首页 > 编程 > Python > 正文

利用python模拟sql语句对员工表格进行增删改查

2020-01-04 16:56:23
字体:
来源:转载
供稿:网友

本文主要给大家介绍了关于python/114102.html">python模拟sql语句对员工表格进行增删改查的相关内容,分享出来供大家参考学习,下面来一起看看详细的介绍:

具体需求:

员工信息表程序,实现增删改查操作:

可进行模糊查询,语法支持下面3种:

  select name,age from staff_data where age > 22                  多个查询参数name,age 用','分割

  select * from staff_data where dept = 人事

  select * from staff_data where enroll_date like 2013

查到的信息,打印后,最后面还要显示查到的条数

可创建新员工纪录,以phone做唯一键,phone存在即提示,staff_id需自增,添加多个记录record1/record2中间用'/'分割

  insert into staff_data values record1/record2

可删除指定员工信息纪录,输入员工id,即可删除

  delete from staff_data where staff_id>=5andstaff_id<=10

可修改员工信息,语法如下:

  update staff_table set dept=Market,phone=13566677787  where dept = 运维   多个set值用','分割

使用re模块,os模块,充分使用函数精简代码,熟练使用 str.split()来解析格式化字符串

由于,sql命令中的几个关键字符串有一定规律,只出现一次,并且有顺序!!!

按照key_lis = ['select', 'insert', 'delete', 'update', 'from', 'into', 'set', 'values', 'where', 'limit']的元素顺序分割sql. 

分割元素作为sql_dic字典的key放进字典中.分割后的列表为b,如果len(b)>1,说明sql字符串中含有分割元素,同时b[0]对应上一个分割元素的值,b[-1]为下一次分割对象!

这样不断迭代直到把sql按出现的所有分割元素分割完毕,但注意这里每次循环都是先分割后赋值!!!当前分割元素比如'select'对应的值,需要等到下一个分割元素

比如'from'执行分割后的列表b,其中b[0]的值才会赋值给sql_dic['select'] ,所以最后一个分割元素的值,不能通过上述循环来完成,必须先处理可能是最后一个分割元素,再正常循环!!

在这sql语句中,有可能成为最后一个分割元素的 'limit' ,'values', 'where',  按优先级别,先处理'limit' ,再处理'values'或 'where'.....

处理完得到sql_dic后,就是你按不同命令执行,对数据文件的增删改查,最后返回处理结果!!

