test: lock Anthropic contract regressions

Align TOOL_FORWARD_ENABLED docs with the current default and add count_tokens/auth/backpressure regressions so Anthropic compatibility stays stable. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-04-22 13:03:25 +08:00
parent 8b012310a2
commit f7fad97073
3 changed files with 100 additions and 2 deletions
--- a/DESIGN.md
+++ b/DESIGN.md
@@ -47,7 +47,7 @@

 - **逆向 Lingma 后端协议**：之前评估过（曾经的"B1 终极方案"），需要反编译二进制，维护成本高、政策风险大，放弃。
 - **多租户 / 水平扩缩**：单容器即可；真要大规模部署 → 套层反代 + N 个网关副本就够，不在进程内解决。
- **请求侧完整 function calling / tools 语义**：仍不是当前目标；现阶段仅支持 `tools`/`tool_choice` 在 `TOOL_FORWARD_ENABLED` 开关下灰度透传（默认关闭）。
+- **请求侧完整 function calling / tools 语义**：仍不是当前目标；现阶段仅支持 `tools`/`tool_choice` 在 `TOOL_FORWARD_ENABLED` 开关下灰度透传（默认开启，可显式关闭）。
 - **响应侧工具事件桥接**：若 Lingma 上游产出 tool 事件，网关会向 OpenAI 输出 `tool_calls`，向 Anthropic 输出 `tool_use` / `tool_result`（stream + non-stream）。
 - **强制工具回退闭环**：OpenAI 在 stream + non-stream 下都支持从文本里解析严格 JSON / `tool_code` 并合成 `tool_calls`；Anthropic 当前只在 non-stream 下合成 `tool_use` / `tool_result`，stream 仍保持原始文本流。

--- a/README.md
+++ b/README.md
@@ -5,7 +5,7 @@
 - OpenAI：`/v1/models`、`/v1/chat/completions`（含 stream）
 - Anthropic：`/v1/messages`、`/v1/messages/count_tokens`（含 stream）
 - 内置：多实例池、会话复用、Prometheus 指标、登录态 bundle 注入
- 工具事件桥接：Lingma 上游返回 `tool` 事件时，网关会输出为 OpenAI `tool_calls`（stream/non-stream）和 Anthropic `tool_use` / `tool_result`（stream/non-stream）；请求侧 `tools` / `tool_choice` 仅在 `TOOL_FORWARD_ENABLED=true` 时透传（默认关闭）
+- 工具事件桥接：Lingma 上游返回 `tool` 事件时，网关会输出为 OpenAI `tool_calls`（stream/non-stream）和 Anthropic `tool_use` / `tool_result`（stream/non-stream）；请求侧 `tools` / `tool_choice` 仅在 `TOOL_FORWARD_ENABLED=true` 时透传（默认开启，可显式关闭）
 - 多模态降级：OpenAI `image_url` / `input_image` 转 `[image]`，`input_audio` 转 `[audio]`；Anthropic `image` 转 `[image]`

 > 架构设计与二开细节请看 [`DESIGN.md`](./DESIGN.md)。
--- a/tests/test_tool_call_bridge.py
+++ b/tests/test_tool_call_bridge.py
@@ -1197,6 +1197,104 @@ class ToolCallBridgeTests(unittest.IsolatedAsyncioTestCase):
        self.assertEqual(fake_cache.keys, [])
        self.assertEqual(fake_cache.get_calls, [])
        self.assertEqual(fake_cache.put_calls, [])
+
+    async def test_anthropic_count_tokens_returns_input_tokens(self) -> None:
+        req = AnthropicMessagesRequest(
+            model="claude-3-5-sonnet-20241022",
+            max_tokens=64,
+            messages=[{"role": "user", "content": "count me"}],
+        )
+
+        with patch.object(main.settings, "api_keys", ["test-key"]):
+            response = await main.v1_messages_count_tokens(
+                req,
+                _make_request(
+                    "/v1/messages/count_tokens",
+                    headers={"x-api-key": "test-key", "anthropic-version": "2023-06-01"},
+                ),
+            )
+
+        payload = json.loads(response.body)
+        self.assertEqual(response.status_code, 200)
+        self.assertEqual(payload, {"input_tokens": main.estimate_tokens(main._messages_to_prompt(main.anthropic_to_internal_messages(req)))})
+
+    async def test_anthropic_count_tokens_requires_authentication(self) -> None:
+        req = AnthropicMessagesRequest(
+            model="claude-3-5-sonnet-20241022",
+            max_tokens=64,
+            messages=[{"role": "user", "content": "count me"}],
+        )
+
+        with patch.object(main.settings, "api_keys", ["test-key"]):
+            response = await main.v1_messages_count_tokens(
+                req,
+                _make_request(
+                    "/v1/messages/count_tokens",
+                    headers={"anthropic-version": "2023-06-01"},
+                ),
+            )
+
+        payload = json.loads(response.body)
+        self.assertEqual(response.status_code, 401)
+        self.assertEqual(payload["type"], "error")
+        self.assertEqual(payload["error"]["type"], "authentication_error")
+
+    async def test_anthropic_messages_requires_authentication(self) -> None:
+        req = AnthropicMessagesRequest(
+            model="claude-3-5-sonnet-20241022",
+            max_tokens=64,
+            messages=[{"role": "user", "content": "hi"}],
+            stream=False,
+        )
+
+        with patch.object(main.settings, "api_keys", ["test-key"]):
+            response = await main.v1_messages(
+                req,
+                _make_request(
+                    "/v1/messages",
+                    headers={"anthropic-version": "2023-06-01"},
+                ),
+            )
+
+        payload = json.loads(response.body)
+        self.assertEqual(response.status_code, 401)
+        self.assertEqual(payload["type"], "error")
+        self.assertEqual(payload["error"]["type"], "authentication_error")
+
+    async def test_anthropic_messages_backpressure_returns_overloaded_error(self) -> None:
+        fake_client = _FakeClient(stream_events=[], complete_result={"text": "ok", "toolEvents": []})
+        req = AnthropicMessagesRequest(
+            model="claude-3-5-sonnet-20241022",
+            max_tokens=64,
+            messages=[{"role": "user", "content": "hi"}],
+            stream=False,
+        )
+        fake_guard = types.SimpleNamespace(
+            in_flight=0,
+            try_acquire=AsyncMock(side_effect=main.BackpressureRejected(2.4)),
+        )
+
+        with (
+            patch.object(main, "pool", _FakePool(_FakeInstance(fake_client))),
+            patch.object(main, "chat_guard", fake_guard),
+            patch.object(main, "_ensure_instance_logged_in", AsyncMock(return_value={"id": "u"})),
+            patch.object(main.settings, "api_keys", ["test-key"]),
+        ):
+            response = await main.v1_messages(
+                req,
+                _make_request(
+                    "/v1/messages",
+                    headers={"x-api-key": "test-key", "anthropic-version": "2023-06-01"},
+                ),
+            )
+
+        payload = json.loads(response.body)
+        self.assertEqual(response.status_code, 429)
+        self.assertEqual(response.headers["Retry-After"], "2")
+        self.assertEqual(payload["type"], "error")
+        self.assertEqual(payload["error"]["type"], "overloaded_error")
+        self.assertIn("retry later", payload["error"]["message"])
+
    async def test_responses_non_stream_maps_chat_payload_shape_and_input(self) -> None:
        req = ResponsesRequest(
            model="org_auto",