Fix xxxtik empty results; replay solved identity on raw requests

Raw requester calls did not adopt the remembered FlareSolverr user agent,
so a RedGifs token minted under it was rejected on every gif lookup.
Tag/creator pages that mix non-RedGifs sources are refilled from later
cursor batches instead of coming back empty.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MUYdCh2nCVsQYgJAT7gwSi

docs/provider-fix-todo.md

@@ -47,7 +47,7 @@ Status: `[ ]` open, `[~]` in progress, `[x]` fixed and verified, `[-]` not repro
 | [x] | `chaturbate` | "girl" and "milf" both 0 | format arguments swapped in the search URL (`girl/api/...?keywords=https://chaturbate.com`, a relative-URL builder error); fixed + unit test. Also stale-cache branch never refreshed. 90 results per query, yt-dlp resolves the live stream |
 | [x] | `cumlouder` | unknown tag takes 146s before returning 0 | a real 404 went through the requester's Jina/FlareSolverr fallback chain; now the raw response is read, 404 returns empty immediately (0s), 5xx/429 still retried. check.py passes |
 | [x] | `vjav` | "girl" 0, "milf" 10 | not term-specific: "girl" exactly matches a member username with no uploads, so the implicit uploader match hijacked the keyword search. Empty uploader/tag matches now fall back to keyword search (site API returns 34k results for "girl"). check.py passes |
-| [ ] | `xxxtik` | "girl" 0 (not re-run serially) | |
+| [x] | `xxxtik` | "girl" 0 (not re-run serially) | three things. (1) shared requester: raw requests did not replay the remembered FlareSolverr identity, so a RedGifs token minted under the solved UA was rejected (401) on every gif lookup after one transient connect error (empty feed, 71s); `get_raw_with_headers_timeout` now adopts it. (2) tag/creator pages mixing non-RedGifs posts (erome/onlyfans links) emptied whole pages; now pulls up to 4 raw batches per page. (3) "girl" itself has no RedGifs posts in its first 600 entries, so empty is correct there; "milf" returns 10. Feed/p2 10 items, check.py passes. First request can still take ~50s when api.redgifs.com has a connect error (FlareSolverr detour) |
 | [x] | `hypnotube` | "milf" 0, "girl" 20 | same root cause as above (second search on a session returned nothing), not term-specific |
 
 ## 5. Slow channels

src/providers/xxxtik.rs

@@ -30,6 +30,7 @@ const RESOLVE_CONCURRENCY: usize = 8;
 // xxxtik's listing endpoints are cursor-based (the numeric `id` of the last item
 // returned), not page-number based, so reaching page N requires walking N
 // sequential requests from the start of the feed.
+const MAX_REFILL_FETCHES: usize = 4;
 const MAX_PAGE_WALK: u16 = 25;
 
 // xxxtik exposes ~60k tags through /tag/options, far too many to background-load
@@ -423,18 +424,43 @@ impl XxxtikProvider {
         let mut items = Vec::new();
 
         for _ in 0..page {
+            // Tag and creator pages mix in posts whose source is not RedGifs
+            // (erome links, empty sources); those cannot be resolved here and
+            // used to leave whole pages empty ("girl" returned nothing). Keep
+            // pulling raw batches, bounded, until a page has enough
+            // resolvable posts.
+            let mut collected: Vec<Post> = Vec::new();
+            let mut exhausted = false;
+            for _ in 0..MAX_REFILL_FETCHES {
                 let batch = self.fetch_one_page(path, cursor, per_page, options).await?;
                 if batch.is_empty() {
-                items = Vec::new();
+                    exhausted = true;
                     break;
                 }
                 cursor = batch.last().map(|post| post.id).unwrap_or(cursor);
-            items = batch;
+                let short_batch = batch.len() < per_page;
+                collected.extend(batch.into_iter().filter(Self::is_redgifs_post));
+                if collected.len() >= per_page || short_batch {
+                    break;
+                }
+            }
+            if exhausted && collected.is_empty() {
+                items = Vec::new();
+                break;
+            }
+            items = collected;
         }
 
         Ok(items)
     }
 
+    fn is_redgifs_post(post: &Post) -> bool {
+        post.source
+            .as_deref()
+            .map(|source| source.contains("redgifs.com/"))
+            .unwrap_or(false)
+    }
+
     async fn fetch_target_posts(
         &self,
         target: &Target,
@@ -791,6 +817,25 @@ impl Provider for XxxtikProvider {
 
 #[cfg(test)]
 mod tests {
+    #[test]
+    fn only_redgifs_sources_are_resolvable() {
+        let mk = |source: Option<&str>| super::Post {
+            id: 1,
+            uuid: "u".to_string(),
+            description: None,
+            source: source.map(str::to_string),
+            status: None,
+            created_at: String::new(),
+            author: Default::default(),
+            tags: vec![],
+        };
+        assert!(XxxtikProvider::is_redgifs_post(&mk(Some("https://www.redgifs.com/watch/abc"))));
+        assert!(XxxtikProvider::is_redgifs_post(&mk(Some("https://redgifs.com/watch/abc"))));
+        assert!(!XxxtikProvider::is_redgifs_post(&mk(Some("https://www.erome.com/a/CE5z46ne#39"))));
+        assert!(!XxxtikProvider::is_redgifs_post(&mk(Some(""))));
+        assert!(!XxxtikProvider::is_redgifs_post(&mk(None)));
+    }
+
     use super::*;
 
     #[test]

src/util/requester.rs

@@ -496,6 +496,10 @@ impl Requester {
         timeout: Option<Duration>,
         http_version: Option<Version>,
     ) -> Result<Response, wreq::Error> {
+        // Replay the solved FlareSolverr identity here too: a token minted
+        // through `get_with_headers` under that user agent is rejected (401)
+        // when the follow-up raw request goes out with the default one.
+        self.adopt_remembered_identity(url);
         let cookie_preview = self.debug_cookie_preview_from_owned_headers(url, &headers);
         #[cfg(not(feature = "debug"))]
         let _ = &cookie_preview;