TA的每日心情 | 开心 2021-3-12 23:18 |
---|
签到天数: 2 天 [LV.1]初来乍到
|
这是一个web搜索的基本程序,从命令行输入搜索条件(起始的URL、处理url的最大数、要搜索的字符串),
它就会逐个对Internet上的URL进行实时搜索,查找并输出匹配搜索条件的页面。 这个程序的原型来自《java编程艺术》,
为了更好的分析,站长去掉了其中的GUI部分,并稍作修改以适用jdk1.5。以这个程序为基础,可以写出在互联网上搜索
诸如图像、邮件、网页下载之类的“爬虫”。
先请看程序运行的过程:
D:java>javac SearchCrawler.java(编译)
D:java>java SearchCrawler http://127.0.0.1:8080/zz3zcwbwebhome/index.jsp 20 java
Start searching...
result:
searchString=java
http://127.0.0.1:8080/zz3zcwbwebhome/index.jsp
http://127.0.0.1:8080/zz3zcwbwebhome/reply.jsp
http://127.0.0.1:8080/zz3zcwbwebhome/learn.jsp
http://127.0.0.1:8080/zz3zcwbwebhome/download.jsp
http://127.0.0.1:8080/zz3zcwbwebhome/article.jsp
http://127.0.0.1:8080/zz3zcwbwebhome/myexample/jlGUIOverview.htm
http://127.0.0.1:8080/zz3zcwbwebhome/myexample/Proxooldoc/index.HTML
http://127.0.0.1:8080/zz3zcwbwebhome/view.jsp?id=301
http://127.0.0.1:8080/zz3zcwbwebhome/view.jsp?id=297
http://127.0.0.1:8080/zz3zcwbwebhome/view.jsp?id=291
http://127.0.0.1:8080/zz3zcwbwebhome/view.jsp?id=286
http://127.0.0.1:8080/zz3zcwbwebhome/view.jsp?id=285
http://127.0.0.1:8080/zz3zcwbwebhome/view.jsp?id=284
http://127.0.0.1:8080/zz3zcwbwebhome/view.jsp?id=276
http://127.0.0.1:8080/zz3zcwbwebhome/view.jsp?id=272
又如:
D:java>java SearchCrawler http://www.sina.com 20 java
Start searching...
result:
searchString=java
http://sina.com
http://redirect.sina.com/WWW/sinaCN/www.sina.com.cn class=a2
http://redirect.sina.com/WWW/sinaCN/www.sina.com.cn class=a8
http://redirect.sina.com/WWW/sinaHK/www.sina.com.hk class=a2
http://redirect.sina.com/WWW/sinaTW/www.sina.com.tw class=a8
http://redirect.sina.com/WWW/sinaUS/home.sina.com class=a8
http://redirect.sina.com/WWW/smsCN/sms.sina.com.cn/ class=a2
http://redirect.sina.com/WWW/smsCN/sms.sina.com.cn/ class=a3
http://redirect.sina.com/WWW/sinaNet/www.sina.net/ class=a3
D:java>
下面是这个程序的源码
- import java.util.*;
- import java.net.*;
- import java.io.*;
- import java.util.regex.*;
- // 搜索Web爬行者
- public class SearchCrawler implements Runnable{
-
- /* disallowListCache缓存robot不允许搜索的URL。 Robot协议在Web站点的根目录下设置一个robots.txt文件,
- *规定站点上的哪些页面是限制搜索的。 搜索程序应该在搜索过程中跳过这些区域,下面是robots.txt的一个例子:
- # robots.txt for http://somehost.com/
- User-agent: *
- Disallow: /cgi-bin/
- Disallow: /registration # /Disallow robots on registration page
- Disallow: /login
- */
- private HashMap< String,ArrayList< String>> disallowListCache = new HashMap< String,ArrayList< String>>();
- ArrayList< String> errorList= new ArrayList< String>();//错误信息
- ArrayList< String> result=new ArrayList< String>(); //搜索到的结果
- String startUrl;//开始搜索的起点
- int maxUrl;//最大处理的url数
- String searchString;//要搜索的字符串(英文)
- boolean caseSensitive=false;//是否区分大小写
- boolean limitHost=false;//是否在限制的主机内搜索
-
- public SearchCrawler(String startUrl,int maxUrl,String searchString){
- this.startUrl=startUrl;
- this.maxUrl=maxUrl;
- this.searchString=searchString;
- }
- public ArrayList< String> getResult(){
- return result;
- }
- public void run(){//启动搜索线程
-
- crawl(startUrl,maxUrl, searchString,limitHost,caseSensitive);
- }
-
- //检测URL格式
- private URL verifyUrl(String url) {
- // 只处理HTTP URLs.
