com.ibm.icu.charset.CharsetCallback Maven / Gradle / Ivy
// © 2016 and later: Unicode, Inc. and others.
// License & terms of use: http://www.unicode.org/copyright.html#License
/**
*******************************************************************************
* Copyright (C) 2006-2014, International Business Machines Corporation and
* others. All Rights Reserved.
*******************************************************************************
*/
package com.ibm.icu.charset;
import java.nio.ByteBuffer;
import java.nio.CharBuffer;
import java.nio.IntBuffer;
import java.nio.charset.CoderResult;
/**
* Callback API for CharsetICU API
*
* CharsetCallback class defines some error behaviour functions called
* by CharsetDecoderICU and CharsetEncoderICU. The class also provides
* the facility by which clients can write their own callbacks.
*
* These functions, although public, should NEVER be called directly.
* They should be used as parameters to the onUmappableCharacter() and
* onMalformedInput() methods, to set the behaviour of a converter
* when it encounters UNMAPPED/INVALID sequences.
* Currently the only way to set callbacks is by using CodingErrorAction.
* In the future we will provide set methods on CharsetEncoder and CharsetDecoder
* that will accept CharsetCallback fields.
*
* @stable ICU 3.6
*/
public class CharsetCallback {
/*
* FROM_U, TO_U context options for sub callback
*/
private static final String SUB_STOP_ON_ILLEGAL = "i";
// /*
// * FROM_U, TO_U context options for skip callback
// */
// private static final String SKIP_STOP_ON_ILLEGAL = "i";
// /*
// * FROM_U_CALLBACK_ESCAPE context option to escape the code unit according to ICU (%UXXXX)
// */
// private static final String ESCAPE_ICU = null;
/*
* FROM_U_CALLBACK_ESCAPE context option to escape the code unit according to JAVA (\\uXXXX)
*/
private static final String ESCAPE_JAVA = "J";
/*
* FROM_U_CALLBACK_ESCAPE context option to escape the code unit according to C (\\uXXXX \\UXXXXXXXX)
* TO_U_CALLBACK_ESCAPE option to escape the character value accoding to C (\\xXXXX)
*/
private static final String ESCAPE_C = "C";
/*
* FROM_U_CALLBACK_ESCAPE context option to escape the code unit according to XML Decimal escape \htmlonly(&#DDDD;)\endhtmlonly
* TO_U_CALLBACK_ESCAPE context option to escape the character value accoding to XML Decimal escape \htmlonly(&#DDDD;)\endhtmlonly
*/
private static final String ESCAPE_XML_DEC = "D";
/*
* FROM_U_CALLBACK_ESCAPE context option to escape the code unit according to XML Hex escape \htmlonly(&#xXXXX;)\endhtmlonly
* TO_U_CALLBACK_ESCAPE context option to escape the character value according to XML Hex escape \htmlonly(&#xXXXX;)\endhtmlonly
*/
private static final String ESCAPE_XML_HEX = "X";
/*
* FROM_U_CALLBACK_ESCAPE context option to escape the code unit according to Unicode (U+XXXXX)
*/
private static final String ESCAPE_UNICODE = "U";
/*
* FROM_U_CALLBACK_ESCAPE context option to escape the code unit according to Unicode (U+XXXXX)
*/
private static final String ESCAPE_CSS2 = "S";
/*
* IS_DEFAULT_IGNORABLE_CODE_POINT
* This is to check if a code point has the default ignorable unicode property.
* As such, this list needs to be updated if the ignorable code point list ever
* changes.
* To avoid dependency on other code, this list is hard coded here.
* When an ignorable code point is found and is unmappable, the default callbacks
* will ignore them.
