coding-proxy 0.3.1a3__tar.gz → 0.3.1a4__tar.gz

{coding_proxy-0.3.1a3 → coding_proxy-0.3.1a4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: coding-proxy
-Version: 0.3.1a3
+Version: 0.3.1a4
 Summary: A High-Availability, Transparent, and Smart Multi-Vendor Proxy for Claude Code. Support Claude Plans, GitHub Copilot, Google Antigravity, ZAI/GLM, MiniMax, Qwen, Xiaomi, Kimi, Doubao...
 Project-URL: Source Code, https://github.com/ThreeFish-AI/coding-proxy
 Project-URL: User Guide, https://github.com/ThreeFish-AI/coding-proxy/blob/master/docs/user-guide.md

{coding_proxy-0.3.1a3 → coding_proxy-0.3.1a4}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "coding-proxy"
-version = "0.3.1a3"
+version = "0.3.1a4"
 description = "A High-Availability, Transparent, and Smart Multi-Vendor Proxy for Claude Code. Support Claude Plans, GitHub Copilot, Google Antigravity, ZAI/GLM, MiniMax, Qwen, Xiaomi, Kimi, Doubao..."
 readme = "README.md"
 requires-python = ">=3.12"

{coding_proxy-0.3.1a3 → coding_proxy-0.3.1a4}/src/coding/proxy/convert/anthropic_to_openai.py

@@ -284,24 +284,34 @@ def _translate_assistant_message(message: dict[str, Any]) -> list[dict[str, Any]
         final_text_parts = text_parts
 
     if tool_uses:
+        tool_calls: list[dict[str, Any]] = []
+        for block in tool_uses:
+            raw_input = block.get("input")
+            if not isinstance(raw_input, dict):
+                logger.debug(
+                    "copilot: tool_use id=%s name=%s has non-dict input (type=%s), "
+                    "defaulting to empty dict",
+                    block.get("id", ""),
+                    block.get("name", ""),
+                    type(raw_input).__name__,
+                )
+                raw_input = {}
+            tool_calls.append(
+                {
+                    "id": block.get("id", ""),
+                    "type": "function",
+                    "function": {
+                        "name": block.get("name", ""),
+                        "arguments": json.dumps(raw_input, ensure_ascii=False),
+                    },
+                }
+            )
         return [
             {
                 "role": "assistant",
                 "content": "\n\n".join(part for part in final_text_parts if part)
                 or None,
-                "tool_calls": [
-                    {
-                        "id": block.get("id", ""),
-                        "type": "function",
-                        "function": {
-                            "name": block.get("name", ""),
-                            "arguments": json.dumps(
-                                block.get("input", {}), ensure_ascii=False
-                            ),
-                        },
-                    }
-                    for block in tool_uses
-                ],
+                "tool_calls": tool_calls,
             }
         ]
 

{coding_proxy-0.3.1a3 → coding_proxy-0.3.1a4}/src/coding/proxy/routing/executor.py

@@ -135,6 +135,10 @@ def _is_likely_request_format_error(
         # 非结构化响应体（非 JSON）
         if not trimmed.startswith("{") and len(trimmed) < 200:
             return True
+        # 结构化 JSON 400 但含 tool_call 格式错误码 → 格式不兼容
+        # （如 Copilot 返回 {"error":{"code":"invalid_tool_call_format",...}}）
+        if "invalid_tool_call_format" in trimmed:
+            return True
     return False
 
 
@@ -795,6 +799,27 @@ class _RouteExecutor:
                     tier.name,
                 )
 
+            # 补充检测：zhipu 500 — tool_result 块触发上游 AttributeError
+            # zhipu 后端在 tool_result 块上错误访问 .id 属性（应为 .tool_use_id），
+            # 此为已知的上游格式缺陷，应视为 format incompatibility 而非真实服务器故障。
+            if (
+                not semantic_rejection
+                and exc.response.status_code == 500
+                and request_body is not None
+                and _has_tool_results(request_body)
+            ):
+                err_text = (exc.response.text or "")[:500]
+                if (
+                    "'ClaudeContentBlockToolResult'" in err_text
+                    and "has no attribute 'id'" in err_text
+                ):
+                    semantic_rejection = True
+                    logger.warning(
+                        "Tier %s zhipu tool_result format error (500), "
+                        "treating as format incompatibility without circuit breaker penalty",
+                        tier.name,
+                    )
+
             if semantic_rejection and not is_last:
                 return True, tier.name, exc
 

