auto-coder 0.1.354__py3-none-any.whl → 0.1.356__py3-none-any.whl

autocoder/common/global_cancel.py

@@ -14,32 +14,41 @@ class GlobalCancel:
         self._token_flags: Dict[str, bool] = {}
         self._lock = threading.Lock()
         self._context: Dict[str, Any] = {}  # 存储与取消相关的上下文信息
+        self._active_tokens: set[str] = set() # 存储当前正在运行的token
     
-    @property
-    def requested(self) -> bool:
-        """检查是否请求了全局取消（向后兼容）"""
+    def register_token(self, token: str) -> None:
+        """注册一个 token，表示一个操作开始，但尚未请求取消"""
+        with self._lock:            
+            self._token_flags[token] = False            
+            self._active_tokens.add(token)
+
+    def get_active_tokens(self) -> set[str]:
+        """获取当前正在运行的token"""
         with self._lock:
-            return self._global_flag
+            return self._active_tokens.copy()
     
     def is_requested(self, token: Optional[str] = None) -> bool:
-        """检查是否请求了特定token或全局的取消"""
-        with self._lock:
-            # 全局标志总是优先
-            if self._global_flag:
-                return True
-            # 如果提供了token，检查该token的标志
-            if token is not None and token in self._token_flags:
-                return self._token_flags[token]
-            return False
-    
+        """检查是否请求了特定token或全局的取消"""                  
+        if token is not None and token in self._token_flags:
+            return self._token_flags[token]
+        
+        if self._global_flag:
+            return True
+        return False 
+
+    def set_active_tokens(self) -> None:
+        """启用所有活跃的token"""        
+        for token in self._active_tokens:
+            self.set(token)            
+
     def set(self, token: Optional[str] = None, context: Optional[Dict[str, Any]] = None) -> None:
         """设置特定token或全局的取消标志"""
         with self._lock:
             if token is None:
                 self._global_flag = True
             else:
-                self._token_flags[token] = True
-            
+                self._token_flags[token] = True                
+
             # 存储上下文
             if context:
                 if token is None:
@@ -49,6 +58,21 @@ class GlobalCancel:
                         self._context["tokens"] = {}
                     self._context["tokens"][token] = context
     
+    def reset_global(self) -> None:
+        """重置全局取消标志"""
+        with self._lock:
+            self._global_flag = False 
+
+    def reset_token(self, token: str) -> None:
+        """重置特定token的取消标志"""
+        with self._lock:
+            if token in self._token_flags:
+                del self._token_flags[token]
+            if "tokens" in self._context and token in self._context["tokens"]:
+                del self._context["tokens"][token]
+            if token:    
+                self._active_tokens.discard(token) # 从活跃集合中移除
+
     def reset(self, token: Optional[str] = None) -> None:
         """重置特定token或全局的取消标志"""
         with self._lock:
@@ -57,12 +81,21 @@ class GlobalCancel:
                 self._global_flag = False
                 self._token_flags.clear()
                 self._context.clear()
+                self._active_tokens.clear() # 清空活跃集合
             else:
                 # 特定token重置
                 if token in self._token_flags:
                     del self._token_flags[token]
                 if "tokens" in self._context and token in self._context["tokens"]:
                     del self._context["tokens"][token]
+                if token:
+                    self._active_tokens.discard(token) # 从活跃集合中移除
+    
+    def reset_active_tokens(self) -> None:
+        """重置所有活跃的token"""
+        with self._lock:
+            for token in self._active_tokens.copy():
+                self.reset_token(token)            
     
     def get_context(self, token: Optional[str] = None) -> Dict[str, Any]:
         """获取与取消相关的上下文信息"""
@@ -77,6 +110,10 @@ class GlobalCancel:
         """检查是否请求了取消，如果是则抛出异常"""
         if self.is_requested(token):
             context = self.get_context(token)
+            if token:
+                self.reset_token(token)
+            else:
+                self.reset_global()
             raise CancelRequestedException(token, context.get("message", "Operation was cancelled"))
 
 global_cancel = GlobalCancel()

autocoder/common/rulefiles/autocoderrules_utils.py

@@ -4,6 +4,9 @@ from threading import Lock
 import threading
 from typing import Dict, List, Optional
 from loguru import logger
+import re
+import yaml
+from pydantic import BaseModel, Field
 
