您好,登录后才能下订单哦!
密码登录
登录注册
点击 登录注册 即表示同意《亿速云用户服务条款》
这篇文章主要介绍了如何使用python实现词法分析器,具有一定借鉴价值,感兴趣的朋友可以参考下,希望大家阅读完这篇文章之后大有收获,下面让小编带着大家一起了解一下。
具体内容如下
词法分析器状态转换图:
词法分析器总流程图:
预处理程序:
词法分析器:
词法分析器程序详细设计
详细代码实现:
#!/usr/bin/env python3.4 # coding=utf-8 import sys import string keywards = {} # 关键字部分 keywards['False'] = 101 keywards['class'] = 102 keywards['finally'] = 103 keywards['is'] = 104 keywards['return'] = 105 keywards['None'] = 106 keywards['continue'] = 107 keywards['for'] = 108 keywards['lambda'] = 109 keywards['try'] = 110 keywards['True'] = 111 keywards['def'] = 112 keywards['from'] = 113 keywards['nonlocal'] = 114 keywards['while'] = 115 keywards['and'] = 116 keywards['del'] = 117 keywards['global'] = 118 keywards['not'] = 119 keywards['with'] = 120 keywards['as'] = 121 keywards['elif'] = 122 keywards['if'] = 123 keywards['or'] = 124 keywards['yield'] = 125 keywards['assert'] = 126 keywards['else'] = 127 keywards['import'] = 128 keywards['pass'] = 129 keywards['break'] = 130 keywards['except'] = 131 keywards['in'] = 132 keywards['raise'] = 133 # 符号 keywards['+'] = 201 keywards['-'] = 202 keywards['*'] = 203 keywards['/'] = 204 keywards['='] = 205 keywards[':'] = 206 keywards['<'] = 207 keywards['>'] = 208 keywards['%'] = 209 keywards['&'] = 210 keywards['!'] = 211 keywards['('] = 212 keywards[')'] = 213 keywards['['] = 214 keywards[']'] = 215 keywards['{'] = 216 keywards['}'] = 217 keywards['#'] = 218 keywards['|'] = 219 keywards[','] = 220 # 变量 # keywards['var'] = 301 # 常量 # keywards['const'] = 401 # Error # keywards['const'] = 501 signlist = {} # 预处理函数,将文件中的空格,换行等无关字符处理掉 def pretreatment(file_name): try: fp_read = open(file_name, 'r') fp_write = open('file.tmp', 'w') sign = 0 while True: read = fp_read.readline() if not read: break length = len(read) i = -1 while i < length - 1: i += 1 if sign == 0: if read[i] == ' ': continue if read[i] == '#': break elif read[i] == ' ': if sign == 1: continue else: sign = 1 fp_write.write(' ') elif read[i] == '\t': if sign == 1: continue else: sign = 1 fp_write.write(' ') elif read[i] == '\n': if sign == 1: continue else: fp_write.write(' ') sign = 1 elif read[i] == '"': fp_write.write(read[i]) i += 1 while i < length and read[i] != '"': fp_write.write(read[i]) i += 1 if i >= length: break fp_write.write(read[i]) elif read[i] == "'": fp_write.write(read[i]) i += 1 while i < length and read[i] != "'": fp_write.write(read[i]) i += 1 if i >= length: break fp_write.write(read[i]) else: sign = 3 fp_write.write(read[i]) except Exception: print(file_name, ': This FileName Not Found!') def save(string): if string in keywards.keys(): if string not in signlist.keys(): signlist[string] = keywards[string] else: try: float(string) save_const(string) except ValueError: save_var(string) def save_var(string): if string not in signlist.keys(): if len(string.strip()) < 1: pass else: if is_signal(string) == 1: signlist[string] = 301 else: signlist[string] = 501 def save_const(string): if string not in signlist.keys(): signlist[string] = 401 def save_error(string): if string not in signlist.keys(): signlist[string] = 501 def is_signal(s): if s[0] == '_' or s[0] in string.ascii_letters: for i in s: if i in string.ascii_letters or i == '_' or i in string.digits: pass else: return 0 return 1 else: return 0 def recognition(filename): try: fp_read = open(filename, 'r') string = "" sign = 0 while True: read = fp_read.read(1) if not read: break if read == ' ': if len(string.strip()) < 1: sign = 0 pass else: if sign == 1 or sign == 2: string += read else: save(string) string = "" sign = 0 elif read == '(': if sign == 1 or sign == 2: string += read else: save(string) string = "" save('(') elif read == ')': if sign == 1 or sign == 2: string += read else: save(string) string = "" save(')') elif read == '[': if sign == 1 or sign == 2: string += read else: save(string) string = "" save('[') elif read == ']': if sign == 1 or sign == 2: string += read else: save(string) string = "" save(']') elif read == '{': if sign == 1 or sign == 2: string += read else: save(string) string = "" save('{') elif read == '}': if sign == 1 or sign == 2: string += read else: save(string) string = "" save('}') elif read == '<': save(string) string = "" save('<') elif read == '>': save(string) string = "" save('>') elif read == ',': save(string) string = "" save(',') elif read == "'": string += read if sign == 1: sign = 0 save_const(string) string = "" else: if sign != 2: sign = 1 elif read == '"': string += read if sign == 2: sign = 0 save_const(string) string = "" else: if sign != 1: sign = 2 elif read == ':': if sign == 1 or sign == 2: string += read else: save(string) string = "" save(':') elif read == '+': save(string) string = "" save('+') elif read == '=': save(string) string = "" save('=') else: string += read except Exception as e: print(e) def main(): if len(sys.argv) < 2: print("Please Input FileName") else: pretreatment(sys.argv[1]) recognition('file.tmp') for i in signlist.keys(): print("(", signlist[i], ",", i, ")") if __name__ == '__main__': main()
感谢你能够认真阅读完这篇文章,希望小编分享的“如何使用python实现词法分析器”这篇文章对大家有帮助,同时也希望大家多多支持亿速云,关注亿速云行业资讯频道,更多相关知识等着你来学习!
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:is@yisu.com进行举报,并提供相关证据,一经查实,将立刻删除涉嫌侵权内容。