{coding_proxy-0.3.1a3 → coding_proxy-0.3.1a4}/src/coding/proxy/routing/usage_parser.py

@@ -210,6 +210,9 @@ def parse_usage_from_chunk(
                 request_id=data.get("id"),
                 model_served=data.get("model"),
             )
+            model_name = data.get("model")
+            if model_name:
+                usage["model_served"] = model_name
 
         # Gemini SSE 格式: data.usageMetadata.{promptTokenCount, candidatesTokenCount, cachedContentTokenCount, thoughtsTokenCount, toolUsePromptTokenCount}
         # Gemini 的流式响应在最后一帧（或每一帧）携带 usageMetadata；字段命名与
@@ -243,6 +246,9 @@ def parse_usage_from_chunk(
                     request_id=data.get("responseId") or data.get("id"),
                     model_served=data.get("modelVersion") or data.get("model"),
                 )
+                model_name = data.get("modelVersion") or data.get("model")
+                if model_name:
+                    usage["model_served"] = model_name
 
         # request_id fallback (OpenAI 格式下 id 在顶层, Gemini 顶层为 responseId)
         if not usage.get("request_id"):

{coding_proxy-0.3.1a3 → coding_proxy-0.3.1a4}/tests/test_copilot_convert_request.py

@@ -472,3 +472,132 @@ def test_image_block_converted_to_image_url():
     image_part = [p for p in user_msg["content"] if p.get("type") == "image_url"]
     assert len(image_part) == 1
     assert "data:image/png;base64,abc123" in image_part[0]["image_url"]["url"]
+
+
+# === Defensive tool_use.input serialization ===
+
+
+def test_tool_use_input_none_defaults_to_empty_dict():
+    """input=None 应被降级为 {} 而非序列化为 'null'."""
+    body = {
+        "model": "claude-sonnet-4-20250514",
+        "messages": [
+            {
+                "role": "assistant",
+                "content": [
+                    {
+                        "type": "tool_use",
+                        "id": "toolu_001",
+                        "name": "read_file",
+                        "input": None,
+                    }
+                ],
+            }
+        ],
+    }
+    result = convert_request(body)
+    assistant_msgs = [m for m in result["messages"] if m["role"] == "assistant"]
+    assert len(assistant_msgs) == 1
+    assert "tool_calls" in assistant_msgs[0]
+    tc = assistant_msgs[0]["tool_calls"][0]
+    assert tc["function"]["arguments"] == "{}"
+
+
+def test_tool_use_input_string_defaults_to_empty_dict():
+    """input='some string' 应被降级为 {} 而非序列化为 '"some string"'."""
+    body = {
+        "model": "claude-sonnet-4-20250514",
+        "messages": [
+            {
+                "role": "assistant",
+                "content": [
+                    {
+                        "type": "tool_use",
+                        "id": "toolu_002",
+                        "name": "run_cmd",
+                        "input": "not a dict",
+                    }
+                ],
+            }
+        ],
+    }
+    result = convert_request(body)
+    assistant_msgs = [m for m in result["messages"] if m["role"] == "assistant"]
+    tc = assistant_msgs[0]["tool_calls"][0]
+    assert tc["function"]["arguments"] == "{}"
+
+
+def test_tool_use_input_missing_defaults_to_empty_dict():
+    """input key 不存在时，block.get('input') 返回 None，应降级为 {}."""
+    body = {
+        "model": "claude-sonnet-4-20250514",
+        "messages": [
+            {
+                "role": "assistant",
+                "content": [
+                    {
+                        "type": "tool_use",
+                        "id": "toolu_003",
+                        "name": "search",
+                    }
+                ],
+            }
+        ],
+    }
+    result = convert_request(body)
+    assistant_msgs = [m for m in result["messages"] if m["role"] == "assistant"]
+    tc = assistant_msgs[0]["tool_calls"][0]
+    assert tc["function"]["arguments"] == "{}"
+
+
+def test_tool_use_input_int_defaults_to_empty_dict():
+    """input=42 应被降级为 {} 而非序列化为 '42'."""
+    body = {
+        "model": "claude-sonnet-4-20250514",
+        "messages": [
+            {
+                "role": "assistant",
+                "content": [
+                    {
+                        "type": "tool_use",
+                        "id": "toolu_004",
+                        "name": "calc",
+                        "input": 42,
+                    }
+                ],
+            }
+        ],
+    }
+    result = convert_request(body)
+    assistant_msgs = [m for m in result["messages"] if m["role"] == "assistant"]
+    tc = assistant_msgs[0]["tool_calls"][0]
+    assert tc["function"]["arguments"] == "{}"
+
+
+def test_tool_use_valid_dict_input_preserved():
+    """正常 dict input 应保持原样."""
+    body = {
+        "model": "claude-sonnet-4-20250514",
+        "messages": [
+            {
+                "role": "assistant",
+                "content": [
+                    {
+                        "type": "tool_use",
+                        "id": "toolu_005",
+                        "name": "read_file",
+                        "input": {"path": "/tmp/test.txt", "offset": 10},
+                    }
+                ],
+            }
+        ],
+    }
+    result = convert_request(body)
+    assistant_msgs = [m for m in result["messages"] if m["role"] == "assistant"]
+    tc = assistant_msgs[0]["tool_calls"][0]
+    import json
+
+    assert json.loads(tc["function"]["arguments"]) == {
+        "path": "/tmp/test.txt",
+        "offset": 10,
+    }

