2109 lines
65 KiB
C++
2109 lines
65 KiB
C++
// © 2016 and later: Unicode, Inc. and others.
|
|
// License & terms of use: http://www.unicode.org/copyright.html
|
|
/*
|
|
******************************************************************************
|
|
*
|
|
* Copyright (C) 1999-2014, International Business Machines
|
|
* Corporation and others. All Rights Reserved.
|
|
*
|
|
******************************************************************************
|
|
* file name: unames.c
|
|
* encoding: UTF-8
|
|
* tab size: 8 (not used)
|
|
* indentation:4
|
|
*
|
|
* created on: 1999oct04
|
|
* created by: Markus W. Scherer
|
|
*/
|
|
|
|
#include "unicode/utypes.h"
|
|
#include "unicode/putil.h"
|
|
#include "unicode/uchar.h"
|
|
#include "unicode/udata.h"
|
|
#include "unicode/utf.h"
|
|
#include "unicode/utf16.h"
|
|
#include "uassert.h"
|
|
#include "ustr_imp.h"
|
|
#include "umutex.h"
|
|
#include "cmemory.h"
|
|
#include "cstring.h"
|
|
#include "ucln_cmn.h"
|
|
#include "udataswp.h"
|
|
#include "uprops.h"
|
|
|
|
U_NAMESPACE_BEGIN
|
|
|
|
/* prototypes ------------------------------------------------------------- */
|
|
|
|
static const char DATA_NAME[] = "unames";
|
|
static const char DATA_TYPE[] = "icu";
|
|
|
|
#define GROUP_SHIFT 5
|
|
#define LINES_PER_GROUP (1L<<GROUP_SHIFT)
|
|
#define GROUP_MASK (LINES_PER_GROUP-1)
|
|
|
|
/*
|
|
* This struct was replaced by explicitly accessing equivalent
|
|
* fields from triples of uint16_t.
|
|
* The Group struct was padded to 8 bytes on compilers for early ARM CPUs,
|
|
* which broke the assumption that sizeof(Group)==6 and that the ++ operator
|
|
* would advance by 6 bytes (3 uint16_t).
|
|
*
|
|
* We can't just change the data structure because it's loaded from a data file,
|
|
* and we don't want to make it less compact, so we changed the access code.
|
|
*
|
|
* For details see ICU tickets 6331 and 6008.
|
|
typedef struct {
|
|
uint16_t groupMSB,
|
|
offsetHigh, offsetLow; / * avoid padding * /
|
|
} Group;
|
|
*/
|
|
enum {
|
|
GROUP_MSB,
|
|
GROUP_OFFSET_HIGH,
|
|
GROUP_OFFSET_LOW,
|
|
GROUP_LENGTH
|
|
};
|
|
|
|
/*
|
|
* Get the 32-bit group offset.
|
|
* @param group (const uint16_t *) pointer to a Group triple of uint16_t
|
|
* @return group offset (int32_t)
|
|
*/
|
|
#define GET_GROUP_OFFSET(group) ((int32_t)(group)[GROUP_OFFSET_HIGH]<<16|(group)[GROUP_OFFSET_LOW])
|
|
|
|
#define NEXT_GROUP(group) ((group)+GROUP_LENGTH)
|
|
#define PREV_GROUP(group) ((group)-GROUP_LENGTH)
|
|
|
|
typedef struct {
|
|
uint32_t start, end;
|
|
uint8_t type, variant;
|
|
uint16_t size;
|
|
} AlgorithmicRange;
|
|
|
|
typedef struct {
|
|
uint32_t tokenStringOffset, groupsOffset, groupStringOffset, algNamesOffset;
|
|
} UCharNames;
|
|
|
|
/*
|
|
* Get the groups table from a UCharNames struct.
|
|
* The groups table consists of one uint16_t groupCount followed by
|
|
* groupCount groups. Each group is a triple of uint16_t, see GROUP_LENGTH
|
|
* and the comment for the old struct Group above.
|
|
*
|
|
* @param names (const UCharNames *) pointer to the UCharNames indexes
|
|
* @return (const uint16_t *) pointer to the groups table
|
|
*/
|
|
#define GET_GROUPS(names) (const uint16_t *)((const char *)names+names->groupsOffset)
|
|
|
|
typedef struct {
|
|
const char *otherName;
|
|
UChar32 code;
|
|
} FindName;
|
|
|
|
#define DO_FIND_NAME NULL
|
|
|
|
static UDataMemory *uCharNamesData=NULL;
|
|
static UCharNames *uCharNames=NULL;
|
|
static icu::UInitOnce gCharNamesInitOnce {};
|
|
|
|
/*
|
|
* Maximum length of character names (regular & 1.0).
|
|
*/
|
|
static int32_t gMaxNameLength=0;
|
|
|
|
/*
|
|
* Set of chars used in character names (regular & 1.0).