* For a list of the default ignorable code points, use this link: http://unicode.org/cldr/utility/list-unicodeset.jsp?a=[%3ADI%3A]&g=
*
* This list should be sync with the one in ucnv_err.c
*
*/
private static boolean IS_DEFAULT_IGNORABLE_CODE_POINT(int c) {
return ((c == 0x00AD) ||
(c == 0x034F) ||
(c == 0x061C) ||
(c == 0x115F) ||
(c == 0x1160) ||
(0x17B4 <= c && c <= 0x17B5) ||
(0x180B <= c && c <= 0x180E) ||
(0x200B <= c && c <= 0x200F) ||
(0x202A <= c && c <= 0x202E) ||
(c == 0x2060) ||
(0x2066 <= c && c <= 0x2069) ||
(0x2061 <= c && c <= 0x2064) ||
(0x206A <= c && c <= 0x206F) ||
(c == 0x3164) ||
(0x0FE00 <= c && c <= 0x0FE0F) ||
(c == 0x0FEFF) ||
(c == 0x0FFA0) ||
(0x01BCA0 <= c && c <= 0x01BCA3) ||
(0x01D173 <= c && c <= 0x01D17A) ||
(c == 0x0E0001) ||
(0x0E0020 <= c && c <= 0x0E007F) ||
(0x0E0100 <= c && c <= 0x0E01EF) ||
(c == 0x2065) ||
(0x0FFF0 <= c && c <= 0x0FFF8) ||
(c == 0x0E0000) ||
(0x0E0002 <= c && c <= 0x0E001F) ||
(0x0E0080 <= c && c <= 0x0E00FF) ||
(0x0E01F0 <= c && c <= 0x0E0FFF)
);
}
/**
* Decoder Callback interface
* @stable ICU 3.6
*/
public interface Decoder {
/**
* This function is called when the bytes in the source cannot be handled,
* and this function is meant to handle or fix the error if possible.
*
* @return Result of decoding action. This returned object is set to an error
* if this function could not handle the conversion.
* @stable ICU 3.6
*/
public CoderResult call(CharsetDecoderICU decoder, Object context,
ByteBuffer source, CharBuffer target, IntBuffer offsets,
char[] buffer, int length, CoderResult cr);
}
/**
* Encoder Callback interface
* @stable ICU 3.6
*/
public interface Encoder {
/**
* This function is called when the Unicode characters in the source cannot be handled,
* and this function is meant to handle or fix the error if possible.
* @return Result of decoding action. This returned object is set to an error
* if this function could not handle the conversion.
* @stable ICU 3.6
*/
public CoderResult call(CharsetEncoderICU encoder, Object context,
CharBuffer source, ByteBuffer target, IntBuffer offsets,
char[] buffer, int length, int cp, CoderResult cr);
}
/**
* Skip callback
* @stable ICU 3.6
*/
public static final Encoder FROM_U_CALLBACK_SKIP = new Encoder() {
@Override
public CoderResult call(CharsetEncoderICU encoder, Object context,
CharBuffer source, ByteBuffer target, IntBuffer offsets,
char[] buffer, int length, int cp, CoderResult cr){
if(context==null){
return CoderResult.UNDERFLOW;
}else if(((String)context).equals(SUB_STOP_ON_ILLEGAL)){
if(!cr.isUnmappable()){
return cr;
}else{
return CoderResult.UNDERFLOW;
}
}
return cr;
}
};
/**
* Skip callback
* @stable ICU 3.6
*/
public static final Decoder TO_U_CALLBACK_SKIP = new Decoder() {
@Override
public CoderResult call(CharsetDecoderICU decoder, Object context,
ByteBuffer source, CharBuffer target, IntBuffer offsets,
char[] buffer, int length, CoderResult cr){
if(context==null){
return CoderResult.UNDERFLOW;
}else if(((String)context).equals(SUB_STOP_ON_ILLEGAL)){
if(!cr.isUnmappable()){
return cr;
}else{
return CoderResult.UNDERFLOW;
}
}
return cr;
}
};
/**
* Write substitute callback
* @stable ICU 3.6
*/
public static final Encoder FROM_U_CALLBACK_SUBSTITUTE = new Encoder(){
@Override
public CoderResult call(CharsetEncoderICU encoder, Object context,
CharBuffer source, ByteBuffer target, IntBuffer offsets,
char[] buffer, int length, int cp, CoderResult cr){
if (cr.isUnmappable() && IS_DEFAULT_IGNORABLE_CODE_POINT(cp)) {
