fix(request-body): 使用 deepcopy 防止请求体在处理流程中被意外修改

handler 基类和格式转换 registry 中，原始请求体通过浅拷贝或直接引用传递，
导致下游处理（模型映射、格式转换、重试整流）可能修改原始数据，
影响后续重试或并发请求的正确性。统一改用 copy.deepcopy 隔离副本。

src/api/handlers/base/chat_handler_base.py

@@ -22,6 +22,7 @@ Chat Handler Base - Chat API 格式的通用基类
 from __future__ import annotations
 
 import asyncio
+import copy
 import json
 from abc import ABC, abstractmethod
 from collections.abc import AsyncGenerator, Awaitable, Callable
@@ -499,7 +500,7 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
 
         # 可变请求体容器：允许 TaskService 在遇到 Thinking 签名错误时整流请求体后重试
         # 结构: {"body": 实际请求体, "_rectified": 是否已整流, "_rectified_this_turn": 本轮是否整流}
-        request_body_ref: dict[str, Any] = {"body": original_request_body}
+        request_body_ref: dict[str, Any] = {"body": copy.deepcopy(original_request_body)}
 
         # 创建类型安全的流式上下文
         ctx = StreamContext(
@@ -723,10 +724,9 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
             )
 
         # 应用模型映射到请求体
+        request_body = copy.deepcopy(original_request_body)
         if mapped_model:
-            request_body = self.apply_mapped_model(original_request_body, mapped_model)
-        else:
-            request_body = dict(original_request_body)
+            request_body = self.apply_mapped_model(request_body, mapped_model)
 
         provider_type = str(getattr(provider, "provider_type", "") or "").lower()
         behavior = get_provider_behavior(

src/api/handlers/base/chat_sync_executor.py

@@ -11,6 +11,7 @@ ChatSyncExecutor - 非流式请求执行器
 
 from __future__ import annotations
 
+import copy
 import json
 from dataclasses import dataclass, field
 from typing import TYPE_CHECKING, Any
@@ -123,7 +124,7 @@ class ChatSyncExecutor:
 
         # 可变请求体容器：允许 TaskService 在遇到 Thinking 签名错误时整流请求体后重试
         # 结构: {"body": 实际请求体, "_rectified": 是否已整流, "_rectified_this_turn": 本轮是否整流}
-        request_body_ref: dict[str, Any] = {"body": original_request_body}
+        request_body_ref: dict[str, Any] = {"body": copy.deepcopy(original_request_body)}
 
         # 捕获的上下文变量
         ctx = self._ctx

src/api/handlers/base/cli_stream_mixin.py

@@ -4,6 +4,7 @@ from __future__ import annotations
 
 import asyncio
 import codecs
+import copy
 import json
 import time
 from collections.abc import AsyncGenerator
@@ -97,7 +98,7 @@ class CliStreamMixin:
 
         # 可变请求体容器：允许 TaskService 在遇到 Thinking 签名错误时整流请求体后重试
         # 结构: {"body": 实际请求体, "_rectified": 是否已整流, "_rectified_this_turn": 本轮是否整流}
-        request_body_ref: dict[str, Any] = {"body": original_request_body}
+        request_body_ref: dict[str, Any] = {"body": copy.deepcopy(original_request_body)}
 
         # 使用子类实现的方法提取 model（不同 API 格式的 model 位置不同）
         # 注意：使用 original_request_body，因为整流只修改 messages，不影响 model 字段
@@ -309,11 +310,10 @@ class CliStreamMixin:
             )
 
         # 应用模型映射到请求体（子类可覆盖此方法处理不同格式）
+        request_body = copy.deepcopy(original_request_body)
         if mapped_model:
             ctx.mapped_model = mapped_model  # 保存映射后的模型名，用于 Usage 记录
-            request_body = self.apply_mapped_model(original_request_body, mapped_model)
-        else:
-            request_body = original_request_body
+            request_body = self.apply_mapped_model(request_body, mapped_model)
 
         client_api_format = (
             ctx.client_api_format.value

src/api/handlers/base/cli_sync_mixin.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import copy
 import json
 import time
 from typing import TYPE_CHECKING, Any
@@ -115,7 +116,7 @@ class CliSyncMixin:
 