{coding_proxy-0.3.1a3 → coding_proxy-0.3.1a4}/tests/test_parse_usage.py

@@ -117,6 +117,34 @@ def test_openai_zhipu_final_chunk():
     assert usage["input_tokens"] == 200
     assert usage["output_tokens"] == 80
     assert usage["request_id"] == "chatcmpl-1"
+    assert usage["model_served"] == "glm-5.1"
+
+
+def test_openai_final_chunk_with_model():
+    """OpenAI 最终 chunk 有 model 字段时应提取到 model_served."""
+    usage: dict = {}
+    parse_usage_from_chunk(
+        _sse(
+            '{"id":"chatcmpl-2","model":"gpt-4o-2024-08-06",'
+            '"usage":{"prompt_tokens":50,"completion_tokens":20}}'
+        ),
+        usage,
+    )
+    assert usage["input_tokens"] == 50
+    assert usage["output_tokens"] == 20
+    assert usage["model_served"] == "gpt-4o-2024-08-06"
+
+
+def test_openai_final_chunk_without_model():
+    """OpenAI 最终 chunk 无 model 字段时不应设置 model_served."""
+    usage: dict = {}
+    parse_usage_from_chunk(
+        _sse('{"id":"chatcmpl-3","usage":{"prompt_tokens":30,"completion_tokens":10}}'),
+        usage,
+    )
+    assert usage["input_tokens"] == 30
+    assert usage["output_tokens"] == 10
+    assert "model_served" not in usage
 
 
 def test_openai_final_chunk_with_cache_tokens():

