2020-08-11 09:10:23 +00:00
|
|
|
// © 2016 and later: Unicode, Inc. and others.
|
|
|
|
// License & terms of use: http://www.unicode.org/copyright.html
|
|
|
|
/*
|
|
|
|
*******************************************************************************
|
|
|
|
*
|
|
|
|
* Copyright (C) 2004-2012, International Business Machines
|
|
|
|
* Corporation and others. All Rights Reserved.
|
|
|
|
*
|
|
|
|
*******************************************************************************
|
|
|
|
* file name: ucase.h
|
|
|
|
* encoding: UTF-8
|
|
|
|
* tab size: 8 (not used)
|
|
|
|
* indentation:4
|
|
|
|
*
|
|
|
|
* created on: 2004aug30
|
|
|
|
* created by: Markus W. Scherer
|
|
|
|
*
|
|
|
|
* Low-level Unicode character/string case mapping code.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#ifndef __UCASE_H__
|
|
|
|
#define __UCASE_H__
|
|
|
|
|
|
|
|
#include "unicode/utypes.h"
|
|
|
|
#include "unicode/uset.h"
|
|
|
|
#include "putilimp.h"
|
|
|
|
#include "uset_imp.h"
|
|
|
|
#include "udataswp.h"
|
|
|
|
#include "utrie2.h"
|
|
|
|
|
|
|
|
#ifdef __cplusplus
|
|
|
|
U_NAMESPACE_BEGIN
|
|
|
|
|
|
|
|
class UnicodeString;
|
|
|
|
|
|
|
|
U_NAMESPACE_END
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/* library API -------------------------------------------------------------- */
|
|
|
|
|
|
|
|
U_CFUNC void U_EXPORT2
|
|
|
|
ucase_addPropertyStarts(const USetAdder *sa, UErrorCode *pErrorCode);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Requires non-NULL locale ID but otherwise does the equivalent of
|
|
|
|
* checking for language codes as if uloc_getLanguage() were called:
|
|
|
|
* Accepts both 2- and 3-letter codes and accepts case variants.
|
|
|
|
*/
|
|
|
|
U_CFUNC int32_t
|
|
|
|
ucase_getCaseLocale(const char *locale);
|
|
|
|
|
|
|
|
/* Casing locale types for ucase_getCaseLocale */
|
|
|
|
enum {
|
|
|
|
UCASE_LOC_UNKNOWN,
|
|
|
|
UCASE_LOC_ROOT,
|
|
|
|
UCASE_LOC_TURKISH,
|
|
|
|
UCASE_LOC_LITHUANIAN,
|
|
|
|
UCASE_LOC_GREEK,
|
|
|
|
UCASE_LOC_DUTCH,
|
|
|
|
UCASE_LOC_ARMENIAN
|
|
|
|
};
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Bit mask for getting just the options from a string compare options word
|
|
|
|
* that are relevant for case-insensitive string comparison.
|
|
|
|
* See stringoptions.h. Also include _STRNCMP_STYLE and U_COMPARE_CODE_POINT_ORDER.
|
|
|
|
* @internal
|
|
|
|
*/
|
|
|
|
#define _STRCASECMP_OPTIONS_MASK 0xffff
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Bit mask for getting just the options from a string compare options word
|
|
|
|
* that are relevant for case folding (of a single string or code point).
|
|
|
|
*
|
|
|
|
* Currently only bit 0 for U_FOLD_CASE_EXCLUDE_SPECIAL_I.
|
|
|
|
* It is conceivable that at some point we might use one more bit for using uppercase sharp s.
|
|
|
|
* It is conceivable that at some point we might want the option to use only simple case foldings
|
|
|
|
* when operating on strings.
|
|
|
|
*
|
|
|
|
* See stringoptions.h.