示例代码

# _*_coding:utf-8_*_# Author:Jaye Heimport reimport osdef sql_parse(sql, key_lis): ''' 解析sql命令字符串,按照key_lis列表里的元素分割sql得到字典形式的命令sql_dic :param sql: :param key_lis: :return: ''' sql_list = [] sql_dic = {} for i in key_lis:  b = [j.strip() for j in sql.split(i)]  if len(b) > 1:   if len(sql.split('limit')) > 1:    sql_dic['limit'] = sql.split('limit')[-1]   if i == 'where' or i == 'values':    sql_dic[i] = b[-1]   if sql_list:    sql_dic[sql_list[-1]] = b[0]   sql_list.append(i)   sql = b[-1]  else:   sql = b[0]  if sql_dic.get('select'):   if not sql_dic.get('from') and not sql_dic.get('where'):    sql_dic['from'] = b[-1] if sql_dic.get('select'):  sql_dic['select'] = sql_dic.get('select').split(',') if sql_dic.get('where'):  sql_dic['where'] = where_parse(sql_dic.get('where')) return sql_dicdef where_parse(where): ''' 格式化where字符串为列表where_list,用'and', 'or', 'not'分割字符串 :param where: :return: ''' casual_l = [where] logic_key = ['and', 'or', 'not'] for j in logic_key:  for i in casual_l:   if i not in logic_key:    if len(i.split(j)) > 1:     ele = i.split(j)     index = casual_l.index(i)     casual_l.pop(index)     casual_l.insert(index, ele[0])     casual_l.insert(index+1, j)     casual_l.insert(index+2, ele[1])     casual_l = [k for k in casual_l if k] where_list = three_parse(casual_l, logic_key) return where_listdef three_parse(casual_l, logic_key): ''' 处理临时列表casual_l中具体的条件,'staff_id>5'-->['staff_id','>','5'] :param casual_l: :param logic_key: :return: ''' where_list = [] for i in casual_l:  if i not in logic_key:   b = i.split('like')   if len(b) > 1:    b.insert(1, 'like')    where_list.append(b)   else:    key = ['<', '=', '>']    new_lis = []    opt = ''    lis = [j for j in re.split('([=<>])', i) if j]    for k in lis:     if k in key:      opt += k     else:      new_lis.append(k)    new_lis.insert(1, opt)    where_list.append(new_lis)  else:   where_list.append(i) return where_listdef sql_action(sql_dic, title): ''' 把解析好的sql_dic分发给相应函数执行处理 :param sql_dic: :param title: :return: ''' key = {'select': select,   'insert': insert,   'delete': delete,   'update': update} res = [] for i in sql_dic:  if i in key:   res = key[i](sql_dic, title) return resdef select(sql_dic, title): ''' 处理select语句命令 :param sql_dic: :param title: :return: ''' with open('staff_data', 'r', encoding='utf-8') as fh:  filter_res = where_action(fh, sql_dic.get('where'), title)  limit_res = limit_action(filter_res, sql_dic.get('limit'))  search_res = search_action(limit_res, sql_dic.get('select'), title) return search_resdef insert(sql_dic, title): ''' 处理insert语句命令 :param sql_dic: :param title: :return: ''' with open('staff_data', 'r+', encoding='utf-8') as f:  data = f.readlines()  phone_list = [i.strip().split(',')[4] for i in data]  ins_count = 0  if not data:   new_id = 1  else:   last = data[-1]   last_id = int(last.split(',')[0])   new_id = last_id+1  record = sql_dic.get('values').split('/')  for i in record:   if i.split(',')[3] in phone_list:    print('/033[1;31m%s 手机号已存在/033[0m' % i)   else:    new_record = '%s,%s/n' % (str(new_id), i)    f.write(new_record)    new_id += 1    ins_count += 1  f.flush() return ['insert successful'], [str(ins_count)]def delete(sql_dic, title): ''' 处理delete语句命令 :param sql_dic: :param title: :return: ''' with open('staff_data', 'r', encoding='utf-8') as r_file,/   open('staff_data_bak', 'w', encoding='utf-8') as w_file:  del_count = 0  for line in r_file:   dic = dict(zip(title.split(','), line.split(',')))   filter_res = logic_action(dic, sql_dic.get('where'))   if not filter_res:    w_file.write(line)   else:    del_count += 1  w_file.flush() os.remove('staff_data') os.rename('staff_data_bak', 'staff_data') return ['delete successful'], [str(del_count)]def update(sql_dic, title): ''' 处理update语句命令 :param sql_dic: :param title: :return: ''' set_l = sql_dic.get('set').strip().split(',') set_list = [i.split('=') for i in set_l] update_count = 0 with open('staff_data', 'r', encoding='utf-8') as r_file,/   open('staff_data_bak', 'w', encoding='utf-8') as w_file:  for line in r_file:   dic = dict(zip(title.split(','), line.strip().split(',')))   filter_res = logic_action(dic, sql_dic.get('where'))   if filter_res:    for i in set_list:     k = i[0]     v = i[-1]     dic[k] = v    line = [dic[i] for i in title.split(',')]    update_count += 1    line = ','.join(line)+'/n'   w_file.write(line)  w_file.flush() os.remove('staff_data') os.rename('staff_data_bak', 'staff_data') return ['update successful'], [str(update_count)]def where_action(fh, where_list, title): ''' 具体处理where_list里的所有条件 :param fh: :param where_list: :param title: :return: ''' res = [] if len(where_list) != 0:  for line in fh:   dic = dict(zip(title.split(','), line.strip().split(',')))   if dic['name'] != 'name':    logic_res = logic_action(dic, where_list)    if logic_res:     res.append(line.strip().split(',')) else:  res = [i.split(',') for i in fh.readlines()] return res passdef logic_action(dic, where_list): ''' 判断数据文件中每一条是否符合where_list条件 :param dic: :param where_list: :return: ''' logic = [] for exp in where_list:  if type(exp) is list:   exp_k, opt, exp_v = exp   if exp[1] == '=':    opt = '=='   logical_char = "'%s'%s'%s'" % (dic[exp_k], opt, exp_v)   if opt != 'like':    exp = str(eval(logical_char))   else:    if exp_v in dic[exp_k]:     exp = 'True'    else:     exp = 'False'  logic.append(exp) res = eval(' '.join(logic)) return resdef limit_action(filter_res, limit_l): ''' 用列表切分处理显示符合条件的数量 :param filter_res: :param limit_l: :return: ''' if limit_l:  index = int(limit_l[0])  res = filter_res[:index] else:  res = filter_res return resdef search_action(limit_res, select_list, title): ''' 处理需要查询并显示的title和相应数据 :param limit_res: :param select_list: :param title: :return: ''' res = [] fields_list = title.split(',') if select_list[0] == '*':  res = limit_res else:  fields_list = select_list  for data in limit_res:   dic = dict(zip(title.split(','), data))   r_l = []   for i in fields_list:    r_l.append((dic[i].strip()))   res.append(r_l) return fields_list, resif __name__ == '__main__': with open('staff_data', 'r', encoding='utf-8') as f: key_lis = ['select', 'insert', 'delete', 'update', 'from', 'into', 'set', 'values', 'where', 'limit'] while True:  sql = input('请输入sql命令,退出请输入exit:').strip()  sql = re.sub(' ', '', sql)  if len(sql) == 0:continue  if sql == 'exit':break  sql_dict = sql_parse(sql, key_lis)  fields_list, fields_data = sql_action(sql_dict, title)  print('/033[1;33m结果如下:/033[0m')  print('-'.join(fields_list))  for data in fields_data:   print('-'.join(data))

总结

以上就是这篇文章的全部内容了,希望本文的内容对大家的学习或者工作能带来一定的帮助,如果有疑问大家可以留言交流,谢谢大家对VEVB武林网的支持。

 
发表评论 共有条评论
用户名: 密码:
验证码: 匿名发表