feat: preserve governed OpenAI tool messages in 1.4.0

Add faithful single-choice chat results, tool stream activity, error-frame retries, and isolated protocol caching.

Reject empty cached completions with the same value predicate as transport, add regression coverage, and adopt Apache-2.0.

tests/e2e/conftest.py

@@ -22,7 +22,13 @@ from polygateway.client import (
 )
 from polygateway.providers import get_provider
 from polygateway.transports.openai_compat import OpenAICompatTransport
-from polygateway.types import Effort, EmbeddingTransportResult, SourceConfig, TransportResult
+from polygateway.types import (
+    Effort,
+    EmbeddingTransportResult,
+    ResponseMode,
+    SourceConfig,
+    TransportResult,
+)
 from tests.live_evidence import (
     AttemptEvidence,
     HttpEvidence,
@@ -293,6 +299,7 @@ class ObservedTransport:
         reasoning_effort: Effort | None,
         first_token_event: asyncio.Event | None,
         content_event: asyncio.Event | None,
+        response_mode: ResponseMode,
     ) -> TransportResult:
         """与生产端口逐参数同签名。"""
         with self._capture.attempt_context(call_id):
@@ -306,6 +313,7 @@ class ObservedTransport:
                 reasoning_effort=reasoning_effort,
                 first_token_event=first_token_event,
                 content_event=content_event,
+                response_mode=response_mode,
             )
 
     async def embed(

tests/e2e/test_openai_tools_live.py

@@ -0,0 +1,114 @@
+"""真实网关工具双轮冒烟；复用安全 Markdown 报告，不运行工具副作用。"""
+
+import dataclasses
+import json
+import os
+from pathlib import Path
+from uuid import uuid4
+
+import pytest
+from dotenv import dotenv_values
+
+from polygateway import GatewayClient, GatewaySettings
+from tests.live_evidence import write_live_round
+
+_ENV = {k: v for k, v in {**dotenv_values(".env"), **os.environ}.items() if v is not None}
+pytestmark = [pytest.mark.slow, pytest.mark.requires_llm]
+
+
+@pytest.mark.parametrize("stream", [True, False])
+async def test_tools_two_rounds(stream):
+    """已配置源中选首源，只改变测试隔离与 raw-only 意图，不收紧生产超时。"""
+    if not any(k.startswith("LLM__") and k.endswith("__API_KEY") for k in _ENV):
+        pytest.skip("未配置真实源，未覆盖")
+    settings = GatewaySettings.from_env("LLM", env=_ENV)
+    source = dataclasses.replace(settings.sources[0], enable_thinking=None, reasoning_effort=None)
+    settings = dataclasses.replace(
+        settings,
+        sources=(source,),
+        cache_backend="none",
+        cache_namespace=None,
+        cache_ttl_s=None,
+        telemetry_backend="none",
+        telemetry_auto_migrate=False,
+        limiter_backend="memory",
+        breaker_backend="memory",
+        hedge_after_s=None,
+        hedge_stall_after_s=None,
+    )
+    tools = [
+        {
+            "type": "function",
+            "function": {
+                "name": "echo_value",
+                "description": "Return the supplied value",
+                "parameters": {
+                    "type": "object",
+                    "properties": {"value": {"type": "string"}},
+                    "required": ["value"],
+                },
+            },
+        }
+    ]
+    messages = [
+        {
+            "role": "user",
+            "content": "Call echo_value with value hello. After receiving the tool result, reply with hello.",
+        }
+    ]
+    status, error_type, counts = "FAIL", None, {}
+    try:
+        async with GatewayClient.from_settings(settings) as client:
+            first = await client.chat(
+                messages,
+                stream=stream,
+                response_mode="openai",
+                overlay={
+                    "tools": tools,
+                    "tool_choice": {"type": "function", "function": {"name": "echo_value"}},
+                },
+            )
+            completion = first.completion
+            assert completion is not None and completion.message["role"] == "assistant"
+            assert completion.response_id and completion.finish_reason == "tool_calls"
+            calls = completion.message["tool_calls"]
+            counts["tools"] = len(calls)
+            assert calls
+            messages.append(completion.message)
+            for call in calls:
+                assert call["function"]["name"] == "echo_value"
+                arguments = json.loads(call["function"]["arguments"])
+                assert arguments["value"] == "hello"
+                messages.append(
+                    {"role": "tool", "tool_call_id": call["id"], "content": arguments["value"]}
+                )
+            second = await client.chat(
+                messages,
+                stream=stream,
+                response_mode="openai",
+                overlay={"tools": tools, "tool_choice": "none"},
+            )
+            assert (
+                second.completion is not None and second.completion.message["role"] == "assistant"
+            )
+            assert "hello" in (second.completion.message.get("content") or "").lower()
+            counts["attempts"] = first.call_stats.attempts + second.call_stats.attempts
+            status = "PASS"
+    except Exception as exc:
+        error_type = type(exc).__name__
+        raise
+    finally:
+        write_live_round(
+            Path("tests/outputs/issue29/live"),
+            run_id=uuid4().hex,
+            matrix_id="tools",
+            round_index=1,
+            safe_fields={
+                "status": status,
+                "stream": stream,
+                "error_type": error_type,
+                "counts": counts,
+                "planned_rounds": 2,
+                "completed_rounds": 2 if status == "PASS" else 0,
+            },
+        )