- if (!url.toLowerCase().startsWith("http://"))
- return null;
- URL verifiedUrl = null;
- try {
- verifiedUrl = new URL(url);
- } catch (Exception e) {
- return null;
- }
- return verifiedUrl;
- }
- // 检测robot是否允许访问给出的URL.
- private boolean isRobotAllowed(URL urlToCheck) {
- String host = urlToCheck.getHost().toLowerCase();//获取给出RUL的主机
- //System.out.println("主机="+host);
- // 获取主机不允许搜索的URL缓存
- ArrayList< String> disallowList =disallowListCache.get(host);
- // 如果还没有缓存,下载并缓存。
- if (disallowList == null) {
- disallowList = new ArrayList< String>();
- try {
- URL robotsFileUrl =new URL("http://" + host + "/robots.txt");
- BufferedReader reader =new BufferedReader(new InputStreamReader(robotsFileUrl.openStream()));
- // 读robot文件,创建不允许访问的路径列表。
- String line;
- while ((line = reader.readLine()) != null) {
- if (line.indexOf("Disallow:") == 0) {//是否包含"Disallow:"
- String disallowPath =line.substring("Disallow:".length());//获取不允许访问路径
- // 检查是否有注释。
- int commentIndex = disallowPath.indexOf("#");
- if (commentIndex != - 1) {
- disallowPath =disallowPath.substring(0, commentIndex);//去掉注释
- }
-
- disallowPath = disallowPath.trim();
- disallowList.add(disallowPath);
- }
- }
- // 缓存此主机不允许访问的路径。
- disallowListCache.put(host, disallowList);
- } catch (Exception e) {
- return true; //web站点根目录下没有robots.txt文件,返回真
- }
- }
-
- String file = urlToCheck.getFile();
- //System.out.println("文件getFile()="+file);
- for (int i = 0; i < disallowList.size(); i++) {
- String disallow = disallowList.get(i);
- if (file.startsWith(disallow)) {
- return false;
- }
- }
- return true;
- }
-
- private String downloadPage(URL pageUrl) {
- try {
- // Open connection to URL for reading.
- BufferedReader reader =
- new BufferedReader(new InputStreamReader(pageUrl.openStream()));
- // Read page into buffer.
- String line;
- StringBuffer pageBuffer = new StringBuffer();
- while ((line = reader.readLine()) != null) {
- pageBuffer.append(line);
- }
-
- return pageBuffer.toString();
- } catch (Exception e) {
- }
- return null;
- }
- // 从URL中去掉"www"
- private String removeWwwFromUrl(String url) {
- int index = url.indexOf("://www.");
- if (index != -1) {
- return url.substring(0, index + 3) +
- url.substring(index + 7);
- }
- return (url);
- }
- // 解析页面并找出链接
- private ArrayList< String> retrieveLinks(URL pageUrl, String pageContents, HashSet crawledList,
- boolean limitHost)
- {
- // 用正则表达式编译链接的匹配模式。
- Pattern p =Pattern.compile("< a\s+href\s*=\s*"?(.*?)["|>]",Pattern.CASE_INSENSITIVE);
- Matcher m = p.matcher(pageContents);
-
- ArrayList< String> linkList = new ArrayList< String>();
- while (m.find()) {
- String link = m.group(1).trim();
-
- if (link.length() < 1) {
- continue;
- }
- // 跳过链到本页面内链接。
- if (link.charAt(0) == "#") {
- continue;
- }
-
- if (link.indexOf("mailto:") != -1) {
- continue;
- }
-
- if (link.toLowerCase().indexOf("javascript") != -1) {
- continue;
- }
- if (link.indexOf("://") == -1){
- if (link.charAt(0) == "/") {//处理绝对地
- link = "http://" + pageUrl.getHost()+":"+pageUrl.getPort()+ link;
- } else {
- String file = pageUrl.getFile();
- if (file.indexOf("/") == -1) {//处理相对地址
- link = "http://" + pageUrl.getHost()+":"+pageUrl.getPort() + "/" + link;
- } else {
- String path =file.substring(0, file.lastIndexOf("/") + 1);
- link = "http://" + pageUrl.getHost() +":"+pageUrl.getPort()+ path + link;
- }
- }
- }
- int index = link.indexOf("#");
- if (index != -1) {
- link = link.substring(0, index);
- }
- link = removeWwwFromUrl(link);
- URL verifiedLink = verifyUrl(link);
- if (verifiedLink == null) {
- continue;
- }
- /* 如果限定主机,排除那些不合条件的URL*/
- if (limitHost &&
- !pageUrl.getHost().toLowerCase().equals(
- verifiedLink.getHost().toLowerCase()))
- {
- continue;
- }
- // 跳过那些已经处理的链接.