|
|
* Chars are platform-dependent (can be EBCDIC).
|
|
*/
|
|
static uint32_t gNameSet[8]={ 0 };
|
|
|
|
#define U_NONCHARACTER_CODE_POINT U_CHAR_CATEGORY_COUNT
|
|
#define U_LEAD_SURROGATE U_CHAR_CATEGORY_COUNT + 1
|
|
#define U_TRAIL_SURROGATE U_CHAR_CATEGORY_COUNT + 2
|
|
|
|
#define U_CHAR_EXTENDED_CATEGORY_COUNT (U_CHAR_CATEGORY_COUNT + 3)
|
|
|
|
static const char * const charCatNames[U_CHAR_EXTENDED_CATEGORY_COUNT] = {
|
|
"unassigned",
|
|
"uppercase letter",
|
|
"lowercase letter",
|
|
"titlecase letter",
|
|
"modifier letter",
|
|
"other letter",
|
|
"non spacing mark",
|
|
"enclosing mark",
|
|
"combining spacing mark",
|
|
"decimal digit number",
|
|
"letter number",
|
|
"other number",
|
|
"space separator",
|
|
"line separator",
|
|
"paragraph separator",
|
|
"control",
|
|
"format",
|
|
"private use area",
|
|
"surrogate",
|
|
"dash punctuation",
|
|
"start punctuation",
|
|
"end punctuation",
|
|
"connector punctuation",
|
|
"other punctuation",
|
|
"math symbol",
|
|
"currency symbol",
|
|
"modifier symbol",
|
|
"other symbol",
|
|
"initial punctuation",
|
|
"final punctuation",
|
|
"noncharacter",
|
|
"lead surrogate",
|
|
"trail surrogate"
|
|
};
|
|
|
|
/* implementation ----------------------------------------------------------- */
|
|
|
|
static UBool U_CALLCONV unames_cleanup(void)
|
|
{
|
|
if(uCharNamesData) {
|
|
udata_close(uCharNamesData);
|
|
uCharNamesData = NULL;
|
|
}
|
|
if(uCharNames) {
|
|
uCharNames = NULL;
|
|
}
|
|
gCharNamesInitOnce.reset();
|
|
gMaxNameLength=0;
|
|
return true;
|
|
}
|
|
|
|
static UBool U_CALLCONV
|
|
isAcceptable(void * /*context*/,
|
|
const char * /*type*/, const char * /*name*/,
|
|
const UDataInfo *pInfo) {
|
|
return (UBool)(
|
|
pInfo->size>=20 &&
|
|
pInfo->isBigEndian==U_IS_BIG_ENDIAN &&
|
|
pInfo->charsetFamily==U_CHARSET_FAMILY &&
|
|
pInfo->dataFormat[0]==0x75 && /* dataFormat="unam" */
|
|
pInfo->dataFormat[1]==0x6e &&
|
|
pInfo->dataFormat[2]==0x61 &&
|
|
pInfo->dataFormat[3]==0x6d &&
|
|
pInfo->formatVersion[0]==1);
|
|
}
|
|
|
|
static void U_CALLCONV
|
|
loadCharNames(UErrorCode &status) {
|
|
U_ASSERT(uCharNamesData == NULL);
|
|
U_ASSERT(uCharNames == NULL);
|
|
|
|
uCharNamesData = udata_openChoice(NULL, DATA_TYPE, DATA_NAME, isAcceptable, NULL, &status);
|
|
if(U_FAILURE(status)) {
|
|
uCharNamesData = NULL;
|
|
} else {
|
|
uCharNames = (UCharNames *)udata_getMemory(uCharNamesData);
|
|
}
|
|
ucln_common_registerCleanup(UCLN_COMMON_UNAMES, unames_cleanup);
|
|
}
|
|
|
|
|
|
static UBool
|
|
isDataLoaded(UErrorCode *pErrorCode) {
|
|
umtx_initOnce(gCharNamesInitOnce, &loadCharNames, *pErrorCode);
|
|
return U_SUCCESS(*pErrorCode);
|
|
}
|
|
|
|
#define WRITE_CHAR(buffer, bufferLength, bufferPos, c) UPRV_BLOCK_MACRO_BEGIN { \
|
|
if((bufferLength)>0) { \
|
|
*(buffer)++=c; \
|
|
--(bufferLength); \
|
|
} \
|
|
++(bufferPos); \
|
|
} UPRV_BLOCK_MACRO_END
|
|
|
|
#define U_ISO_COMMENT U_CHAR_NAME_CHOICE_COUNT
|
|
|
|
/*
|
|
* Important: expandName() and compareName() are almost the same -
|
|
* apply fixes to both.
|
|
*
|
|
* UnicodeData.txt uses ';' as a field separator, so no
|
|
* field can contain ';' as part of its contents.