|
|
|
|
* @internal
|
|
|
|
*/
|
|
|
|
#define _FOLD_CASE_OPTIONS_MASK 7
|
|
|
|
|
|
|
|
/* single-code point functions */
|
|
|
|
|
|
|
|
U_CAPI UChar32 U_EXPORT2
|
|
|
|
ucase_tolower(UChar32 c);
|
|
|
|
|
|
|
|
U_CAPI UChar32 U_EXPORT2
|
|
|
|
ucase_toupper(UChar32 c);
|
|
|
|
|
|
|
|
U_CAPI UChar32 U_EXPORT2
|
|
|
|
ucase_totitle(UChar32 c);
|
|
|
|
|
|
|
|
U_CAPI UChar32 U_EXPORT2
|
|
|
|
ucase_fold(UChar32 c, uint32_t options);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Adds all simple case mappings and the full case folding for c to sa,
|
|
|
|
* and also adds special case closure mappings.
|
|
|
|
* c itself is not added.
|
|
|
|
* For example, the mappings
|
|
|
|
* - for s include long s
|
|
|
|
* - for sharp s include ss
|
|
|
|
* - for k include the Kelvin sign
|
|
|
|
*/
|
|
|
|
U_CFUNC void U_EXPORT2
|
|
|
|
ucase_addCaseClosure(UChar32 c, const USetAdder *sa);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Maps the string to single code points and adds the associated case closure
|
|
|
|
* mappings.
|
|
|
|
* The string is mapped to code points if it is their full case folding string.
|
|
|
|
* In other words, this performs a reverse full case folding and then
|
|
|
|
* adds the case closure items of the resulting code points.
|
|
|
|
* If the string is found and its closure applied, then
|
|
|
|
* the string itself is added as well as part of its code points' closure.
|
|
|
|
* It must be length>=0.
|
|
|
|
*
|
|
|
|
* @return true if the string was found
|
|
|
|
*/
|
|
|
|
U_CFUNC UBool U_EXPORT2
|
|
|
|
ucase_addStringCaseClosure(const UChar *s, int32_t length, const USetAdder *sa);
|
|
|
|
|
|
|
|
#ifdef __cplusplus
|
|
|
|
U_NAMESPACE_BEGIN
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Iterator over characters with more than one code point in the full default Case_Folding.
|
|
|
|
*/
|
|
|
|
class U_COMMON_API FullCaseFoldingIterator {
|
|
|
|
public:
|
|
|
|
/** Constructor. */
|
|
|
|
FullCaseFoldingIterator();
|
|
|
|
/**
|
|
|
|
* Returns the next (cp, full) pair where "full" is cp's full default Case_Folding.
|
|
|
|
* Returns a negative cp value at the end of the iteration.
|
|
|
|
*/
|
|
|
|
UChar32 next(UnicodeString &full);
|
|
|
|
private:
|
|
|
|
FullCaseFoldingIterator(const FullCaseFoldingIterator &); // no copy
|
|
|
|
FullCaseFoldingIterator &operator=(const FullCaseFoldingIterator &); // no assignment
|
|
|
|
|
|
|
|
const UChar *unfold;
|
|
|
|
int32_t unfoldRows;
|
|
|
|
int32_t unfoldRowWidth;
|
|
|
|
int32_t unfoldStringWidth;
|
|
|
|
int32_t currentRow;
|
|
|
|
int32_t rowCpIndex;
|
|
|
|
};
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Fast case mapping data for ASCII/Latin.
|
|
|
|
* Linear arrays of delta bytes: 0=no mapping; EXC=exception.
|
|
|
|
* Deltas must not cross the ASCII boundary, or else they cannot be easily used
|
|
|
|
* in simple UTF-8 code.