         # 可变请求体容器：允许 TaskService 在遇到 Thinking 签名错误时整流请求体后重试
         # 结构: {"body": 实际请求体, "_rectified": 是否已整流, "_rectified_this_turn": 本轮是否整流}
-        request_body_ref: dict[str, Any] = {"body": original_request_body}
+        request_body_ref: dict[str, Any] = {"body": copy.deepcopy(original_request_body)}
 
         async def sync_request_func(
             provider: "Provider",
@@ -136,11 +137,10 @@ class CliSyncMixin:
                 )
 
             # 应用模型映射到请求体（子类可覆盖此方法处理不同格式）
+            request_body = copy.deepcopy(request_body_ref["body"])
             if mapped_model:
                 mapped_model_result = mapped_model  # 保存映射后的模型名，用于 Usage 记录
-                request_body = self.apply_mapped_model(request_body_ref["body"], mapped_model)
-            else:
-                request_body = dict(request_body_ref["body"])
+                request_body = self.apply_mapped_model(request_body, mapped_model)
 
             client_api_format = (
                 api_format.value if hasattr(api_format, "value") else str(api_format)

src/core/api_format/conversion/registry.py

@@ -10,6 +10,7 @@ source -> internal -> target
 """
 
 import ast
+import copy
 import importlib
 import inspect
 import threading
@@ -250,7 +251,7 @@ class FormatConversionRegistry:
         output_limit: int | None = None,
     ) -> dict[str, Any]:
         if self._same_normalizer(source_format, target_format) and not target_variant:
-            return request
+            return copy.deepcopy(request)
 
         # 同 normalizer + variant: 优先尝试轻量补丁（跳过 internal 转换）
         if self._same_normalizer(source_format, target_format) and target_variant:
@@ -258,7 +259,7 @@ class FormatConversionRegistry:
             with _track_conversion_metrics(
                 "request_patch", str(source_format).upper(), str(target_format).upper()
             ):
-                patched = normalizer.patch_for_variant(request, target_variant)
+                patched = normalizer.patch_for_variant(copy.deepcopy(request), target_variant)
             if patched is not None:
                 return patched
 
@@ -269,7 +270,7 @@ class FormatConversionRegistry:
             "request", str(source_format).upper(), str(target_format).upper()
         ):
             try:
-                internal = src.request_to_internal(request)
+                internal = src.request_to_internal(copy.deepcopy(request))
                 internal.output_limit = output_limit
                 repair_stats = self._repair_internal_tool_call_ids(internal)
                 if repair_stats["generated_tool_use_ids"] or repair_stats["filled_tool_result_ids"]:
@@ -295,7 +296,7 @@ class FormatConversionRegistry:
     ) -> dict[str, Any]:
         """异步版本的 convert_request，在 internal 阶段执行图片 URL 下载等异步操作。"""
         if self._same_normalizer(source_format, target_format) and not target_variant:
-            return request
+            return copy.deepcopy(request)
 
         # 同 normalizer + variant: 优先尝试轻量补丁（跳过 internal 转换）
         if self._same_normalizer(source_format, target_format) and target_variant:
@@ -303,7 +304,7 @@ class FormatConversionRegistry:
             with _track_conversion_metrics(
                 "request_patch", str(source_format).upper(), str(target_format).upper()
             ):
-                patched = normalizer.patch_for_variant(request, target_variant)
+                patched = normalizer.patch_for_variant(copy.deepcopy(request), target_variant)
             if patched is not None:
                 return patched
 
