fix(cache): prevent corrupted cache rows from breaking model listings (#730)

Cache corruption (NULL model_name/file_name from legacy DB rows or partial
writes) caused format_response to raise KeyError/AttributeError, failing the
entire /loras/list request and showing no models in the UI.

Fix across three layers:
- format_response (lora/checkpoint/embedding): replace direct dict[] access
  with .get() fallbacks; return None for entries missing file_path
- handlers: filter None entries from list/excluded/fetch/duplicate/conflict
  endpoints instead of letting them crash or appear as null in responses
- model_scanner: always use validate_batch repaired copies (previously
  discarded when no invalid entries, leaving None values in raw_data)
- persistent_model_cache: add or-empty-string guards on read and write for
  nullable TEXT columns (model_name, file_name, folder, base_model, etc.)

tests/routes/test_base_model_routes_smoke.py

@@ -201,6 +201,45 @@ def test_list_models_returns_formatted_items(mock_service, mock_scanner):
     asyncio.run(scenario())
 
 
+def test_list_models_filters_out_corrupted_entries(mock_service, mock_scanner):
+    """Corrupted cache entries (format_response returns None) must not appear
+    in the response items nor cause a 500. See issue #730.
+    """
+    mock_service.paginated_items = [
+        {"file_path": "/tmp/good.safetensors", "name": "Good"},
+        {"file_path": None, "name": "Corrupted"},  # triggers None from format_response
+        {"file_path": "/tmp/also_good.safetensors", "name": "AlsoGood"},
+    ]
+
+    # Override format_response to return None for corrupted entries
+    original_format = mock_service.format_response
+
+    async def conditional_format(item):
+        if item.get("file_path") is None:
+            return None
+        return await original_format(item)
+
+    mock_service.format_response = conditional_format
+
+    async def scenario():
+        client = await create_test_client(mock_service)
+        try:
+            response = await client.get("/api/lm/test-models/list")
+            payload = await response.json()
+
+            assert response.status == 200
+            # Only the 2 non-corrupted entries should appear
+            assert len(payload["items"]) == 2
+            assert payload["items"][0]["name"] == "Good"
+            assert payload["items"][1]["name"] == "AlsoGood"
+            # None should never appear in the items list
+            assert None not in payload["items"]
+        finally:
+            await client.close()
+
+    asyncio.run(scenario())
+
+
 def test_model_types_endpoint_returns_counts(mock_service, mock_scanner):
     mock_service.model_types = [
         {"type": "LoRa", "count": 3},

tests/services/test_service_format_response_sub_type.py

@@ -199,8 +199,107 @@ class TestEmbeddingServiceFormatResponse:
             "from_civitai": True,
             "civitai": {},
         }
-        
+
         result = await embedding_service.format_response(embedding_data)
-        
+
         assert result["sub_type"] == "embedding"
         assert "model_type" not in result  # Removed in refactoring
+
+
+class TestFormatResponseCorruptedEntries:
+    """Test format_response handles corrupted cache entries gracefully (issue #730).
+
+    When cache rows have None/missing critical fields (e.g. from a partially
+    written or legacy DB), format_response must NOT raise KeyError/AttributeError.
+    Instead it returns None so the handler layer can filter the bad entry out
+    instead of failing the entire listing request.
+    """
+
+    @pytest.fixture
+    def mock_scanner(self):
+        scanner = MagicMock()
+        scanner._hash_index = MagicMock()
+        return scanner
+
+    @pytest.fixture
+    def lora_service(self, mock_scanner):
+        return LoraService(mock_scanner)
+
+    @pytest.fixture
+    def checkpoint_service(self, mock_scanner):
+        return CheckpointService(mock_scanner)
+
+    @pytest.fixture
+    def embedding_service(self, mock_scanner):
+        return EmbeddingService(mock_scanner)
+
+    @pytest.mark.asyncio
+    async def test_lora_returns_none_on_missing_file_path(self, lora_service):
+        """format_response returns None when file_path is missing (corrupted row)."""
+        lora_data = {
+            "model_name": "Test LoRA",
+            "file_name": "test_lora",
+            "file_path": None,  # corrupted: missing file_path
+            "folder": "",
+            "sha256": "abc123",
+            "tags": [],
+            "from_civitai": True,
+            "civitai": {},
+        }
+        result = await lora_service.format_response(lora_data)
+        assert result is None
+
+    @pytest.mark.asyncio
+    async def test_lora_handles_none_model_name_gracefully(self, lora_service):
+        """format_response should not crash when model_name is None (legacy DB row)."""
+        lora_data = {
+            "model_name": None,          # NULL from old DB row
+            "file_name": "test_lora",
+            "file_path": "/models/test_lora.safetensors",
+            "folder": "",
+            "sha256": "abc123",
+            "tags": [],
+            "from_civitai": True,
+            "civitai": {},
+        }
+        result = await lora_service.format_response(lora_data)
+        # Should not raise; model_name falls back to file_name
+        assert result is not None
+        assert result["model_name"] == "test_lora"
+
+    @pytest.mark.asyncio
+    async def test_checkpoint_returns_none_on_missing_file_path(self, checkpoint_service):
+        """format_response returns None when file_path is missing (corrupted row)."""
+        checkpoint_data = {
+            "model_name": "Test",
+            "file_name": "test",
+            "file_path": "",  # empty string == corrupted
+            "folder": "",
+            "sha256": "abc",
+            "tags": [],
+            "from_civitai": True,
+            "civitai": {},
+            "sub_type": "checkpoint",
+        }
+        result = await checkpoint_service.format_response(checkpoint_data)
+        assert result is None
+
+    @pytest.mark.asyncio
+    async def test_embedding_handles_none_fields_gracefully(self, embedding_service):
+        """format_response should not crash when optional fields are None."""
+        embedding_data = {
+            "model_name": None,
+            "file_name": None,
+            "file_path": "/models/test.pt",
+            "folder": None,
+            "sha256": "abc",
+            "tags": [],
+            "from_civitai": True,
+            "civitai": {},
+            "sub_type": "embedding",
+        }
+        result = await embedding_service.format_response(embedding_data)
+        assert result is not None
+        assert result["file_path"] == "/models/test.pt"
+        # model_name falls back to file_name which falls back to ""
+        assert result["model_name"] == ""