return CoderResult.UNDERFLOW;
}else if(context==null){
return encoder.cbFromUWriteSub(encoder, source, target, offsets);
}else if(((String)context).equals(SUB_STOP_ON_ILLEGAL)){
if(!cr.isUnmappable()){
return cr;
}else{
return encoder.cbFromUWriteSub(encoder, source, target, offsets);
}
}
return cr;
}
};
private static final char[] kSubstituteChar1 = new char[]{0x1A};
private static final char[] kSubstituteChar = new char[] {0xFFFD};
/**
* Write substitute callback
* @stable ICU 3.6
*/
public static final Decoder TO_U_CALLBACK_SUBSTITUTE = new Decoder() {
@Override
public CoderResult call(CharsetDecoderICU decoder, Object context,
ByteBuffer source, CharBuffer target, IntBuffer offsets,
char[] buffer, int length, CoderResult cr){
CharsetICU cs = (CharsetICU) decoder.charset();
/* Use the specified replacement character if it is different than the default one. */
boolean useReplacement = true;
char [] replacementChar = decoder.replacement().toCharArray();
if (replacementChar.length == 1 && (replacementChar[0] == kSubstituteChar1[0] || replacementChar[0] == kSubstituteChar[0])) {
useReplacement = false;
}
/* could optimize this case, just one uchar */
if(decoder.invalidCharLength == 1 && cs.subChar1 != 0) {
return CharsetDecoderICU.toUWriteUChars(decoder, useReplacement ? replacementChar : kSubstituteChar1, 0, useReplacement ? replacementChar.length : 1, target, offsets, source.position());
} else {
return CharsetDecoderICU.toUWriteUChars(decoder, useReplacement ? replacementChar : kSubstituteChar, 0, useReplacement ? replacementChar.length : 1, target, offsets, source.position());
}
}
};
/**
* Stop callback
* @stable ICU 3.6
*/
public static final Encoder FROM_U_CALLBACK_STOP = new Encoder() {
@Override
public CoderResult call(CharsetEncoderICU encoder, Object context,
CharBuffer source, ByteBuffer target, IntBuffer offsets,
char[] buffer, int length, int cp, CoderResult cr){
if (cr.isUnmappable() && IS_DEFAULT_IGNORABLE_CODE_POINT(cp)) {
return CoderResult.UNDERFLOW;
}
return cr;
}
};
/**
* Stop callback
* @stable ICU 3.6
*/
public static final Decoder TO_U_CALLBACK_STOP = new Decoder() {
@Override
public CoderResult call(CharsetDecoderICU decoder, Object context,
ByteBuffer source, CharBuffer target, IntBuffer offsets,
char[] buffer, int length, CoderResult cr){
return cr;
}
};
private static final int VALUE_STRING_LENGTH = 32;
private static final char UNICODE_PERCENT_SIGN_CODEPOINT = 0x0025;
private static final char UNICODE_U_CODEPOINT = 0x0055;
private static final char UNICODE_X_CODEPOINT = 0x0058;
private static final char UNICODE_RS_CODEPOINT = 0x005C;
private static final char UNICODE_U_LOW_CODEPOINT = 0x0075;
private static final char UNICODE_X_LOW_CODEPOINT = 0x0078;
private static final char UNICODE_AMP_CODEPOINT = 0x0026;
private static final char UNICODE_HASH_CODEPOINT = 0x0023;
private static final char UNICODE_SEMICOLON_CODEPOINT = 0x003B;
private static final char UNICODE_PLUS_CODEPOINT = 0x002B;
private static final char UNICODE_LEFT_CURLY_CODEPOINT = 0x007B;
private static final char UNICODE_RIGHT_CURLY_CODEPOINT = 0x007D;
private static final char UNICODE_SPACE_CODEPOINT = 0x0020;
/**
* Write escape callback
* @stable ICU 4.0
*/
public static final Encoder FROM_U_CALLBACK_ESCAPE = new Encoder() {
@Override
public CoderResult call(CharsetEncoderICU encoder, Object context,
CharBuffer source, ByteBuffer target, IntBuffer offsets,
char[] buffer, int length, int cp, CoderResult cr){
char[] valueString = new char[VALUE_STRING_LENGTH];