|
|
* In unames.dat, it is marked as token[';']==-1 only if the
|
|
* semicolon is used in the data file - which is iff we
|
|
* have Unicode 1.0 names or ISO comments or aliases.
|
|
* So, it will be token[';']==-1 if we store U1.0 names/ISO comments/aliases
|
|
* although we know that it will never be part of a name.
|
|
*/
|
|
static uint16_t
|
|
expandName(UCharNames *names,
|
|
const uint8_t *name, uint16_t nameLength, UCharNameChoice nameChoice,
|
|
char *buffer, uint16_t bufferLength) {
|
|
uint16_t *tokens=(uint16_t *)names+8;
|
|
uint16_t token, tokenCount=*tokens++, bufferPos=0;
|
|
uint8_t *tokenStrings=(uint8_t *)names+names->tokenStringOffset;
|
|
uint8_t c;
|
|
|
|
if(nameChoice!=U_UNICODE_CHAR_NAME && nameChoice!=U_EXTENDED_CHAR_NAME) {
|
|
/*
|
|
* skip the modern name if it is not requested _and_
|
|
* if the semicolon byte value is a character, not a token number
|
|
*/
|
|
if((uint8_t)';'>=tokenCount || tokens[(uint8_t)';']==(uint16_t)(-1)) {
|
|
int fieldIndex= nameChoice==U_ISO_COMMENT ? 2 : nameChoice;
|
|
do {
|
|
while(nameLength>0) {
|
|
--nameLength;
|
|
if(*name++==';') {
|
|
break;
|
|
}
|
|
}
|
|
} while(--fieldIndex>0);
|
|
} else {
|
|
/*
|
|
* the semicolon byte value is a token number, therefore
|
|
* only modern names are stored in unames.dat and there is no
|
|
* such requested alternate name here
|
|
*/
|
|
nameLength=0;
|
|
}
|
|
}
|
|
|
|
/* write each letter directly, and write a token word per token */
|
|
while(nameLength>0) {
|
|
--nameLength;
|
|
c=*name++;
|
|
|
|
if(c>=tokenCount) {
|
|
if(c!=';') {
|
|
/* implicit letter */
|
|
WRITE_CHAR(buffer, bufferLength, bufferPos, c);
|
|
} else {
|
|
/* finished */
|
|
break;
|
|
}
|
|
} else {
|
|
token=tokens[c];
|
|
if(token==(uint16_t)(-2)) {
|
|
/* this is a lead byte for a double-byte token */
|
|
token=tokens[c<<8|*name++];
|
|
--nameLength;
|
|
}
|
|
if(token==(uint16_t)(-1)) {
|
|
if(c!=';') {
|
|
/* explicit letter */
|
|
WRITE_CHAR(buffer, bufferLength, bufferPos, c);
|
|
} else {
|
|
/* stop, but skip the semicolon if we are seeking
|
|
extended names and there was no 2.0 name but there
|
|
is a 1.0 name. */
|
|
if(!bufferPos && nameChoice == U_EXTENDED_CHAR_NAME) {
|
|
if ((uint8_t)';'>=tokenCount || tokens[(uint8_t)';']==(uint16_t)(-1)) {
|
|
continue;
|
|
}
|
|
}
|
|
/* finished */
|
|
break;
|
|
}
|
|
} else {
|
|
/* write token word */
|
|
uint8_t *tokenString=tokenStrings+token;
|
|
while((c=*tokenString++)!=0) {
|
|
WRITE_CHAR(buffer, bufferLength, bufferPos, c);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
/* zero-terminate */
|
|
if(bufferLength>0) {
|
|
*buffer=0;
|
|
}
|
|
|
|
return bufferPos;
|
|
}
|
|
|
|
/*
|
|
* compareName() is almost the same as expandName() except that it compares
|
|
* the currently expanded name to an input name.
|
|
* It returns the match/no match result as soon as possible.
