in the chain via request.meta and bypass the
dupe filter if the new target matches any of them.

Refs #1225
---
 scrapy/downloadermiddlewares/redirect.py | 35 ++++++++++++++++++++++++
 1 file changed, 35 insertions(+)

diff --git a/scrapy/downloadermiddlewares/redirect.py b/scrapy/downloadermiddlewares/redirect.py
index 45af5c67a..2a94bb565 100644
--- a/scrapy/downloadermiddlewares/redirect.py
+++ b/scrapy/downloadermiddlewares/redirect.py
@@ -106,6 +106,23 @@ class BaseRedirectMiddleware:
                 reason,
             ]
             redirected.dont_filter = request.dont_filter
+            if not redirected.dont_filter and self._is_redirect_chain_duplicate(
+                redirected, request
+            ):
+                # The dupe filter marks a request's fingerprint as seen as
+                # soon as it is scheduled, before its response (and any
+                # resulting redirect) is known. If this redirect chain loops
+                # back to a URL that fingerprints the same as an earlier
+                # request in this *same* chain -- a literal self-redirect, a
+                # redirect whose query string is just reordered (which the
+                # fingerprinter canonicalizes away), or a longer
+                # A -> B -> C -> A style loop -- the redirected request would
+                # otherwise be treated as an already-seen duplicate and
+                # silently dropped instead of followed. Bypass the filter for
+                # that specific case only, so normal cross-request dedup
+                # (unrelated pages redirecting to the same canonical URL)
+                # still works as before.
+                redirected.dont_filter = True
             redirected.priority = request.priority + self.priority_adjust
             logger.debug(
                 "Redirecting (%(reason)s) to %(redirected)s from %(request)s",
@@ -120,6 +137,24 @@ class BaseRedirectMiddleware:
         )
         raise IgnoreRequest("max redirections reached")
 
+    def _is_redirect_chain_duplicate(
+        self, redirected: Request, request: Request
+    ) -> bool:
+        """Whether *redirected* fingerprints the same as *request* or as any
+        earlier request in the same redirect chain as *request*.
+
+        The set of fingerprints accumulated so far in the chain is carried
+        forward in ``request.meta["redirect_fingerprints"]``, bounded by
+        :setting:`REDIRECT_MAX_TIMES` since it grows by at most one entry per
+        hop.
+        """
+        assert self.crawler.request_fingerprinter
+        fingerprint = self.crawler.request_fingerprinter.fingerprint
+        seen_fingerprints = set(request.meta.get("redirect_fingerprints", ()))
+        seen_fingerprints.add(fingerprint(request))
+        redirected.meta["redirect_fingerprints"] = seen_fingerprints
+        return fingerprint(redirected) in seen_fingerprints
+
     def _build_redirect_request(
         self, source_request: Request, response: Response, *, url: str, **kwargs: Any
     ) -> Request: