Added JID escaping support. SMACK-167

git-svn-id: http://svn.igniterealtime.org/svn/repos/smack/trunk@5195 b35dd754-fafc-0310-a699-88a17e54d16e
This commit is contained in:
Gaston Dombiak 2006-09-03 03:10:15 +00:00 committed by gato
parent 39662d3b0f
commit 8f98f71ecb
1 changed files with 141 additions and 0 deletions

View File

@ -127,6 +127,147 @@ public class StringUtils {
}
}
/**
* Escapes the node portion of a JID according to "JID Escaping" (JEP-0106).
* Escaping replaces characters prohibited by node-prep with escape sequences,
* as follows:<p>
*
* <table border="1">
* <tr><td><b>Unescaped Character</b></td><td><b>Encoded Sequence</b></td></tr>
* <tr><td>&lt;space&gt;</td><td>\20</td></tr>
* <tr><td>"</td><td>\22</td></tr>
* <tr><td>&</td><td>\26</td></tr>
* <tr><td>'</td><td>\27</td></tr>
* <tr><td>/</td><td>\2f</td></tr>
* <tr><td>:</td><td>\3a</td></tr>
* <tr><td>&lt;</td><td>\3c</td></tr>
* <tr><td>&gt;</td><td>\3e</td></tr>
* <tr><td>@</td><td>\40</td></tr>
* <tr><td>\</td><td>\5c</td></tr>
* </table><p>
*
* This process is useful when the node comes from an external source that doesn't
* conform to nodeprep. For example, a username in LDAP may be "Joe Smith". Because
* the &lt;space&gt; character isn't a valid part of a node, the username should
* be escaped to "Joe\20Smith" before being made into a JID (e.g. "joe\20smith@example.com"
* after case-folding, etc. has been applied).<p>
*
* All node escaping and un-escaping must be performed manually at the appropriate
* time; the JID class will not escape or un-escape automatically.
*
* @param node the node.
* @return the escaped version of the node.
*/
public static String escapeNode(String node) {
if (node == null) {
return null;
}
StringBuilder buf = new StringBuilder(node.length() + 8);
for (int i=0, n=node.length(); i<n; i++) {
char c = node.charAt(i);
switch (c) {
case '"': buf.append("\\22"); break;
case '&': buf.append("\\26"); break;
case '\'': buf.append("\\27"); break;
case '/': buf.append("\\2f"); break;
case ':': buf.append("\\3a"); break;
case '<': buf.append("\\3c"); break;
case '>': buf.append("\\3e"); break;
case '@': buf.append("\\40"); break;
case '\\': buf.append("\\5c"); break;
default: {
if (Character.isWhitespace(c)) {
buf.append("\\20");
}
else {
buf.append(c);
}
}
}
}
return buf.toString();
}
/**
* Un-escapes the node portion of a JID according to "JID Escaping" (JEP-0106).<p>
* Escaping replaces characters prohibited by node-prep with escape sequences,
* as follows:<p>
*
* <table border="1">
* <tr><td><b>Unescaped Character</b></td><td><b>Encoded Sequence</b></td></tr>
* <tr><td>&lt;space&gt;</td><td>\20</td></tr>
* <tr><td>"</td><td>\22</td></tr>
* <tr><td>&</td><td>\26</td></tr>
* <tr><td>'</td><td>\27</td></tr>
* <tr><td>/</td><td>\2f</td></tr>
* <tr><td>:</td><td>\3a</td></tr>
* <tr><td>&lt;</td><td>\3c</td></tr>
* <tr><td>&gt;</td><td>\3e</td></tr>
* <tr><td>@</td><td>\40</td></tr>
* <tr><td>\</td><td>\5c</td></tr>
* </table><p>
*
* This process is useful when the node comes from an external source that doesn't
* conform to nodeprep. For example, a username in LDAP may be "Joe Smith". Because
* the &lt;space&gt; character isn't a valid part of a node, the username should
* be escaped to "Joe\20Smith" before being made into a JID (e.g. "joe\20smith@example.com"
* after case-folding, etc. has been applied).<p>
*
* All node escaping and un-escaping must be performed manually at the appropriate
* time; the JID class will not escape or un-escape automatically.
*
* @param node the escaped version of the node.
* @return the un-escaped version of the node.
*/
public static String unescapeNode(String node) {
if (node == null) {
return null;
}
char [] nodeChars = node.toCharArray();
StringBuilder buf = new StringBuilder(nodeChars.length);
for (int i=0, n=nodeChars.length; i<n; i++) {
compare: {
char c = node.charAt(i);
if (c == '\\' && i+2<n) {
char c2 = nodeChars[i+1];
char c3 = nodeChars[i+2];
if (c2 == '2') {
switch (c3) {
case '0': buf.append(' '); i+=2; break compare;
case '2': buf.append('"'); i+=2; break compare;
case '6': buf.append('&'); i+=2; break compare;
case '7': buf.append('\''); i+=2; break compare;
case 'f': buf.append('/'); i+=2; break compare;
}
}
else if (c2 == '3') {
switch (c3) {
case 'a': buf.append(':'); i+=2; break compare;
case 'c': buf.append('<'); i+=2; break compare;
case 'e': buf.append('>'); i+=2; break compare;
}
}
else if (c2 == '4') {
if (c3 == '0') {
buf.append("@");
i+=2;
break compare;
}
}
else if (c2 == '5') {
if (c3 == 'c') {
buf.append("\\");
i+=2;
break compare;
}
}
}
buf.append(c);
}
}
return buf.toString();
}
/**
* Escapes all necessary characters in the String so that it can be used
* in an XML doc.