From 874f9206136cdef9d5de421bc92cbfbc92c55612 Mon Sep 17 00:00:00 2001
From: STiFLeR7 <hillaniljppatel@gmail.com>
Date: Wed, 23 Sep 2026 10:15:58 +0530
Subject: [PATCH] test(responses): replace no-op regression test with real
 assertions against _stream_usage_tracking_updates

Greptile flagged that the async test body was a bare pass and the only
running assertion checked Python's `in` operator on a hardcoded list,
never touching production code. Parametrized over route_type against
the actual function, verified red-before-green by disabling the gate.
---
 .../responses/test_regression_issue_28553.py  | 63 +++++++------------
 1 file changed, 21 insertions(+), 42 deletions(-)

diff --git a/tests/unit/responses/test_regression_issue_28553.py b/tests/unit/responses/test_regression_issue_28553.py
index a621154c5f9..5ff7f9aeadc 100644
--- a/tests/unit/responses/test_regression_issue_28553.py
+++ b/tests/unit/responses/test_regression_issue_28553.py
@@ -1,44 +1,23 @@
-from litellm.proxy.common_request_processing import ProxyBaseLLMRequestProcessing
+import pytest
+
+from litellm.proxy.common_request_processing import _stream_usage_tracking_updates
 
 
-async def test_regression_issue_28553_stream_usage_whitelist():
-    """
-    Test that stream_options.include_usage is only injected for chat/text completions,
-    and explicitly NOT for the Responses API (aresponses).
-    Fixes #28553.
-    """
-    # Initialize processor with mock data
-    processor = ProxyBaseLLMRequestProcessing(data={"stream": True, "model": "gpt-4o"})
-    assert processor.data["stream"] is True
-
-    # 1. Test case: acompletion (Chat Completions) - SHOULD inject
-
-    # We call common_processing_pre_call_logic
-    # It takes many args, but we only care about usage tracking injection
-    # For simplicity, we can mock the rest of the method or just isolate the block
-
-    # Actually, the block uses:
-    # general_settings.get("always_include_stream_usage", False)
-    # self.data.get("stream", False)
-    # route_type in ["acompletion", "atext_completion"]
-
-    # Since we can't easily call the async method without full setup,
-    # let's verify the logic by running the isolated block if possible,
-    # or just trust the A2A test for now.
-
-    # Wait, I can try to call it by mocking everything it needs.
-    pass
-
-
-def test_logic_verification():
-    # Manual verification of the whitelist logic
-    route_types = ["acompletion", "atext_completion", "aresponses", "arealtime", "auth"]
-    whitelist = ["acompletion", "atext_completion"]
-
-    results = {rt: (rt in whitelist) for rt in route_types}
-
-    assert results["acompletion"] is True
-    assert results["atext_completion"] is True
-    assert results["aresponses"] is False
-    assert results["arealtime"] is False
-    assert results["auth"] is False
+@pytest.mark.parametrize(
+    "route_type,expect_injected",
+    [
+        ("acompletion", True),
+        ("atext_completion", False),
+        ("aresponses", False),
+        ("aget_responses", False),
+        ("arealtime", False),
+    ],
+)
+def test_stream_usage_only_injected_for_whitelisted_route_types(route_type, expect_injected):
+    updates = _stream_usage_tracking_updates(
+        data={"stream": True, "model": "gpt-4o"},
+        general_settings={},
+        route_type=route_type,
+        supports_stream_options=lambda: True,
+    )
+    assert ("stream_options" in updates) is expect_injected