pjs/modules/libnls/headers/unicode.h

649 строки
23 KiB
C++

/*
*****************************************************************************************
* *
* COPYRIGHT: *
* (C) Copyright Taligent, Inc., 1996, 1997 *
* (C) Copyright International Business Machines Corporation, 1996, 1997 *
* Licensed Material - Program-Property of IBM - All Rights Reserved. *
* US Government Users Restricted Rights - Use, duplication, or disclosure *
* restricted by GSA ADP Schedule Contract with IBM Corp. *
* *
*****************************************************************************************
*/
// Copyright (C) 1996-1997 Taligent, Inc. All rights reserved.
//
// FILE NAME : unicode.h
//
// CREATED
// Wednesday, December 11, 1996
//
// CREATED BY
// Helena Shih
//
//
//********************************************************************************************
#ifndef _UNICODE
#define _UNICODE
#include "ptypes.h"
class CompactByteArray;
class CompactShortArray;
#ifdef NLS_MAC
#pragma export on
#endif
/**
* The Unicode class allows you to query the properties associated with individual
* Unicode character values.
* <p>
* The Unicode character information, provided implicitly by the
* Unicode character encoding standard, includes information about the sript
* (for example, symbols or control characters) to which the character belongs,
* as well as semantic information such as whether a character is a digit or
* uppercase, lowercase, or uncased.
* <P>
* @subclassing Do not subclass.
*/
class T_UTILITY_API Unicode
{
public:
/**
* The minimum value a UniChar can have. The lowest value a
* UniChar can have is 0x0000.
*/
static const UniChar MIN_VALUE;
/**
* The maximum value a UniChar can have. The greatest value a
* UniChar can have is 0xffff.
*/
static const UniChar MAX_VALUE;
/**
* Public data for enumerated Unicode general category types
*/
enum EUnicodeGeneralTypes
{
UNASSIGNED = 0,
UPPERCASE_LETTER = 1,
LOWERCASE_LETTER = 2,
TITLECASE_LETTER = 3,
MODIFIER_LETTER = 4,
OTHER_LETTER = 5,
NON_SPACING_MARK = 6,
ENCLOSING_MARK = 7,
COMBINING_SPACING_MARK = 8,
DECIMAL_DIGIT_NUMBER = 9,
LETTER_NUMBER = 10,
OTHER_NUMBER = 11,
SPACE_SEPARATOR = 12,
LINE_SEPARATOR = 13,
PARAGRAPH_SEPARATOR = 14,
CONTROL = 15,
FORMAT = 16,
PRIVATE_USE = 17,
SURROGATE = 18,
DASH_PUNCTUATION = 19,
START_PUNCTUATION = 20,
END_PUNCTUATION = 21,
CONNECTOR_PUNCTUATION = 22,
OTHER_PUNCTUATION = 23,
MATH_SYMBOL = 24,
CURRENCY_SYMBOL = 25,
MODIFIER_SYMBOL = 26,
OTHER_SYMBOL = 27,
GENERAL_TYPES_COUNT = 28
};
/**
* This specifies the language directional property of a character set.
*/
enum EDirectionProperty {
LEFT_TO_RIGHT = 0,
RIGHT_TO_LEFT = 1,
EUROPEAN_NUMBER = 2,
EUROPEAN_NUMBER_SEPARATOR = 3,
EUROPEAN_NUMBER_TERMINATOR = 4,
ARABIC_NUMBER = 5,
COMMON_NUMBER_SEPARATOR = 6,
BLOCK_SEPARATOR = 7,
SEGMENT_SEPARATOR = 8,
WHITE_SPACE_NEUTRAL = 9,
OTHER_NEUTRAL = 10
};
/**
* Values returned by the getCellWidth() function.
* @see Unicode#getCellWidth
*/
enum ECellWidths
{
ZERO_WIDTH = 0,
HALF_WIDTH = 1,
FULL_WIDTH = 2,
NEUTRAL = 3
};
/**
* Determines whether the specified UniChar is a lowercase character
* according to Unicode 2.0.14.
*
* @param ch the character to be tested
* @return true if the character is lowercase; false otherwise.
