class org.dom4j.io.SAXReader
interface org.dom4j.Document
interface org.dom4j.Node
interface org.dom4j.Element
interface org.dom4j.Attribute
interface org.dom4j.Text
interface org.dom4j.CDATA
interface org.dom4j.Comment
getText 获取注释
//先加入dom4j.jar包 import java.util.HashMap; import java.util.Iterator; import java.util.Map; import org.dom4j.Document; import org.dom4j.DocumentException; import org.dom4j.DocumentHelper; import org.dom4j.Element; public class TestDom4j { public void readStringXml(String xml) { Document doc = null; try { // 读取并解析XML文档 // SAXReader就是一个管道,用一个流的方式,把xml文件读出来 // // SAXReader reader = new SAXReader(); //User.hbm.xml表示你要解析的xml文档 // Document document = reader.read(new File("User.hbm.xml")); // 下面的是通过解析xml字符串的 doc = DocumentHelper.parseText(xml); // 将字符串转为XML Element rootElt = doc.getRootElement(); // 获取根节点 System.out.println("根节点:" + rootElt.getName()); // 拿到根节点的名称 Iterator iter = rootElt.elementIterator("head"); // 获取根节点下的子节点head // 遍历head节点 while (iter.hasNext()) { Element recordEle = (Element) iter.next(); String title = recordEle.elementTextTrim("title"); // 拿到head节点下的子节点title值 System.out.println("title:" + title); Iterator iters = recordEle.elementIterator("script"); // 获取子节点head下的子节点script // 遍历Header节点下的Response节点 while (iters.hasNext()) { Element itemEle = (Element) iters.next(); String username = itemEle.elementTextTrim("username"); // 拿到head下的子节点script下的字节点username的值 String password = itemEle.elementTextTrim("password"); System.out.println("username:" + username); System.out.println("password:" + password); } } Iterator iterss = rootElt.elementIterator("body"); ///获取根节点下的子节点body // 遍历body节点 while (iterss.hasNext()) { Element recordEless = (Element) iterss.next(); String result = recordEless.elementTextTrim("result"); // 拿到body节点下的子节点result值 System.out.println("result:" + result); Iterator itersElIterator = recordEless.elementIterator("form"); // 获取子节点body下的子节点form // 遍历Header节点下的Response节点 while (itersElIterator.hasNext()) { Element itemEle = (Element) itersElIterator.next(); String banlce = itemEle.elementTextTrim("banlce"); // 拿到body下的子节点form下的字节点banlce的值 String subID = itemEle.elementTextTrim("subID"); System.out.println("banlce:" + banlce); System.out.println("subID:" + subID); } } } catch (DocumentException e) { e.printStackTrace(); } catch (Exception e) { e.printStackTrace(); } } /** * @description 将xml字符串转换成map * @param xml * @return Map */ public static Map readStringXmlOut(String xml) { Map map = new HashMap(); Document doc = null; try { // 将字符串转为XML doc = DocumentHelper.parseText(xml); // 获取根节点 Element rootElt = doc.getRootElement(); // 拿到根节点的名称 System.out.println("根节点:" + rootElt.getName()); // 获取根节点下的子节点head Iterator iter = rootElt.elementIterator("head"); // 遍历head节点 while (iter.hasNext()) { Element recordEle = (Element) iter.next(); // 拿到head节点下的子节点title值 String title = recordEle.elementTextTrim("title"); System.out.println("title:" + title); map.put("title", title); // 获取子节点head下的子节点script Iterator iters = recordEle.elementIterator("script"); // 遍历Header节点下的Response节点 while (iters.hasNext()) { Element itemEle = (Element) iters.next(); // 拿到head下的子节点script下的字节点username的值 String username = itemEle.elementTextTrim("username"); String password = itemEle.elementTextTrim("password"); System.out.println("username:" + username); System.out.println("password:" + password); map.put("username", username); map.put("password", password); } } //获取根节点下的子节点body Iterator iterss = rootElt.elementIterator("body"); // 遍历body节点 while (iterss.hasNext()) { Element recordEless = (Element) iterss.next(); // 拿到body节点下的子节点result值 String result = recordEless.elementTextTrim("result"); System.out.println("result:" + result); // 获取子节点body下的子节点form Iterator itersElIterator = recordEless.elementIterator("form"); // 遍历Header节点下的Response节点 while (itersElIterator.hasNext()) { Element itemEle = (Element) itersElIterator.next(); // 拿到body下的子节点form下的字节点banlce的值 String banlce = itemEle.elementTextTrim("banlce"); String subID = itemEle.elementTextTrim("subID"); System.out.println("banlce:" + banlce); System.out.println("subID:" + subID); map.put("result", result); map.put("banlce", banlce); map.put("subID", subID); } } } catch (DocumentException e) { e.printStackTrace(); } catch (Exception e) { e.printStackTrace(); } return map; } public static void main(String[] args) { // 下面是需要解析的xml字符串例子 String xmlString = "<html>" + "<head>" + "<title>dom4j解析一个例子</title>" + "<script>" + "<username>yangrong</username>" + "<password>123456</password>" + "</script>" + "</head>" + "<body>" + "<result>0</result>" + "<form>" + "<banlce>1000</banlce>" + "<subID>36242519880716</subID>" + "</form>" + "</body>" + "</html>"; /* * Test2 test = new Test2(); test.readStringXml(xmlString); */ Map map = readStringXmlOut(xmlString); Iterator iters = map.keySet().iterator(); while (iters.hasNext()) { String key = iters.next().toString(); // 拿到键 String val = map.get(key).toString(); // 拿到值 System.out.println(key + "=" + val); } } }
/** * 解析包含有DB连接信息的XML文件 * 格式必须符合如下规范: * 1. 最多三级,每级的node名称自定义; * 2. 二级节点支持节点属性,属性将被视作子节点; * 3. CDATA必须包含在节点中,不能单独出现。 * * 示例1——三级显示: * <db-connections> * <connection> * <name>DBTest</name> * <jndi></jndi> * <url> * <![CDATA[jdbc:mysql://localhost:3306/db_test?useUnicode=true&characterEncoding=UTF8]]> * </url> * <driver>org.gjt.mm.mysql.Driver</driver> * <user>test</user> * <password>test2012</password> * <max-active>10</max-active> * <max-idle>10</max-idle> * <min-idle>2</min-idle> * <max-wait>10</max-wait> * <validation-query>SELECT 1+1</validation-query> * </connection> * </db-connections> * * 示例2——节点属性: * <bookstore> * <book category="cooking"> * <title lang="en">Everyday Italian</title> * <author>Giada De Laurentiis</author> * <year>2005</year> * <price>30.00</price> * </book> * * <book category="children" title="Harry Potter" author="J K. Rowling" year="2005" price="$29.9"/> * </bookstore> * * @param configFile * @return * @throws Exception */ public static List<Map<String, String>> parseDBXML(String configFile) throws Exception { List<Map<String, String>> dbConnections = new ArrayList<Map<String, String>>(); InputStream is = Parser.class.getResourceAsStream(configFile); SAXReader saxReader = new SAXReader(); Document document = saxReader.read(is); Element connections = document.getRootElement(); Iterator<Element> rootIter = connections.elementIterator(); while (rootIter.hasNext()) { Element connection = rootIter.next(); Iterator<Element> childIter = connection.elementIterator(); Map<String, String> connectionInfo = new HashMap<String, String>(); List<Attribute> attributes = connection.attributes(); for (int i = 0; i < attributes.size(); ++i) { // 添加节点属性 connectionInfo.put(attributes.get(i).getName(), attributes.get(i).getValue()); } while (childIter.hasNext()) { // 添加子节点 Element attr = childIter.next(); connectionInfo.put(attr.getName().trim(), attr.getText().trim()); } dbConnections.add(connectionInfo); } return dbConnections; }