fix(wall): conservative cleanup — only delete truly garbage posts, preserve FPT/auto_rewrite with valid images

- _wall_posts_for_view: VIEW-ONLY filtering (non-destructive) of placeholder
image auto posts (pollinations.ai, vnexpress logo, empty). Posts with valid
images are always kept. Sort newest-first, mixed sources.
- _post_has_slide_info: accept YouTube/TikTok/FB embed iframe videos as valid
slide-AI content (Short HOT oEmbed fallback case), not just /api/ai/short-file
- _wall_cleanup_once (startup): only persist-delete auto posts with NO valid
image AND no video AND no slides (true garbage). Never delete FPT embed posts
or auto_rewrite posts with real images — prevents data loss.
- /api/wall/cleanup: same conservative deletion logic

app_v2_entry.py

@@ -938,16 +938,23 @@ def _post_has_valid_image(p):
     return False
 
 def _post_has_slide_info(p):
-    """True if the post has usable slide/rewrite AI content (designed slides
-    or a generated short video with a thumbnail). Posts with neither slides
-    nor a short cannot be properly viewed as 'slide AI' content."""
+    """True if the post has usable slide/rewrite AI content:
+      - designed slides with text+image, OR
+      - a generated short video (/api/ai/short-file/) with a real thumbnail, OR
+      - an oEmbed iframe (YouTube/TikTok/Facebook embed) with a real thumbnail
+        (Short HOT fallback case — embeddable, viewable as a slide)."""
     slides = p.get("slides") or []
     if any((s.get("image") or "").strip() and (s.get("text") or "").strip() for s in slides if isinstance(s, dict)):
         return True
-    # A generated short with a real thumbnail counts as slide-AI content
-    if _post_has_valid_image(p) and (p.get("short_thumb") or "").strip():
-        vid = (p.get("video") or "").strip()
-        if vid and "api/ai/short-file/" in vid:
+    # Generated short video with a real thumbnail
+    vid = (p.get("video") or "").strip()
+    if vid and "api/ai/short-file/" in vid and (p.get("short_thumb") or "").strip():
+        return True
+    # oEmbed iframe (YouTube/TikTok/FB) with a valid image -> viewable embed slide
+    if vid and ("youtube.com/embed" in vid or "tiktok.com/embed" in vid or
+                "facebook.com/plugins" in vid or "instagram.com" in vid or
+                "twitter.com" in vid or "x.com" in vid or "player.vimeo" in vid):
+        if _post_has_valid_image(p):
             return True
     return False
 
@@ -983,11 +990,14 @@ def _created_ts(p):
 
 def _wall_posts_for_view():
     """Apply the user-facing wall filtering + sorting:
-      1) Drop auto-generated posts that have no valid image (broken/placeholder
-         thumb) OR no slide-AI info (summary-style auto posts with a wrong
-         image and no slides/video). User-authored posts (kind/source unset or
-         'user'/'vtv_recorder') are always kept.
-      2) Sort newest-first by creation timestamp (mixed sources)."""
+      1) Hide auto-generated posts that have a wrong/placeholder image (no
+         valid image at all — pollinations.ai, vnexpress logo, empty) or have
+         no slide-AI content (no slides / no short-video / no embed). These
+         are the 'bài đăng tự động bị sai ảnh / ko có thông tin slide AI' that
+         don't match the original article. This filter is VIEW-ONLY — it does
+         NOT delete from the persistent store.
+      2) Sort newest-first by creation timestamp, mixing all sources (FPT,
+         slide rewrite AI, etc.) without source separation."""
     posts = _load_wall_posts()
     if not posts:
         return []
@@ -995,11 +1005,10 @@ def _wall_posts_for_view():
     for p in posts:
         if not isinstance(p, dict):
             continue
-        if _is_auto_post(p):
-            # auto post: must have a valid (non-placeholder) image AND some
-            # slide-AI content, otherwise it shows the wrong image / no content
-            if not _post_has_valid_image(p) or not _post_has_slide_info(p):
-                continue
+        if _is_auto_post(p) and not (_post_has_valid_image(p) and _post_has_slide_info(p)):
+            # auto post with a wrong/placeholder/missing image or no slide AI
+            # content -> hide from the wall (non-destructive)
+            continue
         filtered.append(p)
     filtered.sort(key=_created_ts, reverse=True)
     return filtered
@@ -1244,10 +1253,14 @@ def api_wall_delete(post_id: str):
 