|
|
*/
|
|
static UBool
|
|
compareName(UCharNames *names,
|
|
const uint8_t *name, uint16_t nameLength, UCharNameChoice nameChoice,
|
|
const char *otherName) {
|
|
uint16_t *tokens=(uint16_t *)names+8;
|
|
uint16_t token, tokenCount=*tokens++;
|
|
uint8_t *tokenStrings=(uint8_t *)names+names->tokenStringOffset;
|
|
uint8_t c;
|
|
const char *origOtherName = otherName;
|
|
|
|
if(nameChoice!=U_UNICODE_CHAR_NAME && nameChoice!=U_EXTENDED_CHAR_NAME) {
|
|
/*
|
|
* skip the modern name if it is not requested _and_
|
|
* if the semicolon byte value is a character, not a token number
|
|
*/
|
|
if((uint8_t)';'>=tokenCount || tokens[(uint8_t)';']==(uint16_t)(-1)) {
|
|
int fieldIndex= nameChoice==U_ISO_COMMENT ? 2 : nameChoice;
|
|
do {
|
|
while(nameLength>0) {
|
|
--nameLength;
|
|
if(*name++==';') {
|
|
break;
|
|
}
|
|
}
|
|
} while(--fieldIndex>0);
|
|
} else {
|
|
/*
|
|
* the semicolon byte value is a token number, therefore
|
|
* only modern names are stored in unames.dat and there is no
|
|
* such requested alternate name here
|
|
*/
|
|
nameLength=0;
|
|
}
|
|
}
|
|
|
|
/* compare each letter directly, and compare a token word per token */
|
|
while(nameLength>0) {
|
|
--nameLength;
|
|
c=*name++;
|
|
|
|
if(c>=tokenCount) {
|
|
if(c!=';') {
|
|
/* implicit letter */
|
|
if((char)c!=*otherName++) {
|
|
return false;
|
|
}
|
|
} else {
|
|
/* finished */
|
|
break;
|
|
}
|
|
} else {
|
|
token=tokens[c];
|
|
if(token==(uint16_t)(-2)) {
|
|
/* this is a lead byte for a double-byte token */
|
|
token=tokens[c<<8|*name++];
|
|
--nameLength;
|
|
}
|
|
if(token==(uint16_t)(-1)) {
|
|
if(c!=';') {
|
|
/* explicit letter */
|
|
if((char)c!=*otherName++) {
|
|
return false;
|
|
}
|
|
} else {
|
|
/* stop, but skip the semicolon if we are seeking
|
|
extended names and there was no 2.0 name but there
|
|
is a 1.0 name. */
|
|
if(otherName == origOtherName && nameChoice == U_EXTENDED_CHAR_NAME) {
|
|
if ((uint8_t)';'>=tokenCount || tokens[(uint8_t)';']==(uint16_t)(-1)) {
|
|
continue;
|
|
}
|
|
}
|
|
/* finished */
|
|
break;
|
|
}
|
|
} else {
|
|
/* write token word */
|
|
uint8_t *tokenString=tokenStrings+token;
|
|
while((c=*tokenString++)!=0) {
|
|
if((char)c!=*otherName++) {
|
|
return false;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
/* complete match? */
|
|
return (UBool)(*otherName==0);
|
|
}
|
|
|
|
static uint8_t getCharCat(UChar32 cp) {
|
|
uint8_t cat;
|
|
|
|
if (U_IS_UNICODE_NONCHAR(cp)) {
|
|
return U_NONCHARACTER_CODE_POINT;
|
|
}
|
|
|
|
if ((cat = u_charType(cp)) == U_SURROGATE) {
|
|
cat = U_IS_LEAD(cp) ? U_LEAD_SURROGATE : U_TRAIL_SURROGATE;
|
|
}
|
|
|
|
return cat;
|
|
}
|
|
|
|
static const char *getCharCatName(UChar32 cp) {
|
|
uint8_t cat = getCharCat(cp);
|
|
|
|
/* Return unknown if the table of names above is not up to
|
|
date. */
|
|
|
|
if (cat >= UPRV_LENGTHOF(charCatNames)) {
|
|
return "unknown";
|
|
} else {
|
|
return charCatNames[cat];
|
|
}
|
|
}
|
|
|
|
static uint16_t getExtName(uint32_t code, char *buffer, uint16_t bufferLength) {
|
|
const char *catname = getCharCatName(code);
|
|
uint16_t length = 0;
|
|
|
|
UChar32 cp;
|
|
int ndigits, i;
|
|
|
|
WRITE_CHAR(buffer, bufferLength, length, '<');
|
|
while (catname[length - 1]) {
|
|
WRITE_CHAR(buffer, bufferLength, length, catname[length - 1]);
|
|
}
|
|
WRITE_CHAR(buffer, bufferLength, length, '-');
|
|
for (cp = code, ndigits = 0; cp; ++ndigits, cp >>= 4)
|
|
;
|
|
if (ndigits < 4)
|
|
ndigits = 4;
|
|
for (cp = code, i = ndigits; (cp || i > 0) && bufferLength; cp >>= 4, bufferLength--) {
|
|
uint8_t v = (uint8_t)(cp & 0xf);
|
|
buffer[--i] = (v < 10 ? '0' + v : 'A' + v - 10);
|
|
}
|
|
buffer += ndigits;
|
|
length += static_cast<uint16_t>(ndigits);
|
|
WRITE_CHAR(buffer, bufferLength, length, '>');
|
|
|
|
return length;
|
|
}
|
|
|
|
/*
|
|
* getGroup() does a binary search for the group that contains the
|
|
* Unicode code point "code".
|
|
* The return value is always a valid Group* that may contain "code"
|
|
* or else is the highest group before "code".
|
|
* If the lowest group is after "code", then that one is returned.