 # 尝试导入 FileMonitor
 try:
@@ -15,6 +18,15 @@ except ImportError:
     Change = None
 
 
+class RuleFile(BaseModel):
+    """规则文件的Pydantic模型"""
+    description: str = Field(default="", description="规则的描述")
+    globs: List[str] = Field(default_factory=list, description="文件匹配模式列表")
+    always_apply: bool = Field(default=False, alias="alwaysApply", description="是否总是应用规则")
+    content: str = Field(default="", description="规则文件的正文内容")
+    file_path: str = Field(default="", description="规则文件的路径")
+
+
 class AutocoderRulesManager:
     """
     管理和监控 autocoderrules 目录中的规则文件。
@@ -157,9 +169,66 @@ class AutocoderRulesManager:
             self._load_rules()
             logger.info("已重新加载规则")
 
+    def parse_rule_file(self, file_path: str) -> RuleFile:
+        """
+        解析规则文件并返回结构化的Pydantic模型对象
+        
+        Args:
+            file_path: 规则文件的路径
+            
+        Returns:
+            RuleFile: 包含规则文件结构化内容的Pydantic模型
+        """
+        if not os.path.exists(file_path) or not file_path.endswith('.md'):
+            logger.warning(f"无效的规则文件路径: {file_path}")
+            return RuleFile(file_path=file_path)
+            
+        try:
+            with open(file_path, 'r', encoding='utf-8') as f:
+                content = f.read()
+                
+            # 解析YAML头部和Markdown内容
+            yaml_pattern = re.compile(r'^---\s*\n(.*?)\n---\s*\n', re.DOTALL)
+            yaml_match = yaml_pattern.search(content)
+            
+            metadata = {}
+            markdown_content = content
+            
+            if yaml_match:
+                yaml_content = yaml_match.group(1)
+                try:
+                    metadata = yaml.safe_load(yaml_content)
+                    # 移除YAML部分，仅保留Markdown内容
+                    markdown_content = content[yaml_match.end():]
+                except Exception as e:
+                    logger.warning(f"解析规则文件YAML头部时出错: {e}")
+            
+            # 创建并返回Pydantic模型
+            rule = RuleFile(
+                description=metadata.get('description', ''),
+                globs=metadata.get('globs', []),
+                always_apply=metadata.get('alwaysApply', False),
+                content=markdown_content.strip(),
+                file_path=file_path
+            )
+            
+            return rule
+            
+        except Exception as e:
+            logger.warning(f"解析规则文件时出错: {file_path}, 错误: {e}")
+            return RuleFile(file_path=file_path)
+
     def get_rules(self) -> Dict[str, str]:
         """获取所有规则文件内容"""
         return self._rules.copy()
+        
+    def get_parsed_rules(self) -> List[RuleFile]:
+        """获取所有解析后的规则文件"""
+        parsed_rules = []
+        for file_path in self._rules:
+            parsed_rule = self.parse_rule_file(file_path)
+            parsed_rules.append(parsed_rule)
+        return parsed_rules
 
 
 # 对外提供单例
@@ -171,3 +240,17 @@ def get_rules(project_root: Optional[str] = None) -> Dict[str, str]:
     if _rules_manager is None:
         _rules_manager = AutocoderRulesManager(project_root=project_root)
     return _rules_manager.get_rules()
+
+def get_parsed_rules(project_root: Optional[str] = None) -> List[RuleFile]:
+    """获取所有解析后的规则文件，可指定项目根目录"""
+    global _rules_manager
+    if _rules_manager is None:
+        _rules_manager = AutocoderRulesManager(project_root=project_root)
+    return _rules_manager.get_parsed_rules()
+
+def parse_rule_file(file_path: str, project_root: Optional[str] = None) -> RuleFile:
+    """解析指定的规则文件，可指定项目根目录"""
+    global _rules_manager
+    if _rules_manager is None:
+        _rules_manager = AutocoderRulesManager(project_root=project_root)
+    return _rules_manager.parse_rule_file(file_path)