|
|
|
|
*/
|
|
|
|
namespace LatinCase {
|
|
|
|
|
|
|
|
/** Case mapping/folding data for code points up to U+017F. */
|
|
|
|
constexpr UChar LIMIT = 0x180;
|
|
|
|
/** U+017F case-folds and uppercases crossing the ASCII boundary. */
|
|
|
|
constexpr UChar LONG_S = 0x17f;
|
|
|
|
/** Exception: Complex mapping, or too-large delta. */
|
|
|
|
constexpr int8_t EXC = -0x80;
|
|
|
|
|
|
|
|
/** Deltas for lowercasing for most locales, and default case folding. */
|
|
|
|
extern const int8_t TO_LOWER_NORMAL[LIMIT];
|
|
|
|
/** Deltas for lowercasing for tr/az/lt, and Turkic case folding. */
|
|
|
|
extern const int8_t TO_LOWER_TR_LT[LIMIT];
|
|
|
|
|
|
|
|
/** Deltas for uppercasing for most locales. */
|
|
|
|
extern const int8_t TO_UPPER_NORMAL[LIMIT];
|
|
|
|
/** Deltas for uppercasing for tr/az. */
|
|
|
|
extern const int8_t TO_UPPER_TR[LIMIT];
|
|
|
|
|
|
|
|
} // namespace LatinCase
|
|
|
|
|
|
|
|
U_NAMESPACE_END
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/** @return UCASE_NONE, UCASE_LOWER, UCASE_UPPER, UCASE_TITLE */
|
|
|
|
U_CAPI int32_t U_EXPORT2
|
|
|
|
ucase_getType(UChar32 c);
|
|
|
|
|
|
|
|
/** @return like ucase_getType() but also sets UCASE_IGNORABLE if c is case-ignorable */
|
|
|
|
U_CAPI int32_t U_EXPORT2
|
|
|
|
ucase_getTypeOrIgnorable(UChar32 c);
|
|
|
|
|
|
|
|
U_CAPI UBool U_EXPORT2
|
|
|
|
ucase_isSoftDotted(UChar32 c);
|
|
|
|
|
|
|
|
U_CAPI UBool U_EXPORT2
|
|
|
|
ucase_isCaseSensitive(UChar32 c);
|
|
|
|
|
|
|
|
/* string case mapping functions */
|
|
|
|
|
|
|
|
U_CDECL_BEGIN
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Iterator function for string case mappings, which need to look at the
|
|
|
|
* context (surrounding text) of a given character for conditional mappings.
|
|
|
|
*
|
|
|
|
* The iterator only needs to go backward or forward away from the
|
|
|
|
* character in question. It does not use any indexes on this interface.
|
|
|
|
* It does not support random access or an arbitrary change of
|
|
|
|
* iteration direction.
|
|
|
|
*
|
|
|
|
* The code point being case-mapped itself is never returned by
|
|
|
|
* this iterator.
|
|
|
|
*
|
|
|
|
* @param context A pointer to the iterator's working data.
|
|
|
|
* @param dir If <0 then start iterating backward from the character;
|
|
|
|
* if >0 then start iterating forward from the character;
|
|
|
|
* if 0 then continue iterating in the current direction.
|
|
|
|
* @return Next code point, or <0 when the iteration is done.
|
|
|
|
*/
|
|
|
|
typedef UChar32 U_CALLCONV
|
|
|
|
UCaseContextIterator(void *context, int8_t dir);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Sample struct which may be used by some implementations of
|
|
|
|
* UCaseContextIterator.
|
|
|
|
*/
|
|
|
|
struct UCaseContext {
|
|
|
|
void *p;
|
|
|
|
int32_t start, index, limit;
|
|
|
|
int32_t cpStart, cpLimit;
|
|
|
|
int8_t dir;
|
|
|
|
int8_t b1, b2, b3;
|
|
|
|
};
|
|
|
|
typedef struct UCaseContext UCaseContext;
|
|
|
|
|
|
|
|
U_CDECL_END
|
|
|
|
|
|
|
|
#define UCASECONTEXT_INITIALIZER { NULL, 0, 0, 0, 0, 0, 0, 0, 0, 0 }
|
|
|
|
|
|
|
|
enum {
|
|
|
|
/**
|
|
|
|
* For string case mappings, a single character (a code point) is mapped
|
|
|
|
* either to itself (in which case in-place mapping functions do nothing),
|
|
|
|
* or to another single code point, or to a string.
|
|
|
|
* Aside from the string contents, these are indicated with a single int32_t
|
|
|
|
* value as follows:
|
|
|
|
*
|
|
|
|
* Mapping to self: Negative values (~self instead of -self to support U+0000)
|
|
|
|
*
|
|
|
|
* Mapping to another code point: Positive values >UCASE_MAX_STRING_LENGTH
|
|
|
|
*
|
|
|
|
* Mapping to a string: The string length (0..UCASE_MAX_STRING_LENGTH) is
|
|
|
|
* returned. Note that the string result may indeed have zero length.
|
|
|
|
*/
|
|
|
|
UCASE_MAX_STRING_LENGTH=0x1f
|
|
|
|
};
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Get the full lowercase mapping for c.
|
|
|
|
*
|
|
|
|
* @param csp Case mapping properties.
|
|
|
|
* @param c Character to be mapped.
|
|
|
|
* @param iter Character iterator, used for context-sensitive mappings.
|
|
|
|
* See UCaseContextIterator for details.
|
|
|
|
* If iter==NULL then a context-independent result is returned.
|
|
|
|
* @param context Pointer to be passed into iter.
|
|
|
|
* @param pString If the mapping result is a string, then the pointer is
|
|
|
|
* written to *pString.
|
|
|
|
* @param caseLocale Case locale value from ucase_getCaseLocale().
|
|
|
|
* @return Output code point or string length, see UCASE_MAX_STRING_LENGTH.
|
|
|
|
*
|
|
|
|
* @see UCaseContextIterator
|
|
|
|
* @see UCASE_MAX_STRING_LENGTH
|
|
|
|
* @internal
|
|
|
|
*/
|
|
|
|
U_CAPI int32_t U_EXPORT2
|
|
|
|
ucase_toFullLower(UChar32 c,
|
|
|
|
UCaseContextIterator *iter, void *context,
|
|
|
|
const UChar **pString,
|
|
|
|
int32_t caseLocale);
|
|
|
|
|
|
|
|
U_CAPI int32_t U_EXPORT2
|
|
|
|
ucase_toFullUpper(UChar32 c,
|
|
|
|
UCaseContextIterator *iter, void *context,
|
|
|
|
const UChar **pString,
|
|
|
|
int32_t caseLocale);
|
|
|
|
|
|
|
|
U_CAPI int32_t U_EXPORT2
|
|
|
|
ucase_toFullTitle(UChar32 c,
|
|
|
|
UCaseContextIterator *iter, void *context,
|
|
|
|
const UChar **pString,
|
|
|
|
int32_t caseLocale);
|
|
|
|
|
|
|
|
U_CAPI int32_t U_EXPORT2
|
|
|
|
ucase_toFullFolding(UChar32 c,
|
|
|
|
const UChar **pString,
|
|
|
|
uint32_t options);
|
|
|
|
|
|
|
|
U_CFUNC int32_t U_EXPORT2
|
|
|
|
ucase_hasBinaryProperty(UChar32 c, UProperty which);
|
|
|
|
|
|
|
|
|
|
|
|
U_CDECL_BEGIN
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @internal
|
|
|
|
*/
|
|
|
|
typedef int32_t U_CALLCONV
|
|
|
|
UCaseMapFull(UChar32 c,
|
|
|
|
UCaseContextIterator *iter, void *context,
|
|
|
|
const UChar **pString,
|
|
|
|
int32_t caseLocale);
|
|
|
|
|
|
|
|
U_CDECL_END
|
|
|
|
|
2022-05-17 15:14:19 +00:00
|
|
|