int valueStringLength = 0;
int i = 0;
if (cr.isUnmappable() && IS_DEFAULT_IGNORABLE_CODE_POINT(cp)) {
return CoderResult.UNDERFLOW;
}
if (context == null || !(context instanceof String)) {
while (i < length) {
valueString[valueStringLength++] = UNICODE_PERCENT_SIGN_CODEPOINT; /* adding % */
valueString[valueStringLength++] = UNICODE_U_CODEPOINT; /* adding U */
valueStringLength += itou(valueString, valueStringLength, buffer[i++], 16, 4);
}
} else {
if (((String)context).equals(ESCAPE_JAVA)) {
while (i < length) {
valueString[valueStringLength++] = UNICODE_RS_CODEPOINT; /* adding \ */
valueString[valueStringLength++] = UNICODE_U_LOW_CODEPOINT; /* adding u */
valueStringLength += itou(valueString, valueStringLength, buffer[i++], 16, 4);
}
} else if (((String)context).equals(ESCAPE_C)) {
valueString[valueStringLength++] = UNICODE_RS_CODEPOINT; /* adding \ */
if (length == 2) {
valueString[valueStringLength++] = UNICODE_U_CODEPOINT; /* adding U */
valueStringLength = itou(valueString, valueStringLength, cp, 16, 8);
} else {
valueString[valueStringLength++] = UNICODE_U_LOW_CODEPOINT; /* adding u */
valueStringLength += itou(valueString, valueStringLength, buffer[0], 16, 4);
}
} else if (((String)context).equals(ESCAPE_XML_DEC)) {
valueString[valueStringLength++] = UNICODE_AMP_CODEPOINT; /* adding & */
valueString[valueStringLength++] = UNICODE_HASH_CODEPOINT; /* adding # */
if (length == 2) {
valueStringLength += itou(valueString, valueStringLength, cp, 10, 0);
} else {
valueStringLength += itou(valueString, valueStringLength, buffer[0], 10, 0);
}
valueString[valueStringLength++] = UNICODE_SEMICOLON_CODEPOINT; /* adding ; */
} else if (((String)context).equals(ESCAPE_XML_HEX)) {
valueString[valueStringLength++] = UNICODE_AMP_CODEPOINT; /* adding & */
valueString[valueStringLength++] = UNICODE_HASH_CODEPOINT; /* adding # */
valueString[valueStringLength++] = UNICODE_X_LOW_CODEPOINT; /* adding x */
if (length == 2) {
valueStringLength += itou(valueString, valueStringLength, cp, 16, 0);
} else {
valueStringLength += itou(valueString, valueStringLength, buffer[0], 16, 0);
}
valueString[valueStringLength++] = UNICODE_SEMICOLON_CODEPOINT; /* adding ; */
} else if (((String)context).equals(ESCAPE_UNICODE)) {
valueString[valueStringLength++] = UNICODE_LEFT_CURLY_CODEPOINT; /* adding { */
valueString[valueStringLength++] = UNICODE_U_CODEPOINT; /* adding U */
valueString[valueStringLength++] = UNICODE_PLUS_CODEPOINT; /* adding + */
if (length == 2) {
valueStringLength += itou(valueString, valueStringLength,cp, 16, 4);
} else {
valueStringLength += itou(valueString, valueStringLength, buffer[0], 16, 4);
}
valueString[valueStringLength++] = UNICODE_RIGHT_CURLY_CODEPOINT; /* adding } */
} else if (((String)context).equals(ESCAPE_CSS2)) {
valueString[valueStringLength++] = UNICODE_RS_CODEPOINT; /* adding \ */
valueStringLength += itou(valueString, valueStringLength, cp, 16, 0);
/* Always add space character, because the next character might be whitespace,
which would erroneously be considered the termination of the escape sequence. */
valueString[valueStringLength++] = UNICODE_SPACE_CODEPOINT;
} else {
while (i < length) {
valueString[valueStringLength++] = UNICODE_PERCENT_SIGN_CODEPOINT; /* adding % */
valueString[valueStringLength++] = UNICODE_U_CODEPOINT; /* adding U */
valueStringLength += itou(valueString, valueStringLength, buffer[i++], 16, 4);
}
}
}
return encoder.cbFromUWriteUChars(encoder, CharBuffer.wrap(valueString, 0, valueStringLength), target, offsets);