|
|
*/
|
|
static const uint16_t *
|
|
getGroup(UCharNames *names, uint32_t code) {
|
|
const uint16_t *groups=GET_GROUPS(names);
|
|
uint16_t groupMSB=(uint16_t)(code>>GROUP_SHIFT),
|
|
start=0,
|
|
limit=*groups++,
|
|
number;
|
|
|
|
/* binary search for the group of names that contains the one for code */
|
|
while(start<limit-1) {
|
|
number=(uint16_t)((start+limit)/2);
|
|
if(groupMSB<groups[number*GROUP_LENGTH+GROUP_MSB]) {
|
|
limit=number;
|
|
} else {
|
|
start=number;
|
|
}
|
|
}
|
|
|
|
/* return this regardless of whether it is an exact match */
|
|
return groups+start*GROUP_LENGTH;
|
|
}
|
|
|
|
/*
|
|
* expandGroupLengths() reads a block of compressed lengths of 32 strings and
|
|
* expands them into offsets and lengths for each string.
|
|
* Lengths are stored with a variable-width encoding in consecutive nibbles:
|
|
* If a nibble<0xc, then it is the length itself (0=empty string).
|
|
* If a nibble>=0xc, then it forms a length value with the following nibble.
|
|
* Calculation see below.
|
|
* The offsets and lengths arrays must be at least 33 (one more) long because
|
|
* there is no check here at the end if the last nibble is still used.
|
|
*/
|
|
static const uint8_t *
|
|
expandGroupLengths(const uint8_t *s,
|
|
uint16_t offsets[LINES_PER_GROUP+1], uint16_t lengths[LINES_PER_GROUP+1]) {
|
|
/* read the lengths of the 32 strings in this group and get each string's offset */
|
|
uint16_t i=0, offset=0, length=0;
|
|
uint8_t lengthByte;
|
|
|
|
/* all 32 lengths must be read to get the offset of the first group string */
|
|
while(i<LINES_PER_GROUP) {
|
|
lengthByte=*s++;
|
|
|
|
/* read even nibble - MSBs of lengthByte */
|
|
if(length>=12) {
|
|
/* double-nibble length spread across two bytes */
|
|
length=(uint16_t)(((length&0x3)<<4|lengthByte>>4)+12);
|
|
lengthByte&=0xf;
|
|
} else if((lengthByte /* &0xf0 */)>=0xc0) {
|
|
/* double-nibble length spread across this one byte */
|
|
length=(uint16_t)((lengthByte&0x3f)+12);
|
|
} else {
|
|
/* single-nibble length in MSBs */
|
|
length=(uint16_t)(lengthByte>>4);
|
|
lengthByte&=0xf;
|
|
}
|
|
|
|
*offsets++=offset;
|
|
*lengths++=length;
|
|
|
|
offset+=length;
|
|
++i;
|
|
|
|
/* read odd nibble - LSBs of lengthByte */
|
|
if((lengthByte&0xf0)==0) {
|
|
/* this nibble was not consumed for a double-nibble length above */
|
|
length=lengthByte;
|
|
if(length<12) {
|
|
/* single-nibble length in LSBs */
|
|
*offsets++=offset;
|
|
*lengths++=length;
|
|
|
|
offset+=length;
|
|
++i;
|
|
}
|
|
} else {
|
|
length=0; /* prevent double-nibble detection in the next iteration */
|
|
}
|
|
}
|
|
|
|
/* now, s is at the first group string */
|
|
return s;
|
|
}
|
|
|
|
static uint16_t
|
|
expandGroupName(UCharNames *names, const uint16_t *group,
|
|
uint16_t lineNumber, UCharNameChoice nameChoice,
|
|
char *buffer, uint16_t bufferLength) {
|
|
uint16_t offsets[LINES_PER_GROUP+2], lengths[LINES_PER_GROUP+2];
|
|
const uint8_t *s=(uint8_t *)names+names->groupStringOffset+GET_GROUP_OFFSET(group);
|
|
s=expandGroupLengths(s, offsets, lengths);
|
|
return expandName(names, s+offsets[lineNumber], lengths[lineNumber], nameChoice,
|
|
buffer, bufferLength);
|
|
}
|
|
|
|
static uint16_t
|
|
getName(UCharNames *names, uint32_t code, UCharNameChoice nameChoice,
|
|
char *buffer, uint16_t bufferLength) {
|
|
const uint16_t *group=getGroup(names, code);
|
|
if((uint16_t)(code>>GROUP_SHIFT)==group[GROUP_MSB]) {
|
|
return expandGroupName(names, group, (uint16_t)(code&GROUP_MASK), nameChoice,
|
|
buffer, bufferLength);
|
|
} else {
|
|
/* group not found */
|
|
/* zero-terminate */
|
|
if(bufferLength>0) {
|
|
*buffer=0;
|
|
}
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* enumGroupNames() enumerates all the names in a 32-group
|
|
* and either calls the enumerator function or finds a given input name.
