·您的位置: 首页 » 资源教程 » 编程开发 » JAVA、JSP » JavaScript escape/unescape编码的Java实现

JavaScript escape/unescape编码的Java实现

类别: JAVA教程  评论数:0 总得分:0
 
符合 JavaScript unescape 函数(支持Unicode)的 Java escape 编码实现
最近在弄一个跨域的交互程序,交互过程非常复杂.而且,其中一端只能使用 javascript 来处理返回. 

于是,发现 Java 的 URLEncoder.encode 方法与 JavaScript 的 unescape 方法已经不能兼容了. 

现在浏览器的主流版本中 JavaScript 已经开始使用 Unicode 作为字符串的内码进行编码.即, Unicode 字符 escape 之后的样子类似于 %uABCD 而 java 的 URLEncoder.encode 方法得到的结果是 %AB%CD ,转换出来的结果必定是乱码了. 

方法有两个,其一,用 Java 实现一个生成新编码的方法.其二,用 JavaScript 实现一个针对旧编码的解码方法.想了想,还是用 Java 要方便一点点吧. 



上次贴出的 escape 的 java 实现,似乎对大家有用.最近曾军同学提出 unescape 对于 javascript 与后端的交互也有意义.于是就一并实现了.再次贴上来. 

代码非常简单,注释就免了吧.一贯的,仍然遵循\"以空间换时间\"原则. 

需要的桶子,一贯的,help yourself. 

