Fix: robust empty-pairs handling, CSV trimming, and Python 3.8+ type hints; improve slider bounds and graceful no-op when no pending pairs

app.py

@@ -59,7 +59,12 @@ def _load_task_rows(task_name: str) -> List[Dict[str, str]]:
 
     with open(csv_path, newline="", encoding="utf-8") as csv_file:
         reader = csv.DictReader(csv_file)
-        return [row for row in reader]
+        rows: List[Dict[str, str]] = []
+        for row in reader:
+            # Trim whitespaces in all string fields to avoid path/key mismatches
+            cleaned = {k.strip(): (v.strip() if isinstance(v, str) else v) for k, v in row.items()}
+            rows.append(cleaned)
+        return rows
 
 
 def _build_image_pairs(rows: List[Dict[str, str]], task_name: str) -> List[Dict[str, str]]:
@@ -166,12 +171,13 @@ def load_task(task_name: str):
         raise gr.Error("Please select a task first.")
 
     rows = _load_task_rows(task_name)
-    pairs = _build_image_pairs(rows, task_name)
+    pairs_all = _build_image_pairs(rows, task_name)
     # Filter out already evaluated pairs from persistent CSV
     done_keys = _read_existing_eval_keys(task_name)
     def key_of(p: Dict[str, str]):
         return (p["test_id"], frozenset({p["model1_name"], p["model2_name"]}), p["org_img"])
-    pairs = [p for p in pairs if key_of(p) not in done_keys]
+    pairs = [p for p in pairs_all if key_of(p) not in done_keys]
+    done_len = len([p for p in pairs_all if key_of(p) in done_keys])
 
     # Balanced schedule across test_ids with a stable randomization
     seed_env = os.environ.get("SCHEDULE_SEED")
@@ -183,7 +189,15 @@ def load_task(task_name: str):
         p["swap"] = bool(idx % 2)  # True -> A=B's image; False -> A=A's image
 
     if not pairs:
-        raise gr.Error("No valid image pairs found for evaluation. Please check the data.")
+        try:
+            print("[VisArena] No pending pairs.")
+            print("[VisArena] total_pairs=", len(pairs_all))
+            print("[VisArena] already_done=", done_len)
+            print("[VisArena] persist_csv=", _persist_csv_path_for_task(task_name))
+        except Exception:
+            pass
+        # Return empty list; UI will render an informative message instead of erroring out
+        return []
 
     return pairs
 
@@ -289,10 +303,22 @@ def _upload_eval_record_to_dataset(task_name: str, row: Dict[str, object]) -> Tu
 
 def on_task_change(task_name: str, _state_pairs: List[Dict[str, str]]):
     pairs = load_task(task_name)
-    pair = pairs[0]
-    header = _format_pair_header(pair)
     # Defaults for A and B (8 sliders total)
     default_scores = [3, 3, 3, 3, 3, 3, 3, 3]
+    if not pairs:
+        return (
+            [],
+            gr.update(value=0, minimum=0, maximum=0, visible=False),
+            gr.update(value=""),
+            gr.update(value=None),
+            gr.update(value=None),
+            gr.update(value=None),
+            *default_scores,
+            gr.update(value="No pending pairs. Either all pairs are already evaluated or data paths are invalid."),
+        )
+
+    pair = pairs[0]
+    header = _format_pair_header(pair)
     # Pick display order according to swap flag
     a_path = pair["model2_path"] if pair.get("swap") else pair["model1_path"]
     b_path = pair["model1_path"] if pair.get("swap") else pair["model2_path"]
@@ -311,7 +337,16 @@ def on_task_change(task_name: str, _state_pairs: List[Dict[str, str]]):
 
 def on_pair_navigate(index: int, pairs: List[Dict[str, str]]):
     if not pairs:
-        raise gr.Error("请先选择任务。")
+        # Gracefully no-op when no pairs
+        return (
+            gr.update(value=0, minimum=0, maximum=0, visible=False),
+            gr.update(value=""),
+            gr.update(value=None),
+            gr.update(value=None),
+            gr.update(value=None),
+            3, 3, 3, 3,  # A
+            3, 3, 3, 3,  # B
+        )
     index = int(index)
     index = max(0, min(index, len(pairs) - 1))
     pair = pairs[index]
@@ -343,10 +378,28 @@ def on_submit(
     b_overall_score: int,
 ):
     if not task_name:
-        raise gr.Error("请先选择任务。")
+        return (
+            gr.update(value=0),
+            gr.update(value=""),
+            gr.update(value=None),
+            gr.update(value=None),
+            gr.update(value=None),
+            3, 3, 3, 3,
+            3, 3, 3, 3,
+            gr.update(value="Please select a task first."),
+        )
 
     if not pairs:
-        raise gr.Error("No image pairs loaded for the current task.")
+        return (
+            gr.update(value=0, minimum=0, maximum=0, visible=False),
+            gr.update(value=""),
+            gr.update(value=None),
+            gr.update(value=None),
+            gr.update(value=None),
+            3, 3, 3, 3,
+            3, 3, 3, 3,
+            gr.update(value="No pending pairs to submit."),
+        )
 
     pair = pairs[index]
     score_map = {