fix: use safe orjson imports and fix Black formatting for CI

Replace hard orjson removal with try/except safe imports that preserve
perf when orjson is available. Revert transformation.py to response.json()
matching main. Fix test assertions to use safe_dumps() for orjson-agnostic
JSON comparison. Apply Black formatting to sidecar files.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

litellm/llms/custom_httpx/sidecar_transport.py

@@ -68,7 +68,11 @@ class LiteLLMSidecarTransport(httpx.AsyncBaseTransport):
         provider_base = f"{parsed.scheme}://{parsed.host}"
         if parsed.port and parsed.port not in (80, 443):
             provider_base += f":{parsed.port}"
-        path = parsed.raw_path.decode("ascii") if isinstance(parsed.raw_path, bytes) else str(parsed.raw_path)
+        path = (
+            parsed.raw_path.decode("ascii")
+            if isinstance(parsed.raw_path, bytes)
+            else str(parsed.raw_path)
+        )
 
         # Extract auth header if present
         api_key = ""

litellm/llms/openai_like/chat/transformation.py

@@ -2,7 +2,6 @@
 OpenAI-like chat completion transformation
 """
 
-import json
 from typing import TYPE_CHECKING, Any, List, Optional, Tuple, Union
 
 import httpx
@@ -95,7 +94,7 @@ class OpenAILikeChatConfig(OpenAIGPTConfig):
         custom_llm_provider: Optional[str],
         base_model: Optional[str],
     ) -> ModelResponse:
-        response_json = json.loads(response.content)
+        response_json = response.json()
         logging_obj.post_call(
             input=messages,
             api_key="",

litellm/proxy/proxy_server.py

@@ -946,16 +946,22 @@ async def proxy_startup_event(app: FastAPI):  # noqa: PLR0915
     shared_aiohttp_session = await _initialize_shared_aiohttp_session()
 
     ## Initialize Rust sidecar client (optional, for high-perf forwarding)
-    _use_sidecar = os.environ.get("USE_SIDECAR", "").lower() == "true" or general_settings.get("use_sidecar", False)
+    _use_sidecar = os.environ.get(
+        "USE_SIDECAR", ""
+    ).lower() == "true" or general_settings.get("use_sidecar", False)
     if _use_sidecar:
         # Ensure env var is set so AsyncHTTPHandler._should_use_sidecar_transport() picks it up
         os.environ["USE_SIDECAR"] = "true"
 
         from litellm.proxy.sidecar_client import init_sidecar_client
 
-        _sidecar_port = int(os.environ.get("SIDECAR_PORT", general_settings.get("sidecar_port", 8787)))
+        _sidecar_port = int(
+            os.environ.get("SIDECAR_PORT", general_settings.get("sidecar_port", 8787))
+        )
         os.environ.setdefault("SIDECAR_PORT", str(_sidecar_port))
-        _sidecar_binary = os.environ.get("SIDECAR_BINARY", general_settings.get("sidecar_binary", ""))
+        _sidecar_binary = os.environ.get(
+            "SIDECAR_BINARY", general_settings.get("sidecar_binary", "")
+        )
         await init_sidecar_client(
             port=_sidecar_port,
             binary=_sidecar_binary or None,

litellm/proxy/sidecar_client.py

@@ -58,9 +58,7 @@ class SidecarClient:
 
         self._healthy = await self._check_health()
         if self._healthy:
-            verbose_proxy_logger.info(
-                f"Sidecar client connected to {self.sidecar_url}"
-            )
+            verbose_proxy_logger.info(f"Sidecar client connected to {self.sidecar_url}")
         else:
             verbose_proxy_logger.warning(
                 f"Sidecar not available at {self.sidecar_url}, will use fallback"
@@ -144,9 +142,7 @@ class SidecarClient:
             self._process.terminate()
             try:
                 await asyncio.wait_for(
-                    asyncio.get_event_loop().run_in_executor(
-                        None, self._process.wait
-                    ),
+                    asyncio.get_event_loop().run_in_executor(None, self._process.wait),
                     timeout=5,
                 )
             except asyncio.TimeoutError:

tests/test_litellm/integrations/arize/test_arize_utils.py

@@ -18,6 +18,7 @@ from litellm.integrations._types.open_inference import (
 )
 from litellm.integrations.arize.arize import ArizeLogger
 from litellm.integrations.custom_logger import CustomLogger
+from litellm.litellm_core_utils.safe_json_dumps import safe_dumps
 from litellm.types.utils import Choices, StandardCallbackDynamicParams
 
 
@@ -88,7 +89,7 @@ def test_arize_set_attributes():
 
     # Metadata attached to the span
     span.set_attribute.assert_any_call(
-        SpanAttributes.METADATA, json.dumps({"key_1": "value_1", "key_2": None})
+        SpanAttributes.METADATA, safe_dumps({"key_1": "value_1", "key_2": None})
     )
 
     # Basic LLM information
@@ -147,7 +148,7 @@ def test_arize_set_attributes():
 
     # Invocation parameters
     span.set_attribute.assert_any_call(
-        SpanAttributes.LLM_INVOCATION_PARAMETERS, '{"user": "test_user"}'
+        SpanAttributes.LLM_INVOCATION_PARAMETERS, safe_dumps({"user": "test_user"})
     )
 
     # User ID
@@ -178,10 +179,20 @@ def test_arize_set_attributes_responses_api():
     Verifies that multiple output types are correctly handled.
     """
     from unittest.mock import MagicMock
-    from litellm.types.llms.openai import ResponsesAPIResponse, ResponseAPIUsage, OutputTokensDetails
-    from openai.types.responses import ResponseReasoningItem, ResponseOutputMessage, ResponseOutputText
+
+    from openai.types.responses import (
+        ResponseOutputMessage,
+        ResponseOutputText,
+        ResponseReasoningItem,
+    )
     from openai.types.responses.response_reasoning_item import Summary
 