*
* @see Unicode#isUpperCase
* @see Unicode#isTitleCase
* @see Unicode#toLowerCase
*/
static t_bool isLowerCase( UniChar ch);
/**
* Determines whether the specified character is an uppercase character
* according to Unicode 2.0.14.
*
* @param ch the character to be tested
* @return true if the character is uppercase; false otherwise.
* @see Unicode#isLowerCase
* @see Unicode#isTitleCase
* @see Unicode#toUpperCase
*/
static t_bool isUpperCase( UniChar ch);
/**
* Determines whether the specified character is a titlecase character
* according to Unicode 2.0.14.
*
* @param ch the character to be tested
* @return true if the character is titlecase; false otherwise.
* @see Unicode#isUpperCase
* @see Unicode#isLowerCase
* @see Unicode#toTitleCase
*/
static t_bool isTitleCase( UniChar ch);
/**
* Determines whether the specified character is a digit according to Unicode
* 2.0.14.
*
* @param ch the character to be tested
* @return true if the character is a digit; false otherwise.
*/
static t_bool isDigit( UniChar ch) ;
/**
* Determines whether the specified numeric value is actually a defined character
* according to Unicode 2.0.14.
*
* @param ch the character to be tested
* @return true if the character has a defined Unicode meaning; false otherwise.
*
* @see Unicode#isDigit
* @see Unicode#isLetter
* @see Unicode#isLetterOrDigit
* @see Unicode#isUpperCase
* @see Unicode#isLowerCase
* @see Unicode#isTitleCase
*/
static t_bool isDefined(UniChar ch);
/**
* Determines whether the specified character is a control character according
* to Unicode 2.0.14.
*
* @param ch the character to be tested
* @return true if the Unicode character is a control character; false otherwise.
*
* @see Unicode#isPrintable
*/
static t_bool isControl(UniChar ch);
/**
* Determines whether the specified character is a printable character according
* to Unicode 2.0.14.
*
* @param ch the character to be tested
* @return true if the Unicode character is a printable character; false otherwise.
*
* @see Unicode#isControl
*/
static t_bool isPrintable(UniChar ch);
/**
* Determines whether the specified character is of the base form according
* to Unicode 2.0.14.
*
* @param ch the character to be tested
* @return true if the Unicode character is of the base form; false otherwise.
*
* @see Unicode#isLetter
* @see Unicode#isDigit
*/
static t_bool isBaseForm(UniChar ch);
/**
* Determines whether the specified character is a letter
* according to Unicode 2.0.14.
*
* @param ch the character to be tested
* @return true if the character is a letter; false otherwise.
*
*
* @see Unicode#isDigit
* @see Unicode#isLetterOrDigit
* @see Unicode#isUpperCase
* @see Unicode#isLowerCase
* @see Unicode#isTitleCase
*/
static t_bool isLetter(UniChar ch);
/**
* A convenience method for determining if a Unicode character
* is allowed as the first character in a Java identifier.
* <P>
* A character may start a Java identifier if and only if
* it is one of the following:
* <ul>
* <li> a letter
* <li> a currency symbol (such as "$")
* <li> a connecting punctuation symbol (such as "_").
* </ul>
*
* @param ch the Unicode character.
* @return TRUE if the character may start a Java identifier;
* FALSE otherwise.
* @see isJavaIdentifierPart
* @see isLetter
* @see isUnicodeIdentifierStart
*/
static t_bool isJavaIdentifierStart(UniChar ch);
/**
* A convenience method for determining if a Unicode character
* may be part of a Java identifier other than the starting
* character.
* <P>
* A character may be part of a Java identifier if and only if
* it is one of the following:
* <ul>
* <li> a letter
* <li> a currency symbol (such as "$")
* <li> a connecting punctuation character (such as "_").
* <li> a digit
* <li> a numeric letter (such as a Roman numeral character)
* <li> a combining mark
* <li> a non-spacing mark
* <li> an ignorable control character
* </ul>
*
* @param ch the Unicode character.
* @return TRUE if the character may be part of a Unicode identifier;
* FALSE otherwise.
* @see isIdentifierIgnorable
* @see isJavaIdentifierStart
* @see isLetter
* @see isDigit
* @see isUnicodeIdentifierPart
*/
static t_bool isJavaIdentifierPart(UniChar ch);
/**
* A convenience method for determining if a Unicode character
* is allowed to start in a Unicode identifier.
