从归属地数据库生成手机号正则
背景目标
手机号共11位,第一位是固定的:1;后8位是自由组合的数字;前3个数字代表了运营商的号段,为了保证正则的严谨和与时俱进,根据数据库中的数据来生成正则是个不错的选择。
当然生成正则的这种场景,比较适合于只能使用正则的情景;在可以不使用正则的情况下,直接截取前3位,判断是否在前缀的数组里就可以了。
思路
- 首先我们先从数据库获取到所有前3位的号段并去重
- 然后将前3位的数字解析成一棵树
- 合并相同的第3位的区间,比如
(13\d|18\d)
可以写成1[38]\d
- 合并连续区间, 比如
15[012356789]
可写成15[0-35-9]
- 使用元字符,比如第三步中的
[0-9]
可以写成\d
代码
找个手机归属地数据库:https://github.com/ls0f/phone
首先我们先将 dat 的数据文件导出成 txt 格式
def export(self):
for i in range(0, int(self.phone_record_count)):
current_offset = int(self.first_phone_record_offset + i * self.phone_fmt_length)
buffer = self.buf[current_offset: current_offset + self.phone_fmt_length]
cur_phone, record_offset, phone_type = struct.unpack(self.phone_fmt, buffer)
record_content = get_record_content(self.buf, record_offset)
yield Phone._format_phone_content(cur_phone, record_content, phone_type)
根据步骤1,列出所有的前三位
130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 145, 147, 149, 150, 151, 152, 153, 155, 156, 157, 158, 159, 170, 171, 172, 173, 175, 176, 177, 178, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189
根据步骤2,构建一棵树
threePrefix =[130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 145, 147, 149, 150, 151, 152, 153, 155, 156, 157, 158, 159, 170, 171, 172, 173, 175, 176, 177, 178, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189]
tree = {}
for prefix in threePrefix:
# first = prefix // 100 # always 1
second = prefix // 10 % 10
thirds = tree.setdefault(second, [])
third = prefix % 10
if third not in thirds:
thirds.append(third)
thirds.sort()
{3: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
4: [5, 7, 9],
5: [0, 1, 2, 3, 5, 6, 7, 8, 9],
7: [0, 1, 2, 3, 5, 6, 7, 8],
8: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]}
根据步骤3合并区间
def calcRegexRanges(nums):
start = nums[0]
end = nums[0]
xs = []
for s in nums:
if s > end + 1:
# print(v.get('seconds'), start, end)
if start == end:
xs.append(str(start))
elif end - start == 9:
xs.append('\\d')
else:
xs.append(str(start) + '-' + str(end))
start = s
end = start
else:
end = s
# print(v.get('seconds'), start, end)
if start == end:
xs.append(str(start))
elif end - start == 9:
xs.append('\\d')
else:
xs.append(str(start) + '-' + str(end))
# print('xs: ', xs)
if len(xs) == 1 and '-' not in xs[0]:
rs = xs[0]
else:
rs = '[' + ''.join(xs) + ']'
return rs
groupped = {}
for k,v in sorted(tree.items()):
i = groupped.setdefault(str(v), {})
i.setdefault('thirds', v)
i.setdefault('seconds', []).append(k)
segs = []
for k,v in sorted(groupped.items()):
ssp = calcRegexRanges(v.get('seconds'))
tsp = calcRegexRanges(v.get('thirds'))
segs.append(ssp + tsp)
regex = '1(' + '|'.join(segs) + ')\\d{8}'
print(regex)
结果
1([38]\d|5[0-35-9]|7[0-35-8]|4[579])\d{8}
浏览 14056 次 · 下载PDF
禁止转载