实例Python处理XML文件的方法
程序员文章站
2022-03-14 21:29:34
...
需求
有一个表,里面数据量比较大,每天一更新,其字段可以通过xml配置文件进行配置,即,可能每次建表的字段不一样。
有一个表,里面数据量比较大,每天一更新,其字段可以通过xml配置文件进行配置,即,可能每次建表的字段不一样。
上游跑时会根据配置从源文件中提取,到入库这一步需要根据配置进行建表。
解决
写了一个简单的xml,配置需要字段及类型
上游读取到对应的数据
入库这一步,先把原表删除,根据配置建新表
XML文件
处理脚本
#!/usr/bin/python # -*- coding:utf-8 -*- #author: wklken #desc: use to read db xml config. #----------------------- #2012-02-18 created #---------------------- import sys,os from xml.dom import minidom, Node def read_dbconfig_xml(xml_file_path): content = {} root = minidom.parse(xml_file_path) table = root.getElementsByTagName("table")[0] #read dbname and table name. table_name = table.getAttribute("name") db_name = table.getAttribute("db_name") if len(table_name) > 0 and len(db_name) > 0: db_sql = "create database if not exists `" + db_name +"`; use " + db_name + ";" table_drop_sql = "drop " + table_name + " if exists " + table_name + ";" content.update({"db_sql" : db_sql}) content.update({"table_sql" : table_drop_sql }) else: print "Error:attribute is not define well! db_name=" + db_name + " ;table_name=" + table_name sys.exit(1) #print table_name, db_name table_create_sql = "create table " + table_name +"(" #read primary cell primary_key = table.getElementsByTagName("primary_key")[0] primary_key_name = primary_key.getElementsByTagName("name")[0].childNodes[0].nodeValue table_create_sql += primary_key_name + " INTEGER NOT NULL AUTO_INCREMENT PRIMARY KEY," #print primary_key.toxml() #read ordernary field fields = table.getElementsByTagName("field") f_index = 0 for field in fields: f_index += 1 name = field.getElementsByTagName("name")[0].childNodes[0].nodeValue type = field.getElementsByTagName("type")[0].childNodes[0].nodeValue table_create_sql += name + " " + type if f_index != len(fields): table_create_sql += "," is_index = field.getElementsByTagName("is_index")[0].childNodes[0].nodeValue table_create_sql += ");" content.update({"table_create_sql" : table_create_sql}) #character set latin1 collate latin1_danish_ci; print content if __name__ == "__main__": read_dbconfig_xml(sys.argv[1])
PYTHON解析XML大文件[SAX]
需求
读取xml数据文件,文件较大,需要实时处理插入到数据库
xml文档
100000 男 北京,海淀区437 1989 333 242 null 大学 1 2月14日
处理
sax处理时并不会像dom一样可以以类似节点的维度进行读取,它只有 开始标签 内容 结束标签 之分
处理思想是:通过一个handler,对开始标签,内容,结束标签各有一个处理函数
代码及注解
person 处理类
from xml.sax import handler,parseString class PersonHandler(handler.ContentHandler): def __init__(self, db_ops): #db op obj self.db_ops = db_ops #存储一个person的map self.person = {} #当前的tag self.current_tag = "" #是否是tag之间的内容 ,目的拿到tag间内容,不受空白的干扰 self.in_quote = 0 #开始,清空map def startElement(self, name, attr): #以person,清空map if name == "person": self.person = {} #记录 状态 self.current_tag = name self.in_quote = 1 #结束,插入数据库 def endElement(self, name): #以person结尾 代表读取一个person的信息结束 if name == "person": #do something in_fields = tuple([ ('"' + self.person.get(i,"") + '"') for i in fields ]) print in_sql % in_fields db_ops.insert( in_sql%(in_fields)) #处理 self.in_quote = 0 def characters(self, content): #若是在tag之间的内容,更新到map中 if self.in_quote: self.person.update({self.current_tag: content})
加上入库的完整代码
#!/usr/bin/python # -*- coding:utf-8 -*- #parse_person.py #version : 0.1 #author : wukunliang@163.com #desc : parse person.xml and out sql import sys,os import MySQLdb reload(sys) sys.setdefaultencoding('utf-8') in_sql = "insert into person(id,sex,address,fansNum,summary,wbNum,gzNum,blog,edu,work,renZh,brithday) values(%s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s)" fields = ("id","sex","address","fansNum","summary","wbNum","gzNum","blog","edu","work","renZh","brithday") #数据库方法 class Db_Connect: def __init__(self, db_host, user, pwd, db_name, charset="utf8", use_unicode = True): print "init begin" print db_host, user, pwd, db_name, charset , use_unicode self.conn = MySQLdb.Connection(db_host, user, pwd, db_name, charset=charset , use_unicode=use_unicode) print "init end" def insert(self, sql): try: n = self.conn.cursor().execute(sql) return n except MySQLdb.Warning, e: print "Error: execute sql '",sql,"' failed" def close(self): self.conn.close() #person 处理类 from xml.sax import handler,parseString class PersonHandler(handler.ContentHandler): def __init__(self, db_ops): #db op obj self.db_ops = db_ops #存储一个person的map self.person = {} #当前的tag self.current_tag = "" #是否是tag之间的内容 self.in_quote = 0 #开始,清空map def startElement(self, name, attr): #以person,清空map if name == "person": self.person = {} #记录 状态 self.current_tag = name self.in_quote = 1 #结束,插入数据库 def endElement(self, name): #以person结尾 代表读取一个person的信息结束 if name == "person": #do something in_fields = tuple([ ('"' + self.person.get(i,"") + '"') for i in fields ]) print in_sql % in_fields db_ops.insert( in_sql%(in_fields)) #处理 self.in_quote = 0 def characters(self, content): #若是在tag之间的内容,更新到map中 if self.in_quote: self.person.update({self.current_tag: content}) if __name__ == "__main__": f = open("./person.xml") #如果源文件gbk 转码 若是utf-8,去掉decode.encode db_ops = Db_Connect("127.0.0.1", "root", "root", "test") parseString(f.read().decode("gbk").encode("utf-8"), PersonHandler(db_ops)) f.close() db_ops.close()