+    from litellm.types.llms.openai import (
+        OutputTokensDetails,
+        ResponseAPIUsage,
+        ResponsesAPIResponse,
+    )
+
     span = MagicMock()  # Mocked tracing span to test attribute setting
 
     # Construct kwargs to simulate a real LLM request scenario
@@ -212,11 +223,8 @@ def test_arize_set_attributes_responses_api():
                 id="reasoning-001",
                 type="reasoning",
                 summary=[
-                    Summary(
-                        text="First, I need to analyze...",
-                        type="summary_text"
-                    )
-                ]
+                    Summary(text="First, I need to analyze...", type="summary_text")
+                ],
             ),
             ResponseOutputMessage(
                 id="msg-001",
@@ -229,17 +237,15 @@ def test_arize_set_attributes_responses_api():
                         text="The answer is 42",
                         type="output_text",
                     )
-                ]
-            )
+                ],
+            ),
         ],
         usage=ResponseAPIUsage(
             input_tokens=120,
             output_tokens=250,
             total_tokens=370,
-            output_tokens_details=OutputTokensDetails(
-                reasoning_tokens=180
-            )
-        )
+            output_tokens_details=OutputTokensDetails(reasoning_tokens=180),
+        ),
     )
 
     ArizeLogger.set_arize_attributes(span, kwargs, response_obj)
@@ -247,21 +253,18 @@ def test_arize_set_attributes_responses_api():
     # Verify reasoning summary was set (index 0)
     span.set_attribute.assert_any_call(
         f"{SpanAttributes.LLM_OUTPUT_MESSAGES}.0.{MessageAttributes.MESSAGE_REASONING_SUMMARY}",
-        "First, I need to analyze..."
+        "First, I need to analyze...",
     )
 
     # Verify message content was set (index 1)
-    span.set_attribute.assert_any_call(
-        SpanAttributes.OUTPUT_VALUE,
-        "The answer is 42"
-    )
+    span.set_attribute.assert_any_call(SpanAttributes.OUTPUT_VALUE, "The answer is 42")
     span.set_attribute.assert_any_call(
         f"{SpanAttributes.LLM_OUTPUT_MESSAGES}.1.{MessageAttributes.MESSAGE_CONTENT}",
-        "The answer is 42"
+        "The answer is 42",
     )
     span.set_attribute.assert_any_call(
         f"{SpanAttributes.LLM_OUTPUT_MESSAGES}.1.{MessageAttributes.MESSAGE_ROLE}",
-        "assistant"
+        "assistant",
     )
 
     # Verify token counts including reasoning tokens
@@ -335,42 +338,34 @@ def test_construct_dynamic_arize_headers():
 
     # Test with all parameters present
     dynamic_params_full = StandardCallbackDynamicParams(
-        arize_api_key="test_api_key", 
-        arize_space_id="test_space_id"
+        arize_api_key="test_api_key", arize_space_id="test_space_id"
     )
     arize_logger = ArizeLogger()
-    
+
     headers = arize_logger.construct_dynamic_otel_headers(dynamic_params_full)
-    expected_headers = {
-        "api_key": "test_api_key",
-        "arize-space-id": "test_space_id"
-    }
+    expected_headers = {"api_key": "test_api_key", "arize-space-id": "test_space_id"}
     assert headers == expected_headers
-        
+
     # Test with only space_id
     dynamic_params_space_id_only = StandardCallbackDynamicParams(
         arize_space_id="test_space_id"
     )
-    
+
     headers = arize_logger.construct_dynamic_otel_headers(dynamic_params_space_id_only)
-    expected_headers = {
-        "arize-space-id": "test_space_id"
-    }
+    expected_headers = {"arize-space-id": "test_space_id"}
     assert headers == expected_headers
-    
+
     # Test with empty parameters dict
     dynamic_params_empty = StandardCallbackDynamicParams()
-    
+
     headers = arize_logger.construct_dynamic_otel_headers(dynamic_params_empty)
     assert headers == {}
 
     # test with space key and api key
     dynamic_params_space_key_and_api_key = StandardCallbackDynamicParams(
-        arize_space_key="test_space_key",
-        arize_api_key="test_api_key"
+        arize_space_key="test_space_key", arize_api_key="test_api_key"
     )
-    headers = arize_logger.construct_dynamic_otel_headers(dynamic_params_space_key_and_api_key)
-    expected_headers = {
-        "arize-space-id": "test_space_key",
-        "api_key": "test_api_key"
-    }
+    headers = arize_logger.construct_dynamic_otel_headers(
+        dynamic_params_space_key_and_api_key
+    )
+    expected_headers = {"arize-space-id": "test_space_key", "api_key": "test_api_key"}

tests/test_litellm/proxy/agent_endpoints/test_agent_registry.py

@@ -4,6 +4,7 @@ from unittest.mock import AsyncMock, MagicMock
 
 import pytest
 
+from litellm.litellm_core_utils.safe_json_dumps import safe_dumps
 from litellm.proxy.agent_endpoints.agent_registry import AgentRegistry
 
 
@@ -110,5 +111,5 @@ async def test_update_agent_in_db_preserves_explicit_static_headers_and_extra_he
     call_kwargs = mock_update.call_args.kwargs
     update_data = call_kwargs["data"]
 
-    assert update_data["static_headers"] == '{"Authorization": "Bearer xyz"}'
+    assert update_data["static_headers"] == safe_dumps({"Authorization": "Bearer xyz"})
     assert update_data["extra_headers"] == ["X-Custom-Header"]