}
};
/**
* Write escape callback
* @stable ICU 4.0
*/
public static final Decoder TO_U_CALLBACK_ESCAPE = new Decoder() {
@Override
public CoderResult call(CharsetDecoderICU decoder, Object context,
ByteBuffer source, CharBuffer target, IntBuffer offsets,
char[] buffer, int length, CoderResult cr){
char[] uniValueString = new char[VALUE_STRING_LENGTH];
int valueStringLength = 0;
int i = 0;
if (context == null || !(context instanceof String)) {
while (i < length) {
uniValueString[valueStringLength++] = UNICODE_PERCENT_SIGN_CODEPOINT; /* adding % */
uniValueString[valueStringLength++] = UNICODE_X_CODEPOINT; /* adding U */
valueStringLength += itou(uniValueString, valueStringLength, buffer[i++] & UConverterConstants.UNSIGNED_BYTE_MASK, 16, 2);
}
} else {
if (((String)context).equals(ESCAPE_XML_DEC)) {
while (i < length) {
uniValueString[valueStringLength++] = UNICODE_AMP_CODEPOINT; /* adding & */
uniValueString[valueStringLength++] = UNICODE_HASH_CODEPOINT; /* adding # */
valueStringLength += itou(uniValueString, valueStringLength, buffer[i++] & UConverterConstants.UNSIGNED_BYTE_MASK, 10, 0);
uniValueString[valueStringLength++] = UNICODE_SEMICOLON_CODEPOINT; /* adding ; */
}
} else if (((String)context).equals(ESCAPE_XML_HEX)) {
while (i < length) {
uniValueString[valueStringLength++] = UNICODE_AMP_CODEPOINT; /* adding & */
uniValueString[valueStringLength++] = UNICODE_HASH_CODEPOINT; /* adding # */
uniValueString[valueStringLength++] = UNICODE_X_LOW_CODEPOINT; /* adding x */
valueStringLength += itou(uniValueString, valueStringLength, buffer[i++] & UConverterConstants.UNSIGNED_BYTE_MASK, 16, 0);
uniValueString[valueStringLength++] = UNICODE_SEMICOLON_CODEPOINT; /* adding ; */
}
} else if (((String)context).equals(ESCAPE_C)) {
while (i < length) {
uniValueString[valueStringLength++] = UNICODE_RS_CODEPOINT; /* adding \ */
uniValueString[valueStringLength++] = UNICODE_X_LOW_CODEPOINT; /* adding x */
valueStringLength += itou(uniValueString, valueStringLength, buffer[i++] & UConverterConstants.UNSIGNED_BYTE_MASK, 16, 2);
}
} else {
while (i < length) {
uniValueString[valueStringLength++] = UNICODE_PERCENT_SIGN_CODEPOINT; /* adding % */
uniValueString[valueStringLength++] = UNICODE_X_CODEPOINT; /* adding X */
itou(uniValueString, valueStringLength, buffer[i++] & UConverterConstants.UNSIGNED_BYTE_MASK, 16, 2);
valueStringLength += 2;
}
}
}
cr = CharsetDecoderICU.toUWriteUChars(decoder, uniValueString, 0, valueStringLength, target, offsets, 0);
return cr;
}
};
/***
* Java port of uprv_itou() in ICU4C used by TO_U_CALLBACK_ESCAPE and FROM_U_CALLBACK_ESCAPE.
* Fills in a char string with the radix-based representation of a number padded with zeroes
* to minwidth.
*/
private static final int itou(char[] buffer, int sourceIndex, int i, int radix, int minwidth) {
int length = 0;
int digit;
int j;
char temp;
do {
digit = i % radix;
buffer[sourceIndex + length++] = (char)(digit <= 9 ? (0x0030+digit) : (0x0030+digit+7));
i = i/radix;
} while (i != 0 && (sourceIndex + length) < buffer.length);
while (length < minwidth) {
buffer[sourceIndex + length++] = (char)0x0030; /* zero padding */
}
/* reverses the string */
for (j = 0; j < (length / 2); j++) {
temp = buffer[(sourceIndex + length - 1) - j];
buffer[(sourceIndex + length-1) -j] = buffer[sourceIndex + j];
buffer[sourceIndex + j] = temp;
}
return length;
}
/*
* No need to create an instance
*/
private CharsetCallback() {
}
}
© 2015 - 2024 Weber Informatics LLC | Privacy Policy