/* for icuexportdata -------------------------------------------------------- */
|
|
|
|
|
|
|
|
struct UCaseProps {
|
|
|
|
void *mem; // TODO: was unused, and type UDataMemory -- remove
|
|
|
|
const int32_t *indexes;
|
|
|
|
const uint16_t *exceptions;
|
|
|
|
const uint16_t *unfold;
|
|
|
|
|
|
|
|
UTrie2 trie;
|
|
|
|
uint8_t formatVersion[4];
|
|
|
|
};
|
|
|
|
|
|
|
|
U_CAPI const struct UCaseProps * U_EXPORT2
|
|
|
|
ucase_getSingleton(int32_t *pExceptionsLength, int32_t *pUnfoldLength);
|
|
|
|
|
2020-08-11 09:10:23 +00:00
|
|
|
/* file definitions --------------------------------------------------------- */
|
|
|
|
|
|
|
|
#define UCASE_DATA_NAME "ucase"
|
|
|
|
#define UCASE_DATA_TYPE "icu"
|
|
|
|
|
|
|
|
/* format "cAsE" */
|
|
|
|
#define UCASE_FMT_0 0x63
|
|
|
|
#define UCASE_FMT_1 0x41
|
|
|
|
#define UCASE_FMT_2 0x53
|
|
|
|
#define UCASE_FMT_3 0x45
|
|
|
|
|
|
|
|
/* indexes into indexes[] */
|
|
|
|
enum {
|
|
|
|
UCASE_IX_INDEX_TOP,
|
|
|
|
UCASE_IX_LENGTH,
|
|
|
|
UCASE_IX_TRIE_SIZE,
|
|
|
|
UCASE_IX_EXC_LENGTH,
|
|
|
|
UCASE_IX_UNFOLD_LENGTH,
|
|
|
|
|
|
|
|
UCASE_IX_MAX_FULL_LENGTH=15,
|
|
|
|
UCASE_IX_TOP=16
|
|
|
|
};
|
|
|
|
|
|
|
|
/* definitions for 16-bit case properties word ------------------------------ */
|
|
|
|
|
|
|
|
U_CFUNC const UTrie2 * U_EXPORT2
|
|
|
|
ucase_getTrie();
|
|
|
|
|
|
|
|
/* 2-bit constants for types of cased characters */
|
|
|
|
#define UCASE_TYPE_MASK 3
|
|
|
|
enum {
|
|
|
|
UCASE_NONE,
|
|
|
|
UCASE_LOWER,
|
|
|
|
UCASE_UPPER,
|
|
|
|
UCASE_TITLE
|
|
|
|
};
|
|
|
|
|
|
|
|
#define UCASE_GET_TYPE(props) ((props)&UCASE_TYPE_MASK)
|
|
|
|
#define UCASE_GET_TYPE_AND_IGNORABLE(props) ((props)&7)
|
|
|
|
|
|
|
|
#define UCASE_IS_UPPER_OR_TITLE(props) ((props)&2)
|
|
|
|
|
|
|
|
#define UCASE_IGNORABLE 4
|
|
|
|
#define UCASE_EXCEPTION 8
|
|
|
|
#define UCASE_SENSITIVE 0x10
|
|
|
|
|
|
|
|
#define UCASE_HAS_EXCEPTION(props) ((props)&UCASE_EXCEPTION)
|
|
|
|
|
|
|
|
#define UCASE_DOT_MASK 0x60
|
|
|
|
enum {
|
|
|
|
UCASE_NO_DOT=0, /* normal characters with cc=0 */
|
|
|
|
UCASE_SOFT_DOTTED=0x20, /* soft-dotted characters with cc=0 */
|
|
|
|
UCASE_ABOVE=0x40, /* "above" accents with cc=230 */
|
|
|
|
UCASE_OTHER_ACCENT=0x60 /* other accent character (0<cc!=230) */
|
|
|
|
};
|
|
|
|
|
|
|
|
/* no exception: bits 15..7 are a 9-bit signed case mapping delta */
|
|
|
|
#define UCASE_DELTA_SHIFT 7
|
|
|
|
#define UCASE_DELTA_MASK 0xff80
|
|
|
|
#define UCASE_MAX_DELTA 0xff
|
|
|
|
#define UCASE_MIN_DELTA (-UCASE_MAX_DELTA-1)
|
|
|
|
|
|
