com.ctc.wstx.io.CharsetNames Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of woodstox-core Show documentation
Show all versions of woodstox-core Show documentation
Woodstox is a high-performance XML processor that implements Stax (JSR-173),
SAX2 and Stax2 APIs
package com.ctc.wstx.io;
import java.io.OutputStreamWriter;
import java.io.Writer;
import com.ctc.wstx.util.StringUtil;
/**
* Simple utility class that normalizes given character input character
* set names into canonical (within Woodstox, anyways) names.
*/
public final class CharsetNames
{
/*
//////////////////////////////////////////////////
// Canonical names used internally
//////////////////////////////////////////////////
*/
// // // Unicode variants:
public final static String CS_US_ASCII = "US-ASCII";
public final static String CS_UTF8 = "UTF-8";
/**
* This constants is intentionally vague, so that some other information
* will be needed to determine the endianness.
*/
public final static String CS_UTF16 = "UTF-16";
public final static String CS_UTF16BE = "UTF-16BE";
public final static String CS_UTF16LE = "UTF-16LE";
public final static String CS_UTF32 = "UTF-32";
public final static String CS_UTF32BE = "UTF-32BE";
public final static String CS_UTF32LE = "UTF-32LE";
// // // 8-bit ISO encodings:
public final static String CS_ISO_LATIN1 = "ISO-8859-1";
// // // Japanese non-unicode encodings:
public final static String CS_SHIFT_JIS = "Shift_JIS";
// // // Other oddities:
/* There are tons of EBCDIC varieties, with similar but
* non-identical names. As a result, we can not give or use
* just a single canonical name for general use.
* However, we can choose a single one to use for bootstrapping;
* that is, for parsing xml declaration to know the "real"
* EBCDIC variant.
*/
public final static String CS_EBCDIC_SUBSET = "IBM037";
/*
//////////////////////////////////////////////////
// Utility methods
//////////////////////////////////////////////////
*/
public static String normalize(String csName)
{
if (csName == null || csName.length() < 3) {
return csName;
}
/* Canonical charset names here are from IANA recommendation:
* http://www.iana.org/assignments/character-sets
* but comparison is done loosely (case-insensitive, ignoring
* spacing, underscore vs. hyphen etc) to try to make detection
* as extensive as possible.
*/
/* But first bit of pre-filtering: it seems like 'cs' prefix
* is applicable to pretty much all actual encodings (as per
* IANA recommendations; csASCII, csUcs4 etc). So, let's just
* strip out the prefix if so
*/
boolean gotCsPrefix = false;
char c = csName.charAt(0);
if (c == 'c' || c == 'C'){
char d = csName.charAt(1);
if (d == 's' || d == 'S') {
csName = csName.substring(2);
c = csName.charAt(0);
gotCsPrefix = true;
}
}
switch (c) {
case 'a':
case 'A':
if (StringUtil.equalEncodings(csName, "ASCII")) {
return CS_US_ASCII;
}
break;
case 'c':
case 'C':
/* Tons of variants: let's assume 'cpXXX' is an EBCDIC
* variant, and should read 'IBMXXX'
*/
if (StringUtil.encodingStartsWith(csName, "cp")) {
return "IBM" + StringUtil.trimEncoding(csName, true).substring(2);
}
// Hmmh. There are boatloads of these... but what to do with them?
if (StringUtil.encodingStartsWith(csName, "cs")) {
// Well, "csIBMxx" means EBCDIC of "IBMxx"
if (StringUtil.encodingStartsWith(csName, "csIBM")) {
// So let's just peel off "cs" prefix:
return StringUtil.trimEncoding(csName, true).substring(2);
}
// !!! TBI
}
break;
case 'e':
case 'E':
if (csName.startsWith("EBCDIC-CP-") ||
csName.startsWith("ebcdic-cp-")) {
// EBCDIC, but which variety?
// Let's trim out prefix to make comparison easier:
String type = StringUtil.trimEncoding(csName, true).substring(8);
// Note: these are suggested encodings of Xerces
if (type.equals("US") || type.equals("CA")
|| type.equals("WT") || type.equals("NL")) {
return "IBM037";
}
if (type.equals("DK") || type.equals("NO")) { // Denmark, Norway
return "IBM277";
}
if (type.equals("FI") || type.equals("SE")) { // Finland, Sweden
return "IBM278";
}
if (type.equals("ROECE") || type.equals("YU")) {
return "IBM870";
}
if (type.equals("IT")) return "IBM280";
if (type.equals("ES")) return "IBM284";
if (type.equals("GB")) return "IBM285";
if (type.equals("FR")) return "IBM297";
if (type.equals("AR1")) return "IBM420";
if (type.equals("AR2")) return "IBM918";
if (type.equals("HE")) return "IBM424";
if (type.equals("CH")) return "IBM500";
if (type.equals("IS")) return "IBM871";
// Dunno... let's just default to 037?
