coding-proxy 0.3.1a2__tar.gz → 0.3.1a4__tar.gz

{coding_proxy-0.3.1a2 → coding_proxy-0.3.1a4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: coding-proxy
-Version: 0.3.1a2
+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-0.3.1a2 → coding_proxy-0.3.1a4}/docs/issue.md

@@ -4,6 +4,50 @@
 
 ---
 
+## streaming usage parse failed: 'NoneType' object has no attribute 'get'
+
+**问题描述**
+
+OpenAI 兼容 SSE 流式响应过程中，单次请求日志反复刷出数十条 WARNING：
+
+```
+WARNING streaming usage parse failed: 'NoneType' object has no attribute 'get'
+```
+
+警告本身被上层 `try/except` 吞掉不影响主链路，但日志噪声严重，且每帧都丢失了 usage 累加。
+
+**表因**
+
+`StreamingUsageAccumulator.feed` 调用 `parse_usage_from_chunk` 解析 SSE chunk 时抛出 `AttributeError`。
+
+**根因**
+
+`src/coding/proxy/routing/usage_parser.py::parse_usage_from_chunk` 中 Anthropic message_start 与 Anthropic message_delta / OpenAI 两条分支都使用了脆弱的判空模式：
+
+```python
+if "usage" in data:        # 仅判断 key 存在
+    u = data["usage"]      # 但值可能是 null
+    u.get("output_tokens", 0)  # AttributeError
+```
+
+部分上游（含某些 OpenAI 兼容供应商）在中间 chunk 显式发送 `"usage": null` 占位帧，`in` 检查通过但取出的是 `None`。
+
+**处理方式**
+
+将两处 guard 统一改为 `u = container.get("usage"); if isinstance(u, dict):`，既排除缺省也排除 null，并顺手移除内部冗余的 `if isinstance(u, dict):` 包装层（已被外层 guard 覆盖）。同时新增三个回归用例覆盖 `data.usage = null` / `message.usage = null` / null 帧后跟有效帧三种场景。
+
+**后续防范**
+
+- 解析外部 SSE / JSON 结构时, 不要单独使用 `if key in data` 作为安全 guard, 应统一采用 `value = data.get(key); if isinstance(value, dict):` 的双重保护, 同时排除缺省与显式 null。
+- 对 try/except 包裹的 WARNING 路径要保持警觉: 异常被吞不代表无害，重复刷屏的同类警告往往暗示防御性 guard 过窄，需要回溯至根因修复，而非依赖 except 兜底。
+
+**同类问题影响与处理注意事项**
+
+- 本仓库内 `parse_usage_from_chunk` 的 Gemini `usageMetadata` 分支 (line ~219) 已经使用 `isinstance(um, dict)` 防御, 不受影响, 可作为参考实现。
+- 检查其他解析器 (如 routing / vendor adapter 层) 是否还有 `if "key" in data: v = data["key"]; v.get(...)` 这种模式, 必要时同步加固。
+
+---
+
 ## zhipu 自循环 400 + tool_results 偶发降级
 
 **问题描述**

{coding_proxy-0.3.1a2 → coding_proxy-0.3.1a4}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "coding-proxy"
-version = "0.3.1a2"
+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-0.3.1a2 → 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-0.3.1a2 → 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-0.3.1a2 → coding_proxy-0.3.1a4}/src/coding/proxy/routing/usage_parser.py

@@ -169,8 +169,8 @@ def parse_usage_from_chunk(
 
         # Anthropic 格式: message_start 事件 (data.message.usage)
         msg = data.get("message", {})
-        if isinstance(msg, dict) and "usage" in msg:
-            u = msg["usage"]
+        u = msg.get("usage") if isinstance(msg, dict) else None
+        if isinstance(u, dict):
             input_tokens = u.get("input_tokens", 0) or u.get("prompt_tokens", 0)
             _set_if_nonzero(usage, "input_tokens", input_tokens)
             _set_if_nonzero(
@@ -183,18 +183,17 @@ def parse_usage_from_chunk(
                 usage["request_id"] = msg["id"]
             if "model" in msg:
                 usage["model_served"] = msg["model"]
-            if isinstance(u, dict):
-                _append_usage_evidence(
-                    usage,
-                    evidence_kind="message_usage",
-                    raw_usage=dict(u),
-                    request_id=msg.get("id"),
-                    model_served=msg.get("model"),
-                )
+            _append_usage_evidence(
+                usage,
+                evidence_kind="message_usage",
+                raw_usage=dict(u),
+                request_id=msg.get("id"),
+                model_served=msg.get("model"),
+            )
 
