refactor(json_parser): 统一 LLM 响应的 JSON 解析逻辑，简化代码并提高解析成功率

2025-11-02 12:18:53 +08:00
parent bd1624a018
commit d1c3d2196a
8 changed files with 511 additions and 179 deletions
--- a/src/plugins/built_in/affinity_flow_chatter/chat_stream_impression_tool.py
+++ b/src/plugins/built_in/affinity_flow_chatter/chat_stream_impression_tool.py
@@ -13,6 +13,7 @@ from src.common.logger import get_logger
 from src.config.config import model_config
 from src.llm_models.utils_model import LLMRequest
 from src.plugin_system import BaseTool, ToolParamType
+from src.utils.json_parser import extract_and_parse_json

 logger = get_logger("chat_stream_impression_tool")

@@ -290,9 +291,11 @@ class ChatStreamImpressionTool(BaseTool):
                logger.warning("LLM未返回有效响应")
                return None

-            # 清理并解析响应
-            cleaned_response = self._clean_llm_json_response(llm_response)
-            response_data = json.loads(cleaned_response)
+            # 使用统一的 JSON 解析工具
+            response_data = extract_and_parse_json(llm_response, strict=False)
+            if not response_data or not isinstance(response_data, dict):
+                logger.warning("解析LLM响应失败")
+                return None

            # 提取最终决定的数据
            final_impression = {
@@ -373,35 +376,18 @@ class ChatStreamImpressionTool(BaseTool):
            logger.error(f"更新聊天流印象到数据库失败: {e}", exc_info=True)
            raise

-    def _clean_llm_json_response(self, response: str) -> str:
-        """清理LLM响应，移除可能的JSON格式标记
-
-        Args:
-            response: LLM原始响应
-
-        Returns:
-            str: 清理后的JSON字符串
+    # 已移除自定义的 _clean_llm_json_response 方法，统一使用 src.utils.json_parser.extract_and_parse_json
+    
+    def _clean_llm_json_response_deprecated(self, response: str) -> str:
+        """已废弃，保留仅用于兼容性
+        
+        请使用 src.utils.json_parser.extract_and_parse_json 替代
        """
+        from src.utils.json_parser import extract_and_parse_json
        try:
-            import re
-
-            cleaned = response.strip()
-
-            # 移除 ```json 或 ``` 等标记
-            cleaned = re.sub(r"^```(?:json)?\s*", "", cleaned, flags=re.MULTILINE | re.IGNORECASE)
-            cleaned = re.sub(r"\s*```$", "", cleaned, flags=re.MULTILINE)
-
-            # 尝试找到JSON对象的开始和结束
-            json_start = cleaned.find("{")
-            json_end = cleaned.rfind("}")
-
-            if json_start != -1 and json_end != -1 and json_end > json_start:
-                cleaned = cleaned[json_start : json_end + 1]
-
-            cleaned = cleaned.strip()
-
-            return cleaned
-
+            import json
+            result = extract_and_parse_json(response, strict=False)
+            return json.dumps(result) if result else response
        except Exception as e:
            logger.warning(f"清理LLM响应失败: {e}")
            return response
--- a/src/plugins/built_in/affinity_flow_chatter/proactive_thinking_executor.py
+++ b/src/plugins/built_in/affinity_flow_chatter/proactive_thinking_executor.py
@@ -17,6 +17,7 @@ from src.config.config import global_config, model_config
 from src.individuality.individuality import Individuality
 from src.llm_models.utils_model import LLMRequest
 from src.plugin_system.apis import message_api, send_api
+from src.utils.json_parser import extract_and_parse_json

 logger = get_logger("proactive_thinking_executor")

@@ -339,19 +340,17 @@ class ProactiveThinkingPlanner:
                logger.warning("LLM未返回有效响应")
                return None

-            # 清理并解析JSON响应
-            cleaned_response = self._clean_json_response(response)
-            decision = json.loads(cleaned_response)
+            # 使用统一的 JSON 解析工具
+            decision = extract_and_parse_json(response, strict=False)
+            if not decision or not isinstance(decision, dict):
+                logger.error("解析决策JSON失败")
+                if response:
+                    logger.debug(f"原始响应: {response[:500]}")
+                return None

            logger.info(f"决策结果: {decision.get('action', 'unknown')} - {decision.get('reasoning', '无理由')}")

            return decision
-
-        except json.JSONDecodeError as e:
-            logger.error(f"解析决策JSON失败: {e}")
-            if response:
-                logger.debug(f"原始响应: {response}")
-            return None
        except Exception as e:
            logger.error(f"决策过程失败: {e}", exc_info=True)
            return None
@@ -469,21 +468,7 @@ class ProactiveThinkingPlanner:
            logger.warning(f"获取表达方式失败: {e}")
            return ""