{coding_proxy-0.3.1a3 → coding_proxy-0.3.1a4}/tests/test_parse_usage_gemini.py

@@ -35,6 +35,7 @@ def test_gemini_usage_metadata_basic_fields():
     assert usage["output_tokens"] == 42
     assert usage.get("cache_read_tokens", 0) == 0
     assert usage["request_id"] == "resp_abc"
+    assert usage["model_served"] == "gemini-2.0-flash"
 
 
 def test_gemini_usage_metadata_with_cached_content():
@@ -194,3 +195,18 @@ def test_gemini_partial_fields_ok():
     )
     assert usage["input_tokens"] == 77
     assert "output_tokens" not in usage
+
+
+def test_gemini_model_fallback_to_data_model():
+    """当 modelVersion 不存在时，应回退到 data.model."""
+    usage: dict = {}
+    parse_usage_from_chunk(
+        _sse(
+            '{"usageMetadata":{"promptTokenCount":80,"candidatesTokenCount":20},'
+            '"model":"gemini-1.5-flash"}'
+        ),
+        usage,
+    )
+    assert usage["input_tokens"] == 80
+    assert usage["output_tokens"] == 20
+    assert usage["model_served"] == "gemini-1.5-flash"

{coding_proxy-0.3.1a3 → coding_proxy-0.3.1a4}/tests/test_router_executor.py

@@ -11,6 +11,7 @@ from __future__ import annotations
 
 from unittest.mock import AsyncMock, MagicMock
 
+import httpx
 import pytest
 
 from coding.proxy.compat.canonical import (
@@ -1229,6 +1230,31 @@ class TestIsLikelyRequestFormatError:
             is False
         )
 
+    def test_returns_true_for_invalid_tool_call_format(self):
+        """400 + 结构化 JSON 含 invalid_tool_call_format + tool_result → 格式不兼容."""
+        json_body = '{"error":{"message":"Invalid JSON format in tool call arguments","code":"invalid_tool_call_format"}}'
+        assert (
+            _is_likely_request_format_error(
+                status_code=400,
+                error_body_text=json_body,
+                body=self._body_with_tool_results(),
+            )
+            is True
+        )
+
+    def test_returns_false_for_invalid_tool_call_format_without_tool_results(self):
+        """invalid_tool_call_format 但无 tool_result → 不应匹配."""
+        json_body = '{"error":{"message":"Invalid JSON format in tool call arguments","code":"invalid_tool_call_format"}}'
+        body = {"model": "test", "messages": [{"role": "user", "content": "hi"}]}
+        assert (
+            _is_likely_request_format_error(
+                status_code=400,
+                error_body_text=json_body,
+                body=body,
+            )
+            is False
+        )
+
 
