import os def extract_files_as_code_list(root_dir): """ 递归扫描 root_dir 目录,提取所有 .cpp、.h、.hpp、.cc 文件的内容, 返回符合 code_list 格式的列表: [ {"file_rel_path": "relative/path/to/file.cpp", "code_content": "..."}, ... ] 参数: root_dir (str): 要扫描的根目录路径(绝对或相对) 返回: list[dict]: 符合 AscendC 编译服务要求的 code_list """ code_list = [] root_dir = os.path.abspath(root_dir) # 确保是绝对路径,便于计算相对路径 for dirpath, _, filenames in os.walk(root_dir): for filename in filenames: if filename.endswith(('.cpp', '.h', '.hpp', '.cc')): file_path = os.path.join(dirpath, filename) # 计算相对于 root_dir 的相对路径 rel_path = os.path.relpath(file_path, root_dir).replace(os.sep, '/') # 统一使用正斜杠 try: with open(file_path, 'r', encoding='utf-8') as f: content = f.read() except UnicodeDecodeError: # 如果 UTF-8 失败,尝试 latin-1 或忽略错误(根据实际需求调整) with open(file_path, 'r', encoding='latin-1') as f: content = f.read() except Exception as e: print(f"Warning: Skipping file {file_path} due to read error: {e}") continue code_list.append({ "file_rel_path": rel_path, "code_content": content }) return code_list # 示例用法 if __name__ == "__main__": import json op_root = "${ASCEND_GYM_DATA_ROOT}/models/user/mse_loss_v2" code_list = extract_files_as_code_list(op_root) # 打印结果(或用于后续调用 compile_code) print(json.dumps(code_list, indent=2, ensure_ascii=False))