         # Anthropic message_delta / OpenAI 最后一个 chunk (data.usage)
-        if "usage" in data:
-            u = data["usage"]
+        u = data.get("usage")
+        if isinstance(u, dict):
             output_tokens = u.get("output_tokens", 0) or u.get("completion_tokens", 0)
             input_tokens = u.get("input_tokens", 0) or u.get("prompt_tokens", 0)
             cache_creation_tokens = u.get("cache_creation_input_tokens", 0)
@@ -204,14 +203,16 @@ def parse_usage_from_chunk(
             _set_if_nonzero(usage, "input_tokens", input_tokens)
             _set_if_nonzero(usage, "cache_creation_tokens", cache_creation_tokens)
             _set_if_nonzero(usage, "cache_read_tokens", cache_read_tokens)
-            if isinstance(u, dict):
-                _append_usage_evidence(
-                    usage,
-                    evidence_kind="data_usage",
-                    raw_usage=dict(u),
-                    request_id=data.get("id"),
-                    model_served=data.get("model"),
-                )
+            _append_usage_evidence(
+                usage,
+                evidence_kind="data_usage",
+                raw_usage=dict(u),
+                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
@@ -245,6 +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-0.3.1a2 → coding_proxy-0.3.1a4}/src/coding/proxy/server/dashboard.py

@@ -415,7 +415,7 @@ _DASHBOARD_HTML = """<!DOCTYPE html>
       z-index: 1000;
       opacity: 0;
       transition: opacity .15s ease;
-      max-width: 360px;
+      max-width: 520px;
       max-height: 60vh;
       overflow-y: auto;
       backdrop-filter: blur(8px);
@@ -429,8 +429,8 @@ _DASHBOARD_HTML = """<!DOCTYPE html>
     #chart-tooltip-items { display: flex; flex-direction: column; gap: 3px; }
     .tt-item { display: flex; align-items: center; gap: 8px; line-height: 1.4; }
     .tt-color { width: 8px; height: 8px; border-radius: 50%; flex-shrink: 0; }
-    .tt-label { flex: 1; color: var(--text-primary); }
-    .tt-value { font-family: 'JetBrains Mono', monospace; font-size: 12px; color: var(--text-secondary); }
+    .tt-label { flex: 1; color: var(--text-primary); white-space: nowrap; }
+    .tt-value { font-family: 'JetBrains Mono', monospace; font-size: 12px; color: var(--text-secondary); white-space: nowrap; }
     #chart-tooltip-footer {
       margin-top: 6px; padding-top: 6px; border-top: 1px solid var(--border-subtle);
       font-weight: 500; font-size: 12px; color: var(--text-secondary);

{coding_proxy-0.3.1a2 → 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-0.3.1a2 → 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():
@@ -265,3 +293,48 @@ def test_multiple_sse_lines_in_single_chunk():
     parse_usage_from_chunk(chunk, usage)
     assert usage["input_tokens"] == 80
     assert usage["output_tokens"] == 20
+
+
+# --- null usage 安全保护（防御上游 SSE 字段为 null 的极端格式） ---
+
+
+def test_null_usage_at_top_level_does_not_raise():
+    """data.usage 显式为 null 时应被静默忽略，不抛异常、不产生 WARNING."""
+    usage: dict = {}
+    parse_usage_from_chunk(
+        _sse('{"id":"chatcmpl-null","choices":[{"delta":{}}],"usage":null}'),
+        usage,
+    )
+    # 不应写入任何 token 字段
+    assert usage.get("input_tokens", 0) == 0
+    assert usage.get("output_tokens", 0) == 0
+
+
+def test_null_usage_in_message_does_not_raise():
+    """message.usage 显式为 null 时应被静默忽略."""
+    usage: dict = {}
+    parse_usage_from_chunk(
+        _sse('{"type":"message_start","message":{"id":"msg_null","usage":null}}'),
+        usage,
+    )
+    assert usage.get("input_tokens", 0) == 0
+
+
+def test_null_usage_does_not_break_subsequent_valid_chunks():
+    """null usage 帧之后到来的有效帧仍能正确解析."""
+    usage: dict = {}
+    # 1. null usage 帧
+    parse_usage_from_chunk(
+        _sse('{"id":"chatcmpl-1","choices":[{"delta":{"content":"hi"}}],"usage":null}'),
+        usage,
+    )
+    # 2. 有效的最终帧
+    parse_usage_from_chunk(
+        _sse(
+            '{"id":"chatcmpl-1","choices":[{"finish_reason":"stop","delta":{}}],'
+            '"usage":{"prompt_tokens":12,"completion_tokens":3}}'
+        ),
+        usage,
+    )
+    assert usage["input_tokens"] == 12
+    assert usage["output_tokens"] == 3

{coding_proxy-0.3.1a2 → 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"