return CS_EBCDIC_SUBSET;
}
break;
case 'i':
case 'I':
if (StringUtil.equalEncodings(csName, "ISO-8859-1")
|| StringUtil.equalEncodings(csName, "ISO-Latin1")) {
return CS_ISO_LATIN1;
}
if (StringUtil.encodingStartsWith(csName, "ISO-10646")) {
/* Hmmh. There are boatloads of alternatives here, it
* seems (see http://www.iana.org/assignments/character-sets
* for details)
*/
int ix = csName.indexOf("10646");
String suffix = csName.substring(ix+5);
if (StringUtil.equalEncodings(suffix, "UCS-Basic")) {
return CS_US_ASCII;
}
if (StringUtil.equalEncodings(suffix, "Unicode-Latin1")) {
return CS_ISO_LATIN1;
}
if (StringUtil.equalEncodings(suffix, "UCS-2")) {
return CS_UTF16; // endianness?
}
if (StringUtil.equalEncodings(suffix, "UCS-4")) {
return CS_UTF32; // endianness?
}
if (StringUtil.equalEncodings(suffix, "UTF-1")) {
// "Universal Transfer Format (1), this is the multibyte encoding, that subsets ASCII-7"???
return CS_US_ASCII;
}
if (StringUtil.equalEncodings(suffix, "J-1")) {
// Name: ISO-10646-J-1, Source: ISO 10646 Japanese, see RFC 1815.
// ... so what does that really mean? let's consider it ascii
return CS_US_ASCII;
}
if (StringUtil.equalEncodings(suffix, "US-ASCII")) {
return CS_US_ASCII;
}
} else if (StringUtil.encodingStartsWith(csName, "IBM")) {
// EBCDIC of some kind... what (if anything) to do?
// ... for now, return as is
return csName;
}
break;
case 'j':
case 'J':
if (StringUtil.equalEncodings(csName, "JIS_Encoding")) {
return CS_SHIFT_JIS;
}
break;
case 's':
case 'S':
if (StringUtil.equalEncodings(csName, "Shift_JIS")) {
return CS_SHIFT_JIS;
}
break;
case 'u':
case 'U':
if (csName.length() < 2) { // sanity check
break;
}
switch (csName.charAt(1)) {
case 'c':
case 'C':
if (StringUtil.equalEncodings(csName, "UCS-2")) {
return CS_UTF16;
}
if (StringUtil.equalEncodings(csName, "UCS-4")) {
return CS_UTF32;
}
break;
case 'n': // csUnicodeXxx,
case 'N':
if (gotCsPrefix) {
if (StringUtil.equalEncodings(csName, "Unicode")) {
return CS_UTF16; // need BOM
}
if (StringUtil.equalEncodings(csName, "UnicodeAscii")) {
return CS_ISO_LATIN1;
}
if (StringUtil.equalEncodings(csName, "UnicodeAscii")) {
return CS_US_ASCII;
}
}
break;
case 's':
case 'S':
if (StringUtil.equalEncodings(csName, "US-ASCII")) {
return CS_US_ASCII;
}
break;
case 't':
case 'T':
if (StringUtil.equalEncodings(csName, "UTF-8")) {
return CS_UTF8;
}
if (StringUtil.equalEncodings(csName, "UTF-16BE")) {
return CS_UTF16BE;
}
if (StringUtil.equalEncodings(csName, "UTF-16LE")) {
return CS_UTF16LE;
}
if (StringUtil.equalEncodings(csName, "UTF-16")) {
return CS_UTF16;
}
if (StringUtil.equalEncodings(csName, "UTF-32BE")) {
return CS_UTF32BE;
}
if (StringUtil.equalEncodings(csName, "UTF-32LE")) {
return CS_UTF32LE;
}
if (StringUtil.equalEncodings(csName, "UTF-32")) {
return CS_UTF32;
}
if (StringUtil.equalEncodings(csName, "UTF")) {
// 21-Jan-2006, TSa: ??? What is this to do... ?
return CS_UTF16;
}
}
break;
}
return csName;
}
/**
* Because of legacy encodings used by earlier JDK versions, we
* need to be careful when accessing encoding names via JDK
* classes.
*/
public static String findEncodingFor(Writer w)
{
if (w instanceof OutputStreamWriter) {
String enc = ((OutputStreamWriter) w).getEncoding();
/* [WSTX-146]: It is important that we normalize this, since
* older JDKs return legacy encoding names ("UTF8" instead of
* canonical "UTF-8")
*/
return normalize(enc);
}
return null;
}
}