* A character may start a Unicode identifier if and only if
* it is a letter.
*
* @param ch the Unicode character.
* @return TRUE if the character may start a Unicode identifier;
* FALSE otherwise.
* @see isJavaIdentifierStart
* @see isLetter
* @see isUnicodeIdentifierPart
*/
static t_bool isUnicodeIdentifierStart(UniChar ch);
/**
* A convenience method for determining if a Unicode character
* may be part of a Unicode identifier other than the starting
* character.
* <P>
* A character may be part of a Unicode identifier if and only if
* it is one of the following:
* <ul>
* <li> a letter
* <li> a connecting punctuation character (such as "_").
* <li> a digit
* <li> a numeric letter (such as a Roman numeral character)
* <li> a combining mark
* <li> a non-spacing mark
* <li> an ignorable control character
* </ul>
*
* @param ch the Unicode character.
* @return TRUE if the character may be part of a Unicode identifier;
* FALSE otherwise.
* @see isIdentifierIgnorable
* @see isJavaIdentifierPart
* @see isLetterOrDigit
* @see isUnicodeIdentifierStart
*/
static t_bool isUnicodeIdentifierPart(UniChar ch);
/**
* A convenience method for determining if a Unicode character
* should be regarded as an ignorable character in a Java
* identifier or a Unicode identifier.
* <P>
* The following Unicode characters are ignorable in a Java identifier
* or a Unicode identifier:
* <table>
* <tr><td>0x0000 through 0x0008,</td>
* <td>ISO control characters that</td></tr>
* <tr><td>0x000E through 0x001B,</td> <td>are not whitespace</td></tr>
* <tr><td>and 0x007F through 0x009F</td></tr>
* <tr><td>0x200C through 0x200F</td> <td>join controls</td></tr>
* <tr><td>0x200A through 0x200E</td> <td>bidirectional controls</td></tr>
* <tr><td>0x206A through 0x206F</td> <td>format controls</td></tr>
* <tr><td>0xFEFF</td> <td>zero-width no-break space</td></tr>
* </table>
*
* @param ch the Unicode character.
* @return TRUE if the character may be part of a Unicode identifier;
* FALSE otherwise.
* @see isJavaIdentifierPart
* @see isUnicodeIdentifierPart
*/
static t_bool isIdentifierIgnorable(UniChar ch);
/**
* The given character is mapped to its lowercase equivalent according to
* Unicode 2.0.14; if the character has no lowercase equivalent, the character
* itself is returned.
* <P>
* A character has a lowercase equivalent if and only if a lowercase mapping
* is specified for the character in the Unicode 2.0 attribute table.
* <P>
* Unicode::toLowerCase() only deals with the general letter case conversion.
* For language specific case conversion behavior, use UnicodeString::toLower().
* For example, the case conversion for dot-less i and dotted I in Turkish,
* or for final sigma in Greek.
*
* @param ch the character to be converted
* @return the lowercase equivalent of the character, if any;
* otherwise the character itself.
*
* @see UnicodeString#toLower
* @see Unicode#isLowerCase
* @see Unicode#isUpperCase
* @see Unicode#toUpperCase
* @see Unicode#toTitleCase
*/
static UniChar toLowerCase(UniChar ch);
/**
* The given character is mapped to its uppercase equivalent according to Unicode
* 2.0.14; if the character has no uppercase equivalent, the character itself is
* returned.
* <P>
* Unicode::toUpperCase() only deals with the general letter case conversion.
* For language specific case conversion behavior, use UnicodeString::toUpper().
* For example, the case conversion for dot-less i and dotted I in Turkish,
* or ess-zed (i.e., "sharp S") in German.
*
* @param ch the character to be converted
* @return the uppercase equivalent of the character, if any;
* otherwise the character itself.
*
* @see UnicodeString#toUpper
* @see Unicode#isUpperCase
* @see Unicode#isLowerCase
* @see Unicode#toLowerCase
* @see Unicode#toTitleCase
*/
static UniChar toUpperCase(UniChar ch);
/**
* The given character is mapped to its titlecase equivalent according to Unicode
* 2.0.14. There are only four Unicode characters that are truly titlecase forms
* that are distinct from uppercase forms. As a rule, if a character has no
* true titlecase equivalent, its uppercase equivalent is returned.