 @app.post('/api/wall/cleanup')
 def api_wall_cleanup():
-    """Permanently remove auto-generated wall posts that have a wrong/missing
-    image (placeholder) or no slide-AI content, and persist the cleaned store.
-    Also sorts the remaining posts newest-first so the wall is consistent.
-    Returns the number of removed posts and the new post count."""
+    """Permanently remove auto-generated wall posts that are completely
+    unrecoverable (no valid image, no video, no slides, no embed — i.e. a
+    post that can never be displayed), and sort the remaining posts
+    newest-first. Returns the number of removed posts and the new count.
+
+    NOTE: Conservative — auto posts that HAVE a valid image or a video/embed
+    are preserved (they remain visible at view time via
+    _wall_posts_for_view). Only use this to purge genuine garbage."""
     posts = _load_wall_posts()
     if not isinstance(posts, list):
         return JSONResponse({"error": "No posts"}, status_code=404)
@@ -1258,7 +1271,7 @@ def api_wall_cleanup():
         if not isinstance(p, dict):
             removed.append(str(p))
             continue
-        if _is_auto_post(p) and not (_post_has_valid_image(p) and _post_has_slide_info(p)):
+        if _is_auto_post(p) and not _post_has_valid_image(p) and not _post_has_slide_info(p) and not (p.get("video") or "").strip():
             removed.append(p.get('id', '?'))
         else:
             kept.append(p)
@@ -2532,10 +2545,15 @@ def _scheduler_loop():
 threading.Thread(target=_scheduler_loop, daemon=True, name='auto-rewrite-scheduler').start()
 
 def _wall_cleanup_once():
-    """One-time startup cleanup: remove auto-generated wall posts that have a
-    wrong/placeholder image or no slide-AI content, and sort the remaining
-    posts newest-first. Runs once a few seconds after import so the persistent
-    wall store is clean for every reader."""
+    """One-time startup cleanup: sort the persistent wall store newest-first
+    and remove truly-garbage auto posts (no valid image at ALL, no video, no
+    slides, no embed URL — i.e. a post that can never be displayed).
+
+    NOTE: This is intentionally conservative. Posts with a valid image or a
+    valid embed/video are NEVER deleted here — they are only hidden at view
+    time by _wall_posts_for_view() when the image is a placeholder. This
+    prevents accidental data loss of FPT Short AI posts whose image is valid
+    but may temporarily lack a short_thumb."""
     try:
         time.sleep(8)
         posts = _load_wall_posts()
@@ -2544,19 +2562,35 @@ def _wall_cleanup_once():
         kept = []
         removed = 0
         removed_ids = []
+        needs_sort = False
         for p in posts:
             if not isinstance(p, dict):
                 removed += 1
                 continue
-            if _is_auto_post(p) and not (_post_has_valid_image(p) and _post_has_slide_info(p)):
-                removed += 1
-                removed_ids.append(p.get('id', '?'))
-            else:
-                kept.append(p)
-        if removed > 0 or len(removed_ids) > 0:
+            if _is_auto_post(p):
+                has_img = _post_has_valid_image(p)
+                has_content = _post_has_slide_info(p)
+                vid = (p.get("video") or "").strip()
+                # Only delete auto posts that are completely unrecoverable:
+                # no valid image AND no video/slides/embed at all.
+                if not has_img and not has_content and not vid:
+                    removed += 1
+                    removed_ids.append(p.get('id', '?'))
+                    continue
+            kept.append(p)
+        # Sort if any post lacks proper ordering
+        prev_ts = None
+        is_sorted = True
+        for p in kept:
+            ts = _created_ts(p)
+            if prev_ts is not None and ts > prev_ts:
+                is_sorted = False
+                break
+            prev_ts = ts
+        if removed > 0 or not is_sorted:
             kept.sort(key=_created_ts, reverse=True)
             _save_wall_posts(kept)
-            print(f"[wall] startup cleanup removed {removed} broken placeholder auto posts ({removed_ids[:10]}), kept {len(kept)}")
+            print(f"[wall] startup cleanup: removed {removed} garbage posts ({removed_ids[:10]}), kept {len(kept)}, sorted newest-first")
     except Exception as e:
         print(f"[wall] startup cleanup error: {e}")