|
|
*/
|
|
static UBool
|
|
enumGroupNames(UCharNames *names, const uint16_t *group,
|
|
UChar32 start, UChar32 end,
|
|
UEnumCharNamesFn *fn, void *context,
|
|
UCharNameChoice nameChoice) {
|
|
uint16_t offsets[LINES_PER_GROUP+2], lengths[LINES_PER_GROUP+2];
|
|
const uint8_t *s=(uint8_t *)names+names->groupStringOffset+GET_GROUP_OFFSET(group);
|
|
|
|
s=expandGroupLengths(s, offsets, lengths);
|
|
if(fn!=DO_FIND_NAME) {
|
|
char buffer[200];
|
|
uint16_t length;
|
|
|
|
while(start<=end) {
|
|
length=expandName(names, s+offsets[start&GROUP_MASK], lengths[start&GROUP_MASK], nameChoice, buffer, sizeof(buffer));
|
|
if (!length && nameChoice == U_EXTENDED_CHAR_NAME) {
|
|
buffer[length = getExtName(start, buffer, sizeof(buffer))] = 0;
|
|
}
|
|
/* here, we assume that the buffer is large enough */
|
|
if(length>0) {
|
|
if(!fn(context, start, nameChoice, buffer, length)) {
|
|
return false;
|
|
}
|
|
}
|
|
++start;
|
|
}
|
|
} else {
|
|
const char *otherName=((FindName *)context)->otherName;
|
|
while(start<=end) {
|
|
if(compareName(names, s+offsets[start&GROUP_MASK], lengths[start&GROUP_MASK], nameChoice, otherName)) {
|
|
((FindName *)context)->code=start;
|
|
return false;
|
|
}
|
|
++start;
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
|
|
/*
|
|
* enumExtNames enumerate extended names.
|
|
* It only needs to do it if it is called with a real function and not
|
|
* with the dummy DO_FIND_NAME, because u_charFromName() does a check
|
|
* for extended names by itself.
|
|
*/
|
|
static UBool
|
|
enumExtNames(UChar32 start, UChar32 end,
|
|
UEnumCharNamesFn *fn, void *context)
|
|
{
|
|
if(fn!=DO_FIND_NAME) {
|
|
char buffer[200];
|
|
uint16_t length;
|
|
|
|
while(start<=end) {
|
|
buffer[length = getExtName(start, buffer, sizeof(buffer))] = 0;
|
|
/* here, we assume that the buffer is large enough */
|
|
if(length>0) {
|
|
if(!fn(context, start, U_EXTENDED_CHAR_NAME, buffer, length)) {
|
|
return false;
|
|
}
|
|
}
|
|
++start;
|
|
}
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static UBool
|
|
enumNames(UCharNames *names,
|
|
UChar32 start, UChar32 limit,
|
|
UEnumCharNamesFn *fn, void *context,
|
|
UCharNameChoice nameChoice) {
|
|
uint16_t startGroupMSB, endGroupMSB, groupCount;
|
|
const uint16_t *group, *groupLimit;
|
|
|
|
startGroupMSB=(uint16_t)(start>>GROUP_SHIFT);
|
|
endGroupMSB=(uint16_t)((limit-1)>>GROUP_SHIFT);
|
|
|
|
/* find the group that contains start, or the highest before it */
|
|
group=getGroup(names, start);
|
|
|
|
if(startGroupMSB<group[GROUP_MSB] && nameChoice==U_EXTENDED_CHAR_NAME) {
|
|
/* enumerate synthetic names between start and the group start */
|
|
UChar32 extLimit=((UChar32)group[GROUP_MSB]<<GROUP_SHIFT);
|
|
if(extLimit>limit) {
|
|
extLimit=limit;
|
|
}
|
|
if(!enumExtNames(start, extLimit-1, fn, context)) {
|
|
return false;
|
|
}
|
|
start=extLimit;
|
|
}
|
|
|
|
if(startGroupMSB==endGroupMSB) {
|
|
if(startGroupMSB==group[GROUP_MSB]) {
|
|
/* if start and limit-1 are in the same group, then enumerate only in that one */
|
|
return enumGroupNames(names, group, start, limit-1, fn, context, nameChoice);
|
|
}
|
|
} else {
|
|
const uint16_t *groups=GET_GROUPS(names);
|
|
groupCount=*groups++;
|
|
groupLimit=groups+groupCount*GROUP_LENGTH;
|
|
|
|
if(startGroupMSB==group[GROUP_MSB]) {
|
|
/* enumerate characters in the partial start group */
|
|
if((start&GROUP_MASK)!=0) {
|
|
if(!enumGroupNames(names, group,
|
|
start, ((UChar32)startGroupMSB<<GROUP_SHIFT)+LINES_PER_GROUP-1,
|
|
fn, context, nameChoice)) {
|
|
return false;
|
|
}
|
|
group=NEXT_GROUP(group); /* continue with the next group */
|
|
}
|
|
} else if(startGroupMSB>group[GROUP_MSB]) {
|
|
/* make sure that we start enumerating with the first group after start */