 # ── TokenAcquireError 永久性凭证错误测试 ────────────────────
 
@@ -2101,3 +2127,157 @@ class TestPrepareBodyForTierSelfTransition:
             b for b in result["messages"][0]["content"] if b.get("type") == "thinking"
         )
         assert thinking_block["signature"] == "zhipu_sig"
+
+
+# ── zhipu 500 tool_result 格式错误检测测试 ──────────────────────
+
+
+class TestZhipu500ToolResultFormatError:
+    """验证 _handle_http_error 对 zhipu 500 'ClaudeContentBlockToolResult' 错误的处理.
+
+    zhipu 后端在 tool_result 块上错误访问 .id 属性（应为 .tool_use_id），
+    此为已知的上游格式缺陷，应视为 format incompatibility（semantic rejection）
+    而非真实服务器故障，不应计入熔断器。
+    """
+
+    @pytest.mark.asyncio
+    async def test_zhipu_500_tool_result_error_triggers_semantic_rejection(self):
+        """zhipu 500 + 'ClaudeContentBlockToolResult' + tool_result → semantic rejection."""
+        from coding.proxy.routing.circuit_breaker import CircuitBreaker
+
+        vendor = _mock_vendor("zhipu")
+        error_body = (
+            b'{"error":{"code":"500","message":"\'ClaudeContentBlockToolResult\' '
+            b"object has no attribute 'id'\"}}"
+        )
+        response = httpx.Response(
+            status_code=500,
+            content=error_body,
+            request=httpx.Request("POST", "https://example.com"),
+        )
+        exc = httpx.HTTPStatusError(
+            "zhipu API error: 500", request=response.request, response=response
+        )
+
+        cb = CircuitBreaker(failure_threshold=3)
+        tier = _make_tier(vendor, circuit_breaker=cb)
+        exec_inst = _executor([tier, _make_tier(_mock_vendor("copilot"))])
+
+        body = {
+            "model": "claude-opus-4-6",
+            "messages": [
+                {
+                    "role": "user",
+                    "content": [
+                        {
+                            "type": "tool_result",
+                            "tool_use_id": "tu_1",
+                            "content": "result",
+                        }
+                    ],
+                },
+            ],
+        }
+
+        should_continue, failed_name, _ = await exec_inst._handle_http_error(
+            tier,
+            exc,
+            is_last=False,
+            failed_tier_name=None,
+            last_exc=None,
+            is_stream=True,
+            request_body=body,
+        )
+
+        assert should_continue is True
+        assert failed_name == "zhipu"
+        # 不应计入熔断器
+        assert cb.get_info()["failure_count"] == 0
+
+    @pytest.mark.asyncio
+    async def test_zhipu_500_generic_error_records_failure(self):
+        """zhipu 500 但非 tool_result 格式错误 → 正常记录熔断器."""
+        from coding.proxy.routing.circuit_breaker import CircuitBreaker
+
+        vendor = _mock_vendor("zhipu")
+        error_body = b'{"error":{"code":"500","message":"Internal Server Error"}}'
+        response = httpx.Response(
+            status_code=500,
+            content=error_body,
+            request=httpx.Request("POST", "https://example.com"),
+        )
+        exc = httpx.HTTPStatusError(
+            "zhipu API error: 500", request=response.request, response=response
+        )
+
+        cb = CircuitBreaker(failure_threshold=3)
+        tier = _make_tier(vendor, circuit_breaker=cb)
+        exec_inst = _executor([tier])
+
+        body = {
+            "model": "claude-opus-4-6",
+            "messages": [
+                {
+                    "role": "user",
+                    "content": [
+                        {
+                            "type": "tool_result",
+                            "tool_use_id": "tu_1",
+                            "content": "result",
+                        }
+                    ],
+                },
+            ],
+        }
+
+        should_continue, _, _ = await exec_inst._handle_http_error(
+            tier,
+            exc,
+            is_last=True,
+            failed_tier_name=None,
+            last_exc=None,
+            is_stream=True,
+            request_body=body,
+        )
+
+        # 非 last tier 时 should_continue=False，且应记录熔断器失败
+        assert should_continue is False
+        assert cb.get_info()["failure_count"] == 1
+
+    @pytest.mark.asyncio
+    async def test_zhipu_500_tool_result_error_without_tool_results_body(self):
+        """zhipu 500 tool_result 错误但请求体无 tool_result → 不触发特殊处理."""
+        from coding.proxy.routing.circuit_breaker import CircuitBreaker
+
+        vendor = _mock_vendor("zhipu")
+        error_body = (
+            b'{"error":{"code":"500","message":"\'ClaudeContentBlockToolResult\' '
+            b"object has no attribute 'id'\"}}"
+        )
+        response = httpx.Response(
+            status_code=500,
+            content=error_body,
+            request=httpx.Request("POST", "https://example.com"),
+        )
+        exc = httpx.HTTPStatusError(
+            "zhipu API error: 500", request=response.request, response=response
+        )
+
+        cb = CircuitBreaker(failure_threshold=3)
+        tier = _make_tier(vendor, circuit_breaker=cb)
+        exec_inst = _executor([tier])
+
+        body = {"model": "test", "messages": [{"role": "user", "content": "hello"}]}
+
+        should_continue, _, _ = await exec_inst._handle_http_error(
+            tier,
+            exc,
+            is_last=True,
+            failed_tier_name=None,
+            last_exc=None,
+            is_stream=True,
+            request_body=body,
+        )
+
+        assert should_continue is False
+        assert cb.get_info()["failure_count"] == 1

{coding_proxy-0.3.1a3 → coding_proxy-0.3.1a4}/uv.lock

@@ -74,7 +74,7 @@ wheels = [
 
 [[package]]
 name = "coding-proxy"
-version = "0.3.1a3"
+version = "0.3.1a4"
 source = { editable = "." }
 dependencies = [
     { name = "aiosqlite" },