autocoder/common/v2/agent/agentic_edit.py

@@ -785,7 +785,7 @@ Below are some files the user is focused on, and the content is up to date. Thes
         while True:
             iteration_count += 1            
             logger.info(f"Starting LLM interaction cycle #{iteration_count}")
-            global_cancel.check_and_raise()
+            global_cancel.check_and_raise(token=self.args.event_file)
             last_message = conversations[-1]
             if last_message["role"] == "assistant":
                 logger.info(f"Last message is assistant, skipping LLM interaction cycle")
@@ -814,8 +814,8 @@ Below are some files the user is focused on, and the content is up to date. Thes
             event_count = 0
             for event in parsed_events:
                 event_count += 1
-                logger.info(f"Processing event #{event_count}: {type(event).__name__}")
-                global_cancel.check_and_raise()
+                logger.info(f"Processing event #{event_count}: {type(event).__name__}")                
+                global_cancel.check_and_raise(token=self.args.event_file)
                 if isinstance(event, (LLMOutputEvent, LLMThinkingEvent)):
                     assistant_buffer += event.text
                     logger.debug(f"Accumulated {len(assistant_buffer)} chars in assistant buffer")
@@ -1033,7 +1033,7 @@ Below are some files the user is focused on, and the content is up to date. Thes
                 return None
 
         for content_chunk, metadata in generator:
-            global_cancel.check_and_raise()
+            global_cancel.check_and_raise(token=self.args.event_file)
             meta_holder.meta = metadata            
             if not content_chunk:
                 continue

autocoder/common/v2/code_agentic_editblock_manager.py

@@ -347,7 +347,7 @@ class CodeEditBlockManager:
 
         # 计算这次修复缺失上下文花费的token情况
         token_cost_calculator.track_token_usage_by_generate(
-            llm=self.llm,
+            llm=self.code_generator.llms[0],
             generate=generation_result,
             operation_name="code_generation_complete",
             start_time=start_time,
@@ -433,7 +433,7 @@ class CodeEditBlockManager:
             return (unmerged_formatted_text, merged_formatted_text)
 
         for attempt in range(self.args.auto_fix_merge_max_attempts):
-            global_cancel.check_and_raise()
+            global_cancel.check_and_raise(token=self.args.event_file)
             unmerged_formatted_text, merged_formatted_text = _format_blocks(
                 merge)
             fix_prompt = self.fix_unmerged_blocks.prompt(
@@ -470,7 +470,7 @@ class CodeEditBlockManager:
 
             # 计算这次修复未合并块花费的token情况
             token_cost_calculator.track_token_usage_by_generate(
-                llm=self.llm,
+                llm=self.code_generator.llms[0],
                 generate=generation_result,
                 operation_name="code_generation_complete",
                 start_time=start_time,
@@ -544,7 +544,7 @@ class CodeEditBlockManager:
         token_cost_calculator = TokenCostCalculator(args=self.args)
 
         for attempt in range(self.auto_fix_lint_max_attempts):
-            global_cancel.check_and_raise()
+            global_cancel.check_and_raise(token=self.args.event_file)
             # 代码生成结果更新到影子文件里去
             self.shadow_manager.clean_shadows()
             shadow_files = self._create_shadow_files_from_edits(
@@ -611,7 +611,7 @@ class CodeEditBlockManager:
 
             # 计算这次修复lint问题花费的token情况
             token_cost_calculator.track_token_usage_by_generate(
-                llm=self.llm,
+                llm=self.code_generator.llms[0],
                 generate=generation_result,
                 operation_name="code_generation_complete",
                 start_time=start_time,
@@ -657,7 +657,7 @@ class CodeEditBlockManager:
         token_cost_calculator = TokenCostCalculator(args=self.args)
 
         for attempt in range(self.auto_fix_compile_max_attempts):
-            global_cancel.check_and_raise()
+            global_cancel.check_and_raise(token=self.args.event_file)
             # 先更新增量影子系统的文件
             self.shadow_manager.clean_shadows()
             shadow_files = self._create_shadow_files_from_edits(
@@ -710,7 +710,7 @@ class CodeEditBlockManager:
 