|
|
#if U_SIGNED_RIGHT_SHIFT_IS_ARITHMETIC
|
|
|
|
# define UCASE_GET_DELTA(props) ((int16_t)(props)>>UCASE_DELTA_SHIFT)
|
|
|
|
#else
|
|
|
|
# define UCASE_GET_DELTA(props) (int16_t)(((props)&0x8000) ? (((props)>>UCASE_DELTA_SHIFT)|0xfe00) : ((uint16_t)(props)>>UCASE_DELTA_SHIFT))
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/* exception: bits 15..4 are an unsigned 12-bit index into the exceptions array */
|
|
|
|
#define UCASE_EXC_SHIFT 4
|
|
|
|
#define UCASE_EXC_MASK 0xfff0
|
|
|
|
#define UCASE_MAX_EXCEPTIONS ((UCASE_EXC_MASK>>UCASE_EXC_SHIFT)+1)
|
|
|
|
|
|
|
|
/* definitions for 16-bit main exceptions word ------------------------------ */
|
|
|
|
|
|
|
|
/* first 8 bits indicate values in optional slots */
|
|
|
|
enum {
|
|
|
|
UCASE_EXC_LOWER,
|
|
|
|
UCASE_EXC_FOLD,
|
|
|
|
UCASE_EXC_UPPER,
|
|
|
|
UCASE_EXC_TITLE,
|
|
|
|
UCASE_EXC_DELTA,
|
|
|
|
UCASE_EXC_5, /* reserved */
|
|
|
|
UCASE_EXC_CLOSURE,
|
|
|
|
UCASE_EXC_FULL_MAPPINGS,
|
|
|
|
UCASE_EXC_ALL_SLOTS /* one past the last slot */
|
|
|
|
};
|
|
|
|
|
|
|
|
/* each slot is 2 uint16_t instead of 1 */
|
|
|
|
#define UCASE_EXC_DOUBLE_SLOTS 0x100
|
|
|
|
|
|
|
|
enum {
|
|
|
|
UCASE_EXC_NO_SIMPLE_CASE_FOLDING=0x200,
|
|
|
|
UCASE_EXC_DELTA_IS_NEGATIVE=0x400,
|
|
|
|
UCASE_EXC_SENSITIVE=0x800
|
|
|
|
};
|
|
|
|
|
|
|
|
/* UCASE_EXC_DOT_MASK=UCASE_DOT_MASK<<UCASE_EXC_DOT_SHIFT */
|
|
|
|
#define UCASE_EXC_DOT_SHIFT 7
|
|
|
|
|
|
|
|
/* normally stored in the main word, but pushed out for larger exception indexes */
|
|
|
|
#define UCASE_EXC_DOT_MASK 0x3000
|
|
|
|
enum {
|
|
|
|
UCASE_EXC_NO_DOT=0,
|
|
|
|
UCASE_EXC_SOFT_DOTTED=0x1000,
|
|
|
|
UCASE_EXC_ABOVE=0x2000, /* "above" accents with cc=230 */
|
|
|
|
UCASE_EXC_OTHER_ACCENT=0x3000 /* other character (0<cc!=230) */
|
|
|
|
};
|
|
|
|
|
|
|
|
/* complex/conditional mappings */
|
|
|
|
#define UCASE_EXC_CONDITIONAL_SPECIAL 0x4000
|
|
|
|
#define UCASE_EXC_CONDITIONAL_FOLD 0x8000
|
|
|
|
|
|
|
|
/* definitions for lengths word for full case mappings */
|
|
|
|
#define UCASE_FULL_LOWER 0xf
|
|
|
|
#define UCASE_FULL_FOLDING 0xf0
|
|
|
|
#define UCASE_FULL_UPPER 0xf00
|
|
|
|
#define UCASE_FULL_TITLE 0xf000
|
|
|
|
|
|
|
|
/* maximum lengths */
|
|
|
|
#define UCASE_FULL_MAPPINGS_MAX_LENGTH (4*0xf)
|
|
|
|
#define UCASE_CLOSURE_MAX_LENGTH 0xf
|
|
|
|
|
|
|
|
/* constants for reverse case folding ("unfold") data */
|
|
|
|
enum {
|
|
|
|
UCASE_UNFOLD_ROWS,
|
|
|
|
UCASE_UNFOLD_ROW_WIDTH,
|
|
|
|
UCASE_UNFOLD_STRING_WIDTH
|
|
|
|
};
|
|
|
|
|
|
|
|
#endif
|