需要安装python3环境
需要安装 xlwt、xmindparser 这两个第三方包
XmindExcel.py 文件代码
# coding=utf-8 import time import xlwt from past.builtins import raw_input from xmindparser import xmind_to_dict def resolvePath(dict, lists, title): # title去除首尾空格 title = title.strip() # 如果title是空字符串,则直接获取value if len(title) == 0: concatTitle = dict['title'].strip() elif "makers" in dict.keys(): if "priority-" in str(dict["makers"]): concatTitle = title + '\t' + dict['title'].strip() + "\t" + str(dict["makers"]) else: concatTitle = title + '\t' + dict['title'].strip() else: concatTitle = title + '\t' + dict['title'].strip() if dict.__contains__('topics') == False: lists.append(concatTitle) else: for d in dict['topics']: resolvePath(d, lists, concatTitle def xmind_cat(list, excelname, groupname): f = xlwt.Workbook() sheet = f.add_sheet(groupname, cell_overwrite_ok=True) row0 = ["测试用例编号", "用例标题", "预置条件", "执行方式", "优先级", "测试步骤", "预期结果", "所属项目"] # 生成第一行中固定表头内容 for i in range(0, len(row0)): sheet.write(0, i, row0[i]) # 增量索引 index = 0 # case级别 case_leve_index = "" # 前置条件 case_pre_condition = [] pre_num = 0 for h in range(0, len(list)): # print("list:",list) lists = [] resolvePath(list[h], lists, '') for j in range(0, len(lists)): # 将xmind转成excel lists[j] = lists[j].split('\t') try: # print(index) if "【预置条件】" in lists[j][-1] or "【前置条件】" in lists[j][-1]: case_pre_condition.append(lists[j]) pre_num += 1 else: case_leve = "" for n in range(len(lists[j])): if 'priority-' in str(lists[j][n]): case_leve_index = n-1 if "priority-1" in str(lists[j][n]): case_leve = "High" elif "priority-2" in str(lists[j][n]): case_leve = "Middle" elif "priority-3" in str(lists[j][n]): case_leve = "Low" lists[j].pop(n) break case_name = "-".join(lists[j][:case_leve_index+1]) sheet.write(j + index + 1 - pre_num, 1, case_name) # 标题 if len(lists[j][case_leve_index:-1]) < 2: sheet.write(j + index + 1 - pre_num, 6, lists[j][case_leve_index + 1]) # 期望结果 else: sheet.write(j + index + 1- pre_num, 5, lists[j][case_leve_index + 1]) # 步骤 sheet.write(j + index + 1- pre_num, 6, lists[j][case_leve_index + 2]) # 期望结果 sheet.write(j + index + 1- pre_num, 3, "手动") # 执行方式 sheet.write(j + index + 1 - pre_num, 4, case_leve) # 预置条件 if len(case_pre_condition) > 0: for pre_list in case_pre_condition: if set(pre_list[:-1]) < set(lists[j]): sheet.write(j + index + 1 - pre_num, 2, pre_list[-1]) except: print("请检查编写的用例是否符合规范:", lists[j]) # 遍历结束lists,给增量索引赋值,跳出for j循环,开始for h循环 if j == len(lists) - 1: index += len(lists) f.save(excelname) def maintest(filename, excelname): out = xmind_to_dict(filename) groupname = out[0]['topic']['title'] xmind_cat(out[0]['topic']['topics'], excelname, groupname) if __name__ == '__main__': try: path = raw_input("请输入Xmind用例文件路径,可将文件拖拽到此处:") filename = path excelname = path.rstrip('xmind') + 'xls' maintest(filename, excelname) print('SUCCESS!\n生成用例成功,用例目录:%s' % excelname) except: print('请确认后重试:\n1.用例文件路径中不能有空格换行符\n2.请使用python3运行\n3.检查xmind文件中不能有乱码或无法识别的字符(xmind自带表情字符除外)\n4.检查是否将已生成的excel文件未关闭')
运行XmindExcel.py文件,输入文件目录运行即可。生成的Excel文件会在Xmind文件的同路径下,文件名称与Xmind文件名称一致
免责声明:本站发布的内容(图片、视频和文字)以原创、转载和分享为主,文章观点不代表本网站立场,如果涉及侵权请联系站长邮箱:mmqy2019@163.com进行举报,并提供相关证据,查实之后,将立刻删除涉嫌侵权内容。
长按识别二维码并关注微信
更方便到期提醒、手机管理