|
|
const uint16_t *nextGroup=NEXT_GROUP(group);
|
|
if (nextGroup < groupLimit && nextGroup[GROUP_MSB] > startGroupMSB && nameChoice == U_EXTENDED_CHAR_NAME) {
|
|
UChar32 end = nextGroup[GROUP_MSB] << GROUP_SHIFT;
|
|
if (end > limit) {
|
|
end = limit;
|
|
}
|
|
if (!enumExtNames(start, end - 1, fn, context)) {
|
|
return false;
|
|
}
|
|
}
|
|
group=nextGroup;
|
|
}
|
|
|
|
/* enumerate entire groups between the start- and end-groups */
|
|
while(group<groupLimit && group[GROUP_MSB]<endGroupMSB) {
|
|
const uint16_t *nextGroup;
|
|
start=(UChar32)group[GROUP_MSB]<<GROUP_SHIFT;
|
|
if(!enumGroupNames(names, group, start, start+LINES_PER_GROUP-1, fn, context, nameChoice)) {
|
|
return false;
|
|
}
|
|
nextGroup=NEXT_GROUP(group);
|
|
if (nextGroup < groupLimit && nextGroup[GROUP_MSB] > group[GROUP_MSB] + 1 && nameChoice == U_EXTENDED_CHAR_NAME) {
|
|
UChar32 end = nextGroup[GROUP_MSB] << GROUP_SHIFT;
|
|
if (end > limit) {
|
|
end = limit;
|
|
}
|
|
if (!enumExtNames((group[GROUP_MSB] + 1) << GROUP_SHIFT, end - 1, fn, context)) {
|
|
return false;
|
|
}
|
|
}
|
|
group=nextGroup;
|
|
}
|
|
|
|
/* enumerate within the end group (group[GROUP_MSB]==endGroupMSB) */
|
|
if(group<groupLimit && group[GROUP_MSB]==endGroupMSB) {
|
|
return enumGroupNames(names, group, (limit-1)&~GROUP_MASK, limit-1, fn, context, nameChoice);
|
|
} else if (nameChoice == U_EXTENDED_CHAR_NAME && group == groupLimit) {
|
|
UChar32 next = (PREV_GROUP(group)[GROUP_MSB] + 1) << GROUP_SHIFT;
|
|
if (next > start) {
|
|
start = next;
|
|
}
|
|
} else {
|
|
return true;
|
|
}
|
|
}
|
|
|
|
/* we have not found a group, which means everything is made of
|
|
extended names. */
|
|
if (nameChoice == U_EXTENDED_CHAR_NAME) {
|
|
if (limit > UCHAR_MAX_VALUE + 1) {
|
|
limit = UCHAR_MAX_VALUE + 1;
|
|
}
|
|
return enumExtNames(start, limit - 1, fn, context);
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static uint16_t
|
|
writeFactorSuffix(const uint16_t *factors, uint16_t count,
|
|
const char *s, /* suffix elements */
|
|
uint32_t code,
|
|
uint16_t indexes[8], /* output fields from here */
|
|
const char *elementBases[8], const char *elements[8],
|
|
char *buffer, uint16_t bufferLength) {
|
|
uint16_t i, factor, bufferPos=0;
|
|
char c;
|
|
|
|
/* write elements according to the factors */
|
|
|
|
/*
|
|
* the factorized elements are determined by modulo arithmetic
|
|
* with the factors of this algorithm
|
|
*
|
|
* note that for fewer operations, count is decremented here
|
|
*/
|
|
--count;
|
|
for(i=count; i>0; --i) {
|
|
factor=factors[i];
|
|
indexes[i]=(uint16_t)(code%factor);
|
|
code/=factor;
|
|
}
|
|
/*
|
|
* we don't need to calculate the last modulus because start<=code<=end
|
|
* guarantees here that code<=factors[0]
|
|
*/
|
|
indexes[0]=(uint16_t)code;
|
|
|
|
/* write each element */
|
|
for(;;) {
|
|
if(elementBases!=NULL) {
|
|
*elementBases++=s;
|
|
}
|
|
|
|
/* skip indexes[i] strings */
|
|
factor=indexes[i];
|
|
while(factor>0) {
|
|
while(*s++!=0) {}
|
|
--factor;
|
|
}
|
|
if(elements!=NULL) {
|
|
*elements++=s;
|
|
}
|
|
|
|
/* write element */
|
|
while((c=*s++)!=0) {
|
|
WRITE_CHAR(buffer, bufferLength, bufferPos, c);
|
|
}
|
|
|
|
/* we do not need to perform the rest of this loop for i==count - break here */
|
|
if(i>=count) {
|
|
break;
|
|
}
|
|
|
|
/* skip the rest of the strings for this factors[i] */
|
|
factor=(uint16_t)(factors[i]-indexes[i]-1);
|
|
while(factor>0) {
|
|
while(*s++!=0) {}
|
|
--factor;
|
|
}
|
|
|
|
++i;
|
|
}
|
|
|
|
/* zero-terminate */
|
|
if(bufferLength>0) {
|
|
*buffer=0;
|
|
}
|
|
|
|
return bufferPos;
|
|
}
|
|
|
|
/*
|
|
* Important:
|
|
* Parts of findAlgName() are almost the same as some of getAlgName().