/**
 * JavaScript escape/unescape 编码的 Java 实现
 * author jackyz
 * keep this copyright info while using this method by free
 */
public class Escape {
    private final static String[] hex = {
        \"00\",\"01\",\"02\",\"03\",\"04\",\"05\",\"06\",\"07\",\"08\",\"09\",\"0A\",\"0B\",\"0C\",\"0D\",\"0E\",\"0F\",
        \"10\",\"11\",\"12\",\"13\",\"14\",\"15\",\"16\",\"17\",\"18\",\"19\",\"1A\",\"1B\",\"1C\",\"1D\",\"1E\",\"1F\",
        \"20\",\"21\",\"22\",\"23\",\"24\",\"25\",\"26\",\"27\",\"28\",\"29\",\"2A\",\"2B\",\"2C\",\"2D\",\"2E\",\"2F\",
        \"30\",\"31\",\"32\",\"33\",\"34\",\"35\",\"36\",\"37\",\"38\",\"39\",\"3A\",\"3B\",\"3C\",\"3D\",\"3E\",\"3F\",
        \"40\",\"41\",\"42\",\"43\",\"44\",\"45\",\"46\",\"47\",\"48\",\"49\",\"4A\",\"4B\",\"4C\",\"4D\",\"4E\",\"4F\",
        \"50\",\"51\",\"52\",\"53\",\"54\",\"55\",\"56\",\"57\",\"58\",\"59\",\"5A\",\"5B\",\"5C\",\"5D\",\"5E\",\"5F\",
        \"60\",\"61\",\"62\",\"63\",\"64\",\"65\",\"66\",\"67\",\"68\",\"69\",\"6A\",\"6B\",\"6C\",\"6D\",\"6E\",\"6F\",
        \"70\",\"71\",\"72\",\"73\",\"74\",\"75\",\"76\",\"77\",\"78\",\"79\",\"7A\",\"7B\",\"7C\",\"7D\",\"7E\",\"7F\",
        \"80\",\"81\",\"82\",\"83\",\"84\",\"85\",\"86\",\"87\",\"88\",\"89\",\"8A\",\"8B\",\"8C\",\"8D\",\"8E\",\"8F\",
        \"90\",\"91\",\"92\",\"93\",\"94\",\"95\",\"96\",\"97\",\"98\",\"99\",\"9A\",\"9B\",\"9C\",\"9D\",\"9E\",\"9F\",
        \"A0\",\"A1\",\"A2\",\"A3\",\"A4\",\"A5\",\"A6\",\"A7\",\"A8\",\"A9\",\"AA\",\"AB\",\"AC\",\"AD\",\"AE\",\"AF\",
        \"B0\",\"B1\",\"B2\",\"B3\",\"B4\",\"B5\",\"B6\",\"B7\",\"B8\",\"B9\",\"BA\",\"BB\",\"BC\",\"BD\",\"BE\",\"BF\",
        \"C0\",\"C1\",\"C2\",\"C3\",\"C4\",\"C5\",\"C6\",\"C7\",\"C8\",\"C9\",\"CA\",\"CB\",\"CC\",\"CD\",\"CE\",\"CF\",
        \"D0\",\"D1\",\"D2\",\"D3\",\"D4\",\"D5\",\"D6\",\"D7\",\"D8\",\"D9\",\"DA\",\"DB\",\"DC\",\"DD\",\"DE\",\"DF\",
        \"E0\",\"E1\",\"E2\",\"E3\",\"E4\",\"E5\",\"E6\",\"E7\",\"E8\",\"E9\",\"EA\",\"EB\",\"EC\",\"ED\",\"EE\",\"EF\",
        \"F0\",\"F1\",\"F2\",\"F3\",\"F4\",\"F5\",\"F6\",\"F7\",\"F8\",\"F9\",\"FA\",\"FB\",\"FC\",\"FD\",\"FE\",\"FF\"
    };
    private final static byte[] val = {
        0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x00,0x01,0x02,0x03,0x04,0x05,0x06,0x07,0x08,0x09,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x0A,0x0B,0x0C,0x0D,0x0E,0x0F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x0A,0x0B,0x0C,0x0D,0x0E,0x0F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,
        0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F,0x3F
    };
    public static String escape(String s) {
        StringBuffer sbuf = new StringBuffer();
        int len = s.length();
        for (int i = 0; i < len; i++) {
            int ch = s.charAt(i);
            if (ch == \' \') {                        // space : map to \'+\' 
                sbuf.append(\'+\');
            } else if (\'A\' <= ch && ch <= \'Z\') {    // \'A\'..\'Z\' : as it was
                sbuf.append((char)ch);
            } else if (\'a\' <= ch && ch <= \'z\') {    // \'a\'..\'z\' : as it was
                sbuf.append((char)ch);
            } else if (\'0\' <= ch && ch <= \'9\') {    // \'0\'..\'9\' : as it was
                sbuf.append((char)ch);
            } else if (ch == \'-\' || ch == \'_\'       // unreserved : as it was
                || ch == \'.\' || ch == \'!\'
                || ch == \'~\' || ch == \'*\'
                || ch == \'/\'\' || ch == \'(\'
                || ch == \')\') {
                sbuf.append((char)ch);
            } else if (ch <= 0x007F) {              // other ASCII : map to %XX
                sbuf.append(\'%\');
                sbuf.append(hex[ch]);
            } else {                                // unicode : map to %uXXXX
                sbuf.append(\'%\');
                sbuf.append(\'u\');
                sbuf.append(hex[(ch >>> 8)]);
                sbuf.append(hex[(0x00FF & ch)]);
            }
        }
        return sbuf.toString();
    }
    public static String unescape(String s) {
        StringBuffer sbuf = new StringBuffer();
        int i = 0;
        int len = s.length();
        while (i < len) {
            int ch = s.charAt(i);
            if (ch == \'+\') {                        // + : map to \' \' 
                sbuf.append(\' \');
            } else if (\'A\' <= ch && ch <= \'Z\') {    // \'A\'..\'Z\' : as it was
                sbuf.append((char)ch);
            } else if (\'a\' <= ch && ch <= \'z\') {    // \'a\'..\'z\' : as it was
                sbuf.append((char)ch);
            } else if (\'0\' <= ch && ch <= \'9\') {    // \'0\'..\'9\' : as it was
                sbuf.append((char)ch);
            } else if (ch == \'-\' || ch == \'_\'       // unreserved : as it was
                || ch == \'.\' || ch == \'!\'
                || ch == \'~\' || ch == \'*\'
                || ch == \'/\'\' || ch == \'(\'
                || ch == \')\') {
                sbuf.append((char)ch);
            } else if (ch == \'%\') {
                int cint = 0;
                if (\'u\' != s.charAt(i+1)) {         // %XX : map to ascii(XX)
                    cint = (cint << 4) | val[s.charAt(i+1)];
                    cint = (cint << 4) | val[s.charAt(i+2)];
                    i+=2;
                } else {                            // %uXXXX : map to unicode(XXXX)
                    cint = (cint << 4) | val[s.charAt(i+2)];
                    cint = (cint << 4) | val[s.charAt(i+3)];
                    cint = (cint << 4) | val[s.charAt(i+4)];
                    cint = (cint << 4) | val[s.charAt(i+5)];
                    i+=5;
                }
                sbuf.append((char)cint);
            }
            i++;
        }
        return sbuf.toString();
    }
    public static void main(String[] args) {
        String stest = \"中文1234 abcd[]()<+>,.~//\";
        System.out.println(stest);
        System.out.println(escape(stest));
        System.out.println(unescape(escape(stest)));
    }
}




-= 资 源 教 程 =-
文 章 搜 索
关键词:
类型:
范围:
纯粹空间 softpure.com
Copyright © 2006-2008 暖阳制作 版权所有
QQ: 15242663 (拒绝闲聊)  Email: faisun@sina.com
 纯粹空间 - 韩国酷站|酷站欣赏|教程大全|资源下载|免费博客|美女壁纸|设计素材|技术论坛   Valid XHTML 1.0 Transitional
百度搜索 谷歌搜索 Alexa搜索 | 粤ICP备19116064号-1