test: 清理 backlog——chat_structured 超限检测 + client 上下文管理器

_AI_USAGE_LOG.md

@@ -50,3 +50,4 @@
 | 2026-08-09 10:40 | 反馈迭代 | 里程碑3.1 最终全分支评审（base 2fbff07..8bc5e5e）修正：评审查到 client 两处缺陷并修复——① 2xx 但响应结构损坏（非 JSON/缺 choices 字段）从裸 KeyError/JSONDecodeError 逃逸出 engine 的 except LLMError，改为捕获 (json.JSONDecodeError, KeyError, IndexError, TypeError) 后抛 LLMResponseError（spec §3.8 既有类型首次被触发，封闭 api-design §7 LLM_PARSE_ERROR 来源列虚指）；② 3xx 重定向被当作成功（httpx 不自动跟随），补显式 2xx 判定，非 2xx 一律 LLMNetworkError。TDD：新增 2 用例（test_chat_3xx_no_retry、test_chat_malformed_response_raises_llm_response_error 含 not-json/missing-key 两种 handler）先 RED→实现→GREEN；聚焦 9 passed；pytest 全量 117 passed 覆盖 100.00%（785 stmts/186 br），fail_under=99 达标 | src/genesis/inference/client.py, tests/test_inference_client.py, _AI_USAGE_LOG.md | deepseek-v4-flash-free |
 
 (End of file - total 50 lines)
+| 2026-08-09 10:50 | 反馈迭代 | 里程碑3.1 backlog 清理（retro 3 项）：① chat_structured 增加 token 超限检测（渲染后估算超限→truncate_cb→重渲染，与 chat 流程一致）TDD RED→GREEN 新增 test_chat_structured_truncation_callback_triggered；② 清理 test_inference_engine.py 未用导入（pytest/json/ChatMessage）；③ HttpLLMClient 增加上下文管理器（__enter__/__exit__ 退出时关闭底层 httpx.Client 释放连接）TDD 新增 test_client_context_manager_closes_transport；pytest 全量 119 passed 覆盖 100.00%，fail_under=99 达标；提交见 git log | src/genesis/inference/engine.py, src/genesis/inference/client.py, tests/test_inference_engine.py, tests/test_inference_client.py, _AI_USAGE_LOG.md | deepseek-v4-flash-free |

src/genesis/inference/client.py

@@ -48,6 +48,13 @@ class HttpLLMClient:
         self._retry_backoff = retry_backoff
         self._client = httpx.Client(timeout=timeout_sec, transport=transport)
 
+    def __enter__(self) -> HttpLLMClient:
+        """支持 with 块：退出时自动关闭底层连接。"""
+        return self
+
+    def __exit__(self, exc_type, exc_val, exc_tb) -> None:
+        self._client.close()
+
     def chat(
         self,
         *,

src/genesis/inference/engine.py

@@ -133,6 +133,9 @@ class InferenceEngine:
         retry_count: int = 2,
     ) -> StructuredResult:
         rendered = self._render_prompt(prompt, variables)
+        if self._estimator(rendered) > self._max_context_tokens:
+            variables = self._apply_truncation(rendered, variables)
+            rendered = self._render_prompt(prompt, variables)
         # 追加 schema 约束说明（不强制模板支持）
         schema_hint = json.dumps(schema, ensure_ascii=False) if schema else ""
         base_rendered = rendered + (f'\n\n请输出符合以下 JSON Schema 的 JSON：{schema_hint}' if schema_hint else "")

tests/test_inference_client.py

@@ -54,6 +54,24 @@ def test_chat_requires_api_key():
         HttpLLMClient(base_url="https://x", api_key="")
 
 
+def test_client_context_manager_closes_transport():
+    # with 块退出后底层 httpx.Client 应被关闭（释放连接）
+    with HttpLLMClient(
+        base_url="https://api.test.local",
+        api_key="sk-test",
+        transport=httpx.MockTransport(_ok_handler),
+    ) as client:
+        assert isinstance(client, HttpLLMClient)
+        text, _ = client.chat(
+            model="deepseek-chat",
+            messages=[ChatMessage(role="user", content="hi")],
+            temperature=0.2,
+            max_tokens=100,
+        )
+        assert text == "Hello"
+    assert client._client.is_closed is True
+
+
 def test_chat_timeout():
     def slow(request):
         raise httpx.ReadTimeout("slow")

tests/test_inference_engine.py

@@ -1,14 +1,10 @@
 from __future__ import annotations
 
-import json
-
-import pytest
-
 from genesis.inference.engine import InferenceEngine
 from genesis.inference.exceptions import LLMError
 from genesis.inference.prompt_registry import PromptRegistry
 from genesis.inference.token import approximate_token_count
-from genesis.inference.types import ChatMessage, Prompt
+from genesis.inference.types import Prompt
 from tests.inference_helpers import FakeLLMClient
 
 
@@ -161,6 +157,32 @@ def test_chat_structured_failed_on_network():
     assert r.status == "failed"
 
 
+def test_chat_structured_truncation_callback_triggered():
+    # chat_structured 超限时同样触发 truncate_cb（与 chat 流程一致）
+    seen = {}
+
+    def truncate_cb(prompt_text, variables):
+        seen["called"] = True
+        return {**variables, "chapter": "裁剪版"}
+
+    eng = InferenceEngine(
+        client=FakeLLMClient([("ok", '{"a": 1}')]),
+        models=Models(),
+        registry=PromptRegistry(),
+        estimator=approximate_token_count,
+        truncate_cb=truncate_cb,
+    )
+    eng._max_context_tokens = 2  # 强制超限（渲染约 3 token > 2）
+    r = eng.chat_structured(
+        session_id="s1",
+        prompt=Prompt(name="p", version="v1", template="abcd{{ chapter }}"),
+        variables={"chapter": "很长很长的标题"},
+        schema={"type": "object", "properties": {"a": {"type": "number"}}},
+    )
+    assert seen["called"] is True
+    assert r.status == "ok"
+
+
 # ---------- 补充分支覆盖（defensive / 缺失配置） ----------
 
 def test_chat_fallback_all_failed_when_only_primary():