* <P>
* A character has a titlecase equivalent if and only if a titlecase mapping
* is specified for the character in the Unicode 2.0.14 data.
*
* @param ch the character to be converted
* @return the titlecase equivalent of the character, if any;
* otherwise the character itself.
* @see Unicode#isTitleCase
* @see Unicode#toUpperCase
* @see Unicode#toLowerCase
*/
static UniChar toTitleCase(UniChar ch);
/**
* Determines if the specified character is a Unicode space character
* according to Unicode 2.0.14.
*
* @param ch the character to be tested
* @return true if the character is a space character; false otherwise.
*/
static t_bool isSpaceChar(UniChar ch);
/**
* Returns a value indicating a character category according to Unicode
* 2.0.14.
* @param ch the character to be tested
* @return a value of type int, the character category.
* @see Unicode#UNASSIGNED
* @see Unicode#UPPERCASE_LETTER
* @see Unicode#LOWERCASE_LETTER
* @see Unicode#TITLECASE_LETTER
* @see Unicode#MODIFIER_LETTER
* @see Unicode#OTHER_LETTER
* @see Unicode#NON_SPACING_MARK
* @see Unicode#ENCLOSING_MARK
* @see Unicode#COMBINING_SPACING_MARK
* @see Unicode#DECIMAL_DIGIT_NUMBER
* @see Unicode#OTHER_NUMBER
* @see Unicode#SPACE_SEPARATOR
* @see Unicode#LINE_SEPARATOR
* @see Unicode#PARAGRAPH_SEPARATOR
* @see Unicode#CONTROL
* @see Unicode#PRIVATE_USE
* @see Unicode#SURROGATE
* @see Unicode#DASH_PUNCTUATION
* @see Unicode#OPEN_PUNCTUATION
* @see Unicode#CLOSE_PUNCTUATION
* @see Unicode#CONNECTOR_PUNCTUATION
* @see Unicode#OTHER_PUNCTUATION
* @see Unicode#LETTER_NUMBER
* @see Unicode#MATH_SYMBOL
* @see Unicode#CURRENCY_SYMBOL
* @see Unicode#MODIFIER_SYMBOL
* @see Unicode#OTHER_SYMBOL
*/
static t_int8 getType(UniChar ch);
/**
* Returns the linguistic direction property of a character.
* <P>
* Returns the linguistic direction property of a character.
* For example, 0x0041 (letter A) has the LEFT_TO_RIGHT directional
* property.
* @see #EDirectionProperty
*/
static EDirectionProperty characterDirection(UniChar ch);
/**
* Returns a value indicating the display-cell width of the character
* when used in Asian text, according to the Unicode standard (see p. 6-130
* of The Unicode Standard, Version 2.0). The results for various characters
* are as follows:
* <P>
* ZERO_WIDTH: Characters which are considered to take up no display-cell space:
* control characters
* format characters
* line and paragraph separators
* non-spacing marks
* combining Hangul jungseong
* combining Hangul jongseong
* unassigned Unicode values
* <P>
* HALF_WIDTH: Characters which take up half a cell in standard Asian text:
* all characters in the General Scripts Area except combining Hangul choseong
* and the characters called out specifically above as ZERO_WIDTH
* alphabetic and Arabic presentation forms
* halfwidth CJK punctuation
* halfwidth Katakana
* halfwidth Hangul Jamo
* halfwidth forms, arrows, and shapes
* <P>
* FULL_WIDTH: Characters which take up a full cell in standard Asian text:
* combining Hangul choseong
* all characters in the CJK Phonetics and Symbols Area
* all characters in the CJK Ideographs Area
* all characters in the Hangul Syllables Area
* CJK compatibility ideographs
* CJK compatibility forms
* small form variants
* fullwidth ASCII
* fullwidth punctuation and currency signs
* <P>
* NEUTRAL: Characters whose cell width is context-dependent:
* all characters in the Symbols Area, except those specifically called out above
* all characters in the Surrogates Area
* all charcaters in the Private Use Area
* <P>
* For Korean text, this algorithm should work properly with properly normalized Korean
* text. Precomposed Hangul syllables and non-combining jamo are all considered full-
* width characters. For combining jamo, we treat we treat choseong (initial consonants)
* as double-width characters and junseong (vowels) and jongseong (final consonants)
* as non-spacing marks. This will work right in text that uses the precomposed
* choseong characters instead of teo choseong characters in a row, and which uses the
* choseong filler character at the beginning of syllables that don't have an initial
* consonant. The results may be slightly off with Korean text following different
* conventions.