-    def _clean_json_response(self, response: str) -> str:
-        """清理LLM响应中的JSON格式标记"""
-        import re
-
-        cleaned = response.strip()
-        cleaned = re.sub(r"^```(?:json)?\s*", "", cleaned, flags=re.MULTILINE | re.IGNORECASE)
-        cleaned = re.sub(r"\s*```$", "", cleaned, flags=re.MULTILINE)
-
-        json_start = cleaned.find("{")
-        json_end = cleaned.rfind("}")
-
-        if json_start != -1 and json_end != -1 and json_end > json_start:
-            cleaned = cleaned[json_start : json_end + 1]
-
-        return cleaned.strip()
+    # 已移除自定义的 _clean_json_response 方法，统一使用 src.utils.json_parser.extract_and_parse_json


 # 全局规划器实例
--- a/src/plugins/built_in/affinity_flow_chatter/user_profile_tool.py
+++ b/src/plugins/built_in/affinity_flow_chatter/user_profile_tool.py
@@ -16,6 +16,7 @@ from src.common.logger import get_logger
 from src.config.config import global_config, model_config
 from src.llm_models.utils_model import LLMRequest
 from src.plugin_system import BaseTool, ToolParamType
+from src.utils.json_parser import extract_and_parse_json

 logger = get_logger("user_profile_tool")

@@ -269,9 +270,12 @@ class UserProfileTool(BaseTool):
                logger.warning("LLM未返回有效响应")
                return None

-            # 清理并解析响应
-            cleaned_response = self._clean_llm_json_response(llm_response)
-            response_data = orjson.loads(cleaned_response)
+            # 使用统一的 JSON 解析工具
+            response_data = extract_and_parse_json(llm_response, strict=False)
+            if not response_data or not isinstance(response_data, dict):
+                logger.error("LLM响应JSON解析失败")
+                logger.debug(f"LLM原始响应: {llm_response[:500] if llm_response else 'N/A'}")
+                return None

            # 提取最终决定的数据
            final_profile = {
@@ -285,11 +289,6 @@ class UserProfileTool(BaseTool):
            logger.debug(f"决策理由: {response_data.get('reasoning', '无')}")

            return final_profile
-
-        except orjson.JSONDecodeError as e:
-            logger.error(f"LLM响应JSON解析失败: {e}")
-            logger.debug(f"LLM原始响应: {llm_response if 'llm_response' in locals() else 'N/A'}")
-            return None
        except Exception as e:
            logger.error(f"LLM决策失败: {e}", exc_info=True)
            return None
@@ -336,35 +335,4 @@ class UserProfileTool(BaseTool):
            logger.error(f"更新用户画像到数据库失败: {e}", exc_info=True)
            raise

-    def _clean_llm_json_response(self, response: str) -> str:
-        """清理LLM响应，移除可能的JSON格式标记
-
-        Args:
-            response: LLM原始响应
-
-        Returns:
-            str: 清理后的JSON字符串
-        """
-        try:
-            import re
-
-            cleaned = response.strip()
-
-            # 移除 ```json 或 ``` 等标记
-            cleaned = re.sub(r"^```(?:json)?\s*", "", cleaned, flags=re.MULTILINE | re.IGNORECASE)
-            cleaned = re.sub(r"\s*```$", "", cleaned, flags=re.MULTILINE)
-
-            # 尝试找到JSON对象的开始和结束
-            json_start = cleaned.find("{")
-            json_end = cleaned.rfind("}")
-
-            if json_start != -1 and json_end != -1 and json_end > json_start:
-                cleaned = cleaned[json_start:json_end + 1]
-
-            cleaned = cleaned.strip()
-
-            return cleaned
-
-        except Exception as e:
-            logger.warning(f"清理LLM响应失败: {e}")
-            return response
+    # 已移除自定义的 _clean_llm_json_response 方法，统一使用 src.utils.json_parser.extract_and_parse_json