             # 计算这次修复compile问题花费的token情况
             token_cost_calculator.track_token_usage_by_generate(
-                llm=self.llm,
+                llm=self.code_generator.llms[0],
                 generate=generation_result,
                 operation_name="code_generation_complete",
                 start_time=start_time,
@@ -749,7 +749,7 @@ class CodeEditBlockManager:
 
         token_cost_calculator = TokenCostCalculator(args=self.args)
         token_cost_calculator.track_token_usage_by_generate(
-            llm=self.llm,
+            llm=self.code_generator.llms[0],
             generate=generation_result,
             operation_name="code_generation_complete",
             start_time=start_time,
@@ -804,7 +804,7 @@ class CodeEditBlockManager:
         # 生成代码并自动修复lint错误
 
         generation_result = self.generate_and_fix(query, source_code_list)
-        global_cancel.check_and_raise()
+        global_cancel.check_and_raise(token=self.args.event_file)
 
         # 合并代码
         self.code_merger.merge_code(generation_result)

autocoder/common/v2/code_diff_manager.py

@@ -251,7 +251,7 @@ class CodeDiffManager:
         
         # 最多尝试修复5次
         for attempt in range(self.max_correction_attempts):
-            global_cancel.check_and_raise()
+            global_cancel.check_and_raise(token=self.args.event_file)
             # 代码生成结果更新到影子文件里去
             shadow_files = self._create_shadow_files_from_edits(generation_result)
             
@@ -326,7 +326,7 @@ class CodeDiffManager:
         """
         # 生成代码并自动修复lint错误
         generation_result = self.generate_and_fix(query, source_code_list)
-        global_cancel.check_and_raise()
+        global_cancel.check_and_raise(token=self.args.event_file)
         # 合并代码        
         self.code_merger.merge_code(generation_result)
         

autocoder/common/v2/code_editblock_manager.py

@@ -24,6 +24,7 @@ from autocoder.shadows.shadow_manager import ShadowManager
 from autocoder.linters.shadow_linter import ShadowLinter
 from autocoder.linters.models import IssueSeverity
 from loguru import logger
+from autocoder.utils.llms import get_llm_names
 from autocoder.common.global_cancel import global_cancel
 from autocoder.linters.models import ProjectLintResult
 from autocoder.common.token_cost_caculate import TokenCostCalculator
@@ -347,7 +348,7 @@ class CodeEditBlockManager:
 
         # 计算这次修复缺失上下文花费的token情况
         token_cost_calculator.track_token_usage_by_generate(
-            llm=self.llm,
+            llm=self.code_generator.llms[0],
             generate=generation_result,
             operation_name="code_generation_complete",
             start_time=start_time,
@@ -445,7 +446,7 @@ class CodeEditBlockManager:
             return (unmerged_formatted_text, merged_formatted_text)
 
         for attempt in range(self.args.auto_fix_merge_max_attempts):
-            global_cancel.check_and_raise()
+            global_cancel.check_and_raise(token=self.args.event_file)
             unmerged_formatted_text, merged_formatted_text = _format_blocks(
                 merge)
             fix_prompt = self.fix_unmerged_blocks.prompt(
@@ -482,7 +483,7 @@ class CodeEditBlockManager:
 
             # 计算这次修复未合并块花费的token情况
             token_cost_calculator.track_token_usage_by_generate(
-                llm=self.llm,
+                llm=self.code_generator.llms[0],
                 generate=generation_result,
                 operation_name="code_generation_complete",
                 start_time=start_time,
@@ -556,7 +557,7 @@ class CodeEditBlockManager:
         token_cost_calculator = TokenCostCalculator(args=self.args)
 
         for attempt in range(self.auto_fix_lint_max_attempts):
-            global_cancel.check_and_raise()
+            global_cancel.check_and_raise(token=self.args.event_file)
             # 代码生成结果更新到影子文件里去
             self.shadow_manager.clean_shadows()
             shadow_files = self._create_shadow_files_from_edits(
@@ -623,7 +624,7 @@ class CodeEditBlockManager:
 
