Fix: Wrap feature access in try/except to prevent torchcodec ImportError

utils/status.py

@@ -34,27 +34,62 @@ def check_setup_status():
     
     if _hf_dataset is not None:
         # Check if audio column exists without iterating (which would trigger decoding)
+        # Even accessing .features might trigger torchcodec requirement, so wrap carefully
         audio_count = 0
         try:
-            # Check if audio feature exists without decoding
-            if 'train' in _hf_dataset and 'audio' in _hf_dataset['train'].features:
-                # Use dataset info to check if audio exists without iterating
-                # If we can't check without iterating, assume audio is available if feature exists
+            # Try to check features - this might trigger torchcodec ImportError
+            has_train_audio = False
+            has_test_audio = False
+            
+            try:
+                if 'train' in _hf_dataset:
+                    features = _hf_dataset['train'].features
+                    has_train_audio = 'audio' in features if features else False
+            except ImportError as e:
+                if 'torchcodec' in str(e).lower():
+                    # Can't check - torchcodec not available
+                    has_train_audio = False
+                else:
+                    raise
+            
+            try:
+                if 'test' in _hf_dataset:
+                    features = _hf_dataset['test'].features
+                    has_test_audio = 'audio' in features if features else False
+            except ImportError as e:
+                if 'torchcodec' in str(e).lower():
+                    # Can't check - torchcodec not available
+                    has_test_audio = False
+                else:
+                    raise
+            
+            if has_train_audio:
                 status["audio_from_hf"] = True
-                # Estimate count from dataset size (safer than iterating)
-                audio_count = len(_hf_dataset['train']) if 'train' in _hf_dataset else 0
-            if 'test' in _hf_dataset and 'audio' in _hf_dataset['test'].features:
-                audio_count += len(_hf_dataset['test']) if 'test' in _hf_dataset else 0
+                try:
+                    audio_count = len(_hf_dataset['train'])
+                except:
+                    pass
+            if has_test_audio:
                 status["audio_from_hf"] = True
+                try:
+                    audio_count += len(_hf_dataset['test'])
+                except:
+                    pass
             
             if audio_count > 0:
                 status["audio_count_hf"] = audio_count
+                
+        except ImportError as e:
+            if 'torchcodec' in str(e).lower():
+                # torchcodec not installed - can't check audio features
+                # Don't set audio_from_hf to avoid false positives
+                pass
+            else:
+                raise
         except Exception as e:
-            # If we can't check audio (e.g., torchcodec not installed), 
-            # just mark that audio feature exists
-            if 'train' in _hf_dataset and 'audio' in _hf_dataset['train'].features:
-                status["audio_from_hf"] = True
-                # Don't set count if we can't safely check
+            # Other errors - don't fail, just skip audio check
+            if 'torchcodec' not in str(e).lower():
+                print(f"⚠ Warning checking audio features: {e}")
     
     # Check ESPnet
     try: