Class DBResultProcessor

src/server/tasks/dbbench/result_processor.py:3–280

处理数据库查询结果和比较的类 只对外暴露compare_results和calculate_tables_hash接口

from .interaction import Database

class DBResultProcessor:
    """
    处理数据库查询结果和比较的类
    只对外暴露compare_results和calculate_tables_hash接口
    """
    
    @staticmethod
    def compare_results(answer, ground_truth, query_type):
        """
        比较答案和标准答案
        
        参数:
        answer - 模型输出的答案
        ground_truth - 标准答案
        query_type - 查询类型 (SELECT/INSERT/UPDATE/DELETE)
        
        返回:
        bool - 答案是否匹配
        """
        try:
            # 处理answer和ground_truth
            processed_answer = DBResultProcessor._clean_answer(answer)
            processed_ground_truth = DBResultProcessor._clean_answer(ground_truth)
            
            if query_type in ("INSERT", "DELETE", "UPDATE"):
                return processed_answer == processed_ground_truth
                
            # 打印处理后的结果用于调试
            print("Processed answer:", processed_answer)
            print("Processed ground_truth:", processed_ground_truth)
            
            # 比较逻辑
            if len(processed_answer) == 1 and len(processed_ground_truth) == 1:
                # 获取处理后的值
                ans_val = processed_answer[0]
                gt_val = processed_ground_truth[0]
                
                # 如果两个值都是特殊值（0、undefined等），认为它们相等
                if ans_val == "0" and gt_val == "0":
                    return True
                    
                # 浮点数比较
                if DBResultProcessor._is_float(ans_val) and DBResultProcessor._is_float(gt_val):
                    return DBResultProcessor._float_equal(ans_val, gt_val)

                # 字符串比较
                return ans_val == gt_val
            else:
                # 如果都是浮点数，执行浮点比较
                if (all(DBResultProcessor._is_float(x) for x in processed_answer) and 
                    all(DBResultProcessor._is_float(x) for x in processed_ground_truth)):
                    # 检查每个答案是否都有匹配的标准答案（考虑精度）
                    if len(processed_answer) != len(processed_ground_truth):
                        return False
                        
                    # 创建匹配标记
                    matched_gt = [False] * len(processed_ground_truth)