             # 计算这次修复lint问题花费的token情况
             token_cost_calculator.track_token_usage_by_generate(
-                llm=self.llm,
+                llm=self.code_generator.llms[0],
                 generate=generation_result,
                 operation_name="code_generation_complete",
                 start_time=start_time,
@@ -669,7 +670,7 @@ class CodeEditBlockManager:
         token_cost_calculator = TokenCostCalculator(args=self.args)
 
         for attempt in range(self.auto_fix_compile_max_attempts):
-            global_cancel.check_and_raise()
+            global_cancel.check_and_raise(token=self.args.event_file)
             # 先更新增量影子系统的文件
             self.shadow_manager.clean_shadows()
             shadow_files = self._create_shadow_files_from_edits(
@@ -722,7 +723,7 @@ class CodeEditBlockManager:
 
             # 计算这次修复compile问题花费的token情况
             token_cost_calculator.track_token_usage_by_generate(
-                llm=self.llm,
+                llm=self.code_generator.llms[0],
                 generate=generation_result,
                 operation_name="code_generation_complete",
                 start_time=start_time,
@@ -759,9 +760,9 @@ class CodeEditBlockManager:
         generation_result = self.code_generator.single_round_run(
             query, source_code_list)
 
-        token_cost_calculator = TokenCostCalculator(args=self.args)
+        token_cost_calculator = TokenCostCalculator(args=self.args)        
         token_cost_calculator.track_token_usage_by_generate(
-            llm=self.llm,
+            llm=self.code_generator.llms[0],
             generate=generation_result,
             operation_name="code_generation_complete",
             start_time=start_time,
@@ -816,7 +817,7 @@ class CodeEditBlockManager:
         # 生成代码并自动修复lint错误
 
         generation_result = self.generate_and_fix(query, source_code_list)
-        global_cancel.check_and_raise()
+        global_cancel.check_and_raise(token=self.args.event_file)
 
         # 合并代码
         self.code_merger.merge_code(generation_result)

autocoder/common/v2/code_strict_diff_manager.py

@@ -32,6 +32,7 @@ class CodeStrictDiffManager:
         self.code_merger = CodeAutoMergeStrictDiff(llm, args)
         self.shadow_manager = ShadowManager(args.source_dir, args.event_file)
         self.shadow_linter = ShadowLinter(self.shadow_manager, verbose=False)
+        self.args = args
 
     @byzerllm.prompt()
     def fix_linter_errors(self, query: str, lint_issues: str) -> str:
@@ -159,7 +160,7 @@ class CodeStrictDiffManager:
         
         # 最多尝试修复5次
         for attempt in range(self.max_correction_attempts):
-            global_cancel.check_and_raise()
+            global_cancel.check_and_raise(token=self.args.event_file)
             # 代码生成结果更新到影子文件里去
             shadow_files = self._create_shadow_files_from_edits(generation_result)
             
@@ -234,7 +235,7 @@ class CodeStrictDiffManager:
         """
         # 生成代码并自动修复lint错误
         generation_result = self.generate_and_fix(query, source_code_list)
-        global_cancel.check_and_raise()
+        global_cancel.check_and_raise(token=self.args.event_file)
         # 合并代码        
         self.code_merger.merge_code(generation_result)
         

autocoder/dispacher/actions/action.py

@@ -113,7 +113,7 @@ class ActionTSProject(BaseAction):
                     f"Content(send to model) is {content_length} tokens, which is larger than the maximum input length {self.args.model_max_input_length}"
                 )
 
-        global_cancel.check_and_raise()
+        global_cancel.check_and_raise(token=self.args.event_file)
 
         if (args.enable_auto_fix_merge or args.enable_auto_fix_lint) and args.execute and args.auto_merge=="editblock":
             code_merge_manager = CodeEditBlockManager(llm=self.llm, args=self.args,action=self)
@@ -179,7 +179,7 @@ class ActionTSProject(BaseAction):
                     action_file=self.args.file
                 ).to_dict())
 
-            global_cancel.check_and_raise()
+            global_cancel.check_and_raise(token=self.args.event_file)
 
             merge_result = None
             if args.execute and args.auto_merge:
@@ -267,7 +267,7 @@ class ActionPyProject(BaseAction):
                     max_length=self.args.model_max_input_length
                 )
 
-        global_cancel.check_and_raise()
+        global_cancel.check_and_raise(token=self.args.event_file)
 
         if (args.enable_auto_fix_merge or args.enable_auto_fix_lint) and args.execute and args.auto_merge=="editblock":
             code_merge_manager = CodeEditBlockManager(llm=self.llm, args=self.args,action=self)
@@ -334,7 +334,7 @@ class ActionPyProject(BaseAction):
                     action_file=self.args.file
                 ).to_dict())
 
-            global_cancel.check_and_raise()
+            global_cancel.check_and_raise(token=self.args.event_file)
 
             merge_result = None
             if args.execute and args.auto_merge:
@@ -415,7 +415,7 @@ class ActionSuffixProject(BaseAction):
                     f"Content(send to model) is {content_length} tokens, which is larger than the maximum input length {self.args.model_max_input_length}"
                 )
 
-        global_cancel.check_and_raise()
+        global_cancel.check_and_raise(token=self.args.event_file)
 
         if (args.enable_auto_fix_merge or args.enable_auto_fix_lint) and args.execute and args.auto_merge=="editblock":
             code_merge_manager = CodeEditBlockManager(llm=self.llm, args=self.args,action=self)
@@ -481,7 +481,7 @@ class ActionSuffixProject(BaseAction):
                     action_file=self.args.file
                 ).to_dict())
 
-        global_cancel.check_and_raise()
+        global_cancel.check_and_raise(token=self.args.event_file)
 
         merge_result = None
         if args.execute and args.auto_merge:

autocoder/dispacher/actions/plugins/action_regex_project.py

@@ -67,7 +67,7 @@ class ActionRegexProject:
 
         start_time = time.time()
         
-        global_cancel.check_and_raise()
+        global_cancel.check_and_raise(token=self.args.event_file)
 
         if (args.enable_auto_fix_merge or args.enable_auto_fix_lint) and args.execute and args.auto_merge=="editblock":
             code_merge_manager = CodeEditBlockManager(llm=self.llm, args=self.args,action=self)
@@ -128,7 +128,7 @@ class ActionRegexProject:
                     action_file=self.args.file
                 ).to_dict())
 
-            global_cancel.check_and_raise()
+            global_cancel.check_and_raise(token=self.args.event_file)
             
             merge_result = None
             if args.execute and args.auto_merge:

autocoder/events/event_manager_singleton.py

@@ -71,7 +71,7 @@ class EventManagerSingleton:
 
 def get_event_file_path(file_id:str,project_path: Optional[str] = None) -> str:
     if project_path is None:
-        return os.path.join(".auto-coder", "events", f"{file_id}.jsonl")
+        return os.path.join(os.getcwd(),".auto-coder", "events", f"{file_id}.jsonl")
     else:
         return os.path.join(project_path, ".auto-coder", "events", f"{file_id}.jsonl")
 

autocoder/index/index.py

@@ -298,7 +298,7 @@ class IndexManager:
         return False
 
     def build_index_for_single_source(self, source: SourceCode):
-        global_cancel.check_and_raise()
+        global_cancel.check_and_raise(token=self.args.event_file)
 
         file_path = source.module_name
         if not os.path.exists(file_path):
@@ -575,7 +575,7 @@ class IndexManager:
                 for source in wait_to_build_files
             ]
             for future in as_completed(futures):
-                global_cancel.check_and_raise()
+                global_cancel.check_and_raise(token=self.args.event_file)
                 result = future.result()
                 if result is not None:
                     counter += 1

autocoder/rag/cache/local_byzer_storage_cache.py

@@ -269,7 +269,7 @@ class LocalByzerStorageCache(BaseCacheManager):
                         "file_path": file_info.file_path,
                         "content": chunk[0:self.chunk_size*2],
                         "raw_content": chunk[0:self.chunk_size*2],
-                        "vector": chunk[0:self.chunk_size*2],
+                        "vector": chunk[0:self.args.rag_emb_text_size],
                         "mtime": file_info.modify_time,
                     }
                     items.append(chunk_item)

autocoder/rag/cache/local_duckdb_storage_cache.py

@@ -88,6 +88,7 @@ class LocalDuckdbStorage:
         table_name: str = "documents",
         embed_dim: Optional[int] = None,
         persist_dir: str = "./storage",
+        args: Optional[AutoCoderArgs] = None,
     ) -> None:
         self.llm = llm
         self.database_name = database_name
@@ -95,6 +96,7 @@ class LocalDuckdbStorage:
         self.embed_dim = embed_dim
         self.persist_dir = persist_dir
         self.cache_dir = os.path.join(self.persist_dir, ".cache")
+        self.args = args
         logger.info(f"正在启动 DuckDBVectorStore.")
 
         if self.database_name != ":memory:":
@@ -239,6 +241,11 @@ class LocalDuckdbStorage:
     def _node_to_table_row(
         self, context_chunk: Dict[str, str | float], dim: int | None = None
     ) -> Any:
+        
+        if not context_chunk["raw_content"]:
+            context_chunk["raw_content"] = "empty"        
+        context_chunk["raw_content"] = context_chunk["raw_content"][: self.args.rag_emb_text_size]
+            
         return (
             context_chunk["_id"],
             context_chunk["file_path"],
@@ -332,6 +339,7 @@ class LocalDuckDBStorageCache(BaseCacheManager):
             database_name="byzerai_store_duckdb.db",
             table_name="rag_duckdb",
             persist_dir=self.path,
+            args=args,
         )
         self.queue = []
         self.chunk_size = 1000

autocoder/rag/loaders/image_loader.py

@@ -33,6 +33,9 @@ class ImageLoader:
     and converting the content to markdown format.
     """
     
+    # 存储不同参数组合的PaddleOCR实例
+    _ocr_instances = {}
+    
     @staticmethod
     def parse_diff(diff_content: str) -> List[Tuple[str, str]]:
         """
@@ -106,19 +109,28 @@ class ImageLoader:
             print("paddleocr not installed")
             return ""
 
-        # 初始化 OCR
-        try:
-            ocr = PaddleOCR(
-                use_angle_cls=use_angle_cls,
-                lang=lang,
-                page_num=page_num,
-                det_model_dir=det_model_dir,
-                rec_model_dir=rec_model_dir,
-                **kwargs
-            )
-        except Exception:
-            traceback.print_exc()
-            return ""
+        # 创建一个参数的哈希键，用于在缓存中存储OCR实例
+        param_key = f"{lang}_{use_angle_cls}_{page_num}_{det_model_dir}_{rec_model_dir}_{hash(frozenset(kwargs.items()) if kwargs else 0)}"
+        
+        # 检查是否已经有对应参数的OCR实例
+        if param_key not in ImageLoader._ocr_instances:
+            try:
+                # 初始化OCR并缓存
+                ImageLoader._ocr_instances[param_key] = PaddleOCR(
+                    use_angle_cls=use_angle_cls,
+                    lang=lang,
+                    page_num=page_num,
+                    det_model_dir=det_model_dir,
+                    rec_model_dir=rec_model_dir,
+                    **kwargs
+                )
+                logger.info(f"初始化新的PaddleOCR实例，参数：{param_key}")
+            except Exception:
+                traceback.print_exc()
+                return ""
+        
+        # 使用缓存的OCR实例
+        ocr = ImageLoader._ocr_instances[param_key]
 
         try:
             ext = os.path.splitext(file_path)[1].lower()