- if (crawledList.contains(link)) {
- continue;
- }
- linkList.add(link);
- }
- return (linkList);
- }
- // 搜索下载Web页面的内容,判断在该页面内有没有指定的搜索字符串
- private boolean searchStringMatches(String pageContents, String searchString, boolean caseSensitive){
- String searchContents = pageContents;
- if (!caseSensitive) {//如果不区分大小写
- searchContents = pageContents.toLowerCase();
- }
-
- Pattern p = Pattern.compile("[\s]+");
- String[] terms = p.split(searchString);
- for (int i = 0; i < terms.length; i++) {
- if (caseSensitive) {
- if (searchContents.indexOf(terms[i]) == -1) {
- return false;
- }
- } else {
- if (searchContents.indexOf(terms[i].toLowerCase()) == -1) {
- return false;
- }
- } }
- return true;
- }
-
- //执行实际的搜索操作
- public ArrayList< String> crawl(String startUrl, int maxUrls, String searchString,boolean limithost,boolean caseSensitive )
- {
-
- System.out.println("searchString="+searchString);
- HashSet< String> crawledList = new HashSet< String>();
- LinkedHashSet< String> toCrawlList = new LinkedHashSet< String>();
- if (maxUrls < 1) {
- errorList.add("Invalid Max URLs value.");
- System.out.println("Invalid Max URLs value.");
- }
-
-
- if (searchString.length() < 1) {
- errorList.add("Missing Search String.");
- System.out.println("Missing search String");
- }
-
- if (errorList.size() > 0) {
- System.out.println("err!!!");
- return errorList;
- }
-
- // 从开始URL中移出www
- startUrl = removeWwwFromUrl(startUrl);
-
- toCrawlList.add(startUrl);
- while (toCrawlList.size() > 0) {
-
- if (maxUrls != -1) {
- if (crawledList.size() == maxUrls) {
- break;
- }
- }
- // Get URL at bottom of the list.
- String url = toCrawlList.iterator().next();
- // Remove URL from the to crawl list.
- toCrawlList.remove(url);
- // Convert string url to URL object.
- URL verifiedUrl = verifyUrl(url);
- // Skip URL if robots are not allowed to access it.
- if (!isRobotAllowed(verifiedUrl)) {
- continue;
- }
-
- // 增加已处理的URL到crawledList
- crawledList.add(url);
- String pageContents = downloadPage(verifiedUrl);
-
- if (pageContents != null && pageContents.length() > 0){
- // 从页面中获取有效的链接
- ArrayList< String> links =retrieveLinks(verifiedUrl, pageContents, crawledList,limitHost);
-
- toCrawlList.addAll(links);
- if (searchStringMatches(pageContents, searchString,caseSensitive))
- {
- result.add(url);
- System.out.println(url);
- }
- }
-
- }
- return result;
- }
- // 主函数
- public static void main(String[] args) {
- if(args.length!=3){
- System.out.println("Usage:java SearchCrawler startUrl maxUrl searchString");
- return;
- }
- int max=Integer.parseInt(args[1]);
- SearchCrawler crawler = new SearchCrawler(args[0],max,args[2]);
- Thread search=new Thread(crawler);
- System.out.println("Start searching...");
- System.out.println("result:");
- search.start();
-
- }
- }
复制代码
function TempSave(ElementID)
{
CommentsPersistDiv.setAttribute("CommentContent",document.getElementById(ElementID).value);
CommentsPersistDiv.save("CommentXMLStore");
}
function Restore(ElementID)
{
CommentsPersistDiv.load("CommentXMLStore");
document.getElementById(ElementID).value=CommentsPersistDiv.getAttribute("CommentContent");
}
源码下载:http://file.javaxxz.com/2014/10/31/000017218.zip |
|