|
|
* Fixes must be applied to both.
|
|
*/
|
|
static uint16_t
|
|
getAlgName(AlgorithmicRange *range, uint32_t code, UCharNameChoice nameChoice,
|
|
char *buffer, uint16_t bufferLength) {
|
|
uint16_t bufferPos=0;
|
|
|
|
/* Only the normative character name can be algorithmic. */
|
|
if(nameChoice!=U_UNICODE_CHAR_NAME && nameChoice!=U_EXTENDED_CHAR_NAME) {
|
|
/* zero-terminate */
|
|
if(bufferLength>0) {
|
|
*buffer=0;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
switch(range->type) {
|
|
case 0: {
|
|
/* name = prefix hex-digits */
|
|
const char *s=(const char *)(range+1);
|
|
char c;
|
|
|
|
uint16_t i, count;
|
|
|
|
/* copy prefix */
|
|
while((c=*s++)!=0) {
|
|
WRITE_CHAR(buffer, bufferLength, bufferPos, c);
|
|
}
|
|
|
|
/* write hexadecimal code point value */
|
|
count=range->variant;
|
|
|
|
/* zero-terminate */
|
|
if(count<bufferLength) {
|
|
buffer[count]=0;
|
|
}
|
|
|
|
for(i=count; i>0;) {
|
|
if(--i<bufferLength) {
|
|
c=(char)(code&0xf);
|
|
if(c<10) {
|
|
c+='0';
|
|
} else {
|
|
c+='A'-10;
|
|
}
|
|
buffer[i]=c;
|
|
}
|
|
code>>=4;
|
|
}
|
|
|
|
bufferPos+=count;
|
|
break;
|
|
}
|
|
case 1: {
|
|
/* name = prefix factorized-elements */
|
|
uint16_t indexes[8];
|
|
const uint16_t *factors=(const uint16_t *)(range+1);
|
|
uint16_t count=range->variant;
|
|
const char *s=(const char *)(factors+count);
|
|
char c;
|
|
|
|
/* copy prefix */
|
|
while((c=*s++)!=0) {
|
|
WRITE_CHAR(buffer, bufferLength, bufferPos, c);
|
|
}
|
|
|
|
bufferPos+=writeFactorSuffix(factors, count,
|
|
s, code-range->start, indexes, NULL, NULL, buffer, bufferLength);
|
|
break;
|
|
}
|
|
default:
|
|
/* undefined type */
|
|
/* zero-terminate */
|
|
if(bufferLength>0) {
|
|
*buffer=0;
|
|
}
|
|
break;
|
|
}
|
|
|
|
return bufferPos;
|
|
}
|
|
|
|
/*
|
|
* Important: enumAlgNames() and findAlgName() are almost the same.
|
|
* Any fix must be applied to both.
|
|
*/
|
|
static UBool
|
|
enumAlgNames(AlgorithmicRange *range,
|
|
UChar32 start, UChar32 limit,
|
|
UEnumCharNamesFn *fn, void *context,
|
|
UCharNameChoice nameChoice) {
|
|
char buffer[200];
|
|
uint16_t length;
|
|
|
|
if(nameChoice!=U_UNICODE_CHAR_NAME && nameChoice!=U_EXTENDED_CHAR_NAME) {
|
|
return true;
|
|
}
|
|
|
|
switch(range->type) {
|
|
case 0: {
|
|
char *s, *end;
|
|
char c;
|
|
|
|
/* get the full name of the start character */
|
|
length=getAlgName(range, (uint32_t)start, nameChoice, buffer, sizeof(buffer));
|
|
if(length<=0) {
|
|
return true;
|
|
}
|
|
|
|
/* call the enumerator function with this first character */
|
|
if(!fn(context, start, nameChoice, buffer, length)) {
|
|
return false;
|
|
}
|
|
|
|
/* go to the end of the name; all these names have the same length */
|
|
end=buffer;
|
|
while(*end!=0) {
|
|
++end;
|
|
}
|
|
|
|
/* enumerate the rest of the names */
|
|
while(++start<