@@ -314,7 +315,7 @@ class FormatConversionRegistry:
             "request", str(source_format).upper(), str(target_format).upper()
         ):
             try:
-                internal = src.request_to_internal(request)
+                internal = src.request_to_internal(copy.deepcopy(request))
                 internal.output_limit = output_limit
                 repair_stats = self._repair_internal_tool_call_ids(internal)
                 if repair_stats["generated_tool_use_ids"] or repair_stats["filled_tool_result_ids"]:
@@ -352,15 +353,15 @@ class FormatConversionRegistry:
                             而不是上游返回的映射后模型名。
         """
         if self._same_normalizer(source_format, target_format):
+            response_copy = copy.deepcopy(response)
             # 即使格式相同，也需要替换 model 字段
-            if requested_model and isinstance(response, dict):
-                response = dict(response)  # 避免修改原始响应
+            if requested_model and isinstance(response_copy, dict):
                 # 支持不同格式的 model 字段名
-                if "model" in response:
-                    response["model"] = requested_model
-                elif "modelVersion" in response:
-                    response["modelVersion"] = requested_model
-            return response
+                if "model" in response_copy:
+                    response_copy["model"] = requested_model
+                elif "modelVersion" in response_copy:
+                    response_copy["modelVersion"] = requested_model
+            return response_copy
 
         src = self._require_normalizer(source_format)
         tgt = self._require_normalizer(target_format)
@@ -369,7 +370,7 @@ class FormatConversionRegistry:
             "response", str(source_format).upper(), str(target_format).upper()
         ):
             try:
-                internal = src.response_to_internal(response)
+                internal = src.response_to_internal(copy.deepcopy(response))
                 return tgt.response_from_internal(internal, requested_model=requested_model)
             except Exception as e:
                 raise FormatConversionError(source_format, target_format, str(e)) from e
@@ -381,7 +382,7 @@ class FormatConversionRegistry:
         target_format: str,
     ) -> dict[str, Any]:
         if self._same_normalizer(source_format, target_format):
-            return error_response
+            return copy.deepcopy(error_response)
 
         src = self._require_normalizer(source_format)
         tgt = self._require_normalizer(target_format)
@@ -400,7 +401,7 @@ class FormatConversionRegistry:
             "error", str(source_format).upper(), str(target_format).upper()
         ):
             try:
-                internal = src.error_to_internal(error_response)
+                internal = src.error_to_internal(copy.deepcopy(error_response))
                 return tgt.error_from_internal(internal)
             except Exception as e:
                 raise FormatConversionError(source_format, target_format, str(e)) from e
@@ -428,7 +429,7 @@ class FormatConversionRegistry:
         tgt_base = self._video_format_to_base(target_format)
 
         if src_base == tgt_base:
-            return request
+            return copy.deepcopy(request)
 
         src = self._require_normalizer(src_base)
         tgt = self._require_normalizer(tgt_base)
@@ -437,7 +438,7 @@ class FormatConversionRegistry:
             "video_request", str(source_format).upper(), str(target_format).upper()
         ):
             try:
-                internal = src.video_request_to_internal(request)
+                internal = src.video_request_to_internal(copy.deepcopy(request))
                 return tgt.video_request_from_internal(internal)
             except Exception as e:
                 raise FormatConversionError(source_format, target_format, str(e)) from e
@@ -462,7 +463,7 @@ class FormatConversionRegistry:
         tgt_base = self._video_format_to_base(target_format)
 
         if src_base == tgt_base:
-            return task_response
+            return copy.deepcopy(task_response)
 
         src = self._require_normalizer(src_base)
         tgt = self._require_normalizer(tgt_base)
@@ -471,7 +472,7 @@ class FormatConversionRegistry:
             "video_task", str(source_format).upper(), str(target_format).upper()
         ):
             try:
-                internal = src.video_task_to_internal(task_response)
+                internal = src.video_task_to_internal(copy.deepcopy(task_response))
                 return tgt.video_task_from_internal(internal)
             except Exception as e:
                 raise FormatConversionError(source_format, target_format, str(e)) from e