*/
static t_uint16 getCellWidth(UniChar ch);
static UniChar* toLowerCase(UniChar* ch, t_int32 size);
static UniChar* toUpperCase(UniChar* ch, t_int32 size);
static int compare(const UniChar* ch, t_int32 size, const UniChar* ch1, t_int32 size1);
static int compareIgnoreCase(const UniChar* ch, t_int32 size, const UniChar* ch1, t_int32 size1);
static TextOffset indexOf(const UniChar* ch, t_int32 size, const UniChar* character, t_int32 length, TextOffset fromOffset = 0);
static TextOffset lastIndexOf(const UniChar* ch, t_int32 size, const UniChar* character, t_int32 length, TextOffset fromOffset = 0);
static size_t length(const UniChar* str);
static UniChar* copy(const UniChar* str);
//
// Netscape Added Methods
//
/**
* Releases a UniChar array allocated by one of the Unicode methods.
*/
static void release(UniChar* buffer);
/**
* Returns a pointer to the first instance of ch in the
* Unicode string.
* @param str string to search
* @param strLength length of str in UniChars
* @param ch UniChar to search for
*/
static const UniChar* strchr(const UniChar* str, size_t strLength, UniChar ch);
/**
* Returns a pointer to the last instance of ch in the
* Unicode string.
* @param str string to search
* @param strLength length of str in UniChars
* @param ch UniChar to search for
*/
static const UniChar* strrchr(const UniChar* str, size_t strLength, UniChar ch);
/**
* Compares two UniChar arrays based on the underlying file system
* mode of comparison.
* @param source array to be compared
* @param sourceLength length of source in UniChars
* @param target array to be compared
* @param targetLength length of target in UniChars
*/
static int fileNameCompare(const UniChar* source, size_t sourceLength, const UniChar* target, size_t targetLength);
/**
* Concatenates two UniChar arrays and returns the result in a new buffer
* @param str1 first string
* @param sourceLength length of str1 in UniChars
* @param str2 second string
* @param str2Length length of second string
*/
static UniChar* appendString(const UniChar* str1,size_t str1Length, const UniChar* str2,size_t str2Length);
/**
* Regular Expression-like search through a UniChar string.
* @param string string to search
* @param stringLength length of string in UniChars
* @param expression regular expression (as a C-string)
*/
static TextOffset regularExpressionSearch(const UniChar* string, size_t stringLength, const char* expression);
/**
* Regular Expression-like search through a UniChar string.
* @param source string to search
* @param sourceLength length of source in UniChars
* @param expression regular expression (as a UniChar)
* @param expressionLength length of regular expression in UniChars
*/
static TextOffset regularExpressionSearch(const UniChar* string, size_t stringLength, const UniChar* expression, size_t expressionLength);
protected:
// These constructors, destructor, and assignment operator must
// be protected (not private, as they semantically are) to make
// various UNIX compilers happy. [LIU]
Unicode();
Unicode( const Unicode& other);
~Unicode();
const Unicode& operator=( const Unicode& other);
private:
static const t_int8 isLetterMask;
static const t_uint16 indicies[];
static const t_int8 values[];
static const t_int32 offsetCount;
static const t_uint16 caseIndex[];
static const t_int16 caseValues[];
static const t_int32 caseCount;
static const t_uint16 fCharDirIndices[];
static const t_int8 fCharDirValues[];
static const t_int32 fCharDirCount;
static CompactByteArray *tables;
static CompactShortArray *ulTables;
static CompactByteArray *dirTables;
static t_bool tablesCreated;
static t_bool ulTablesCreated;
static t_bool dirTablesCreated;
static void createTables();
static void createUlTables();
static void createDirTables();
};
#ifdef NLS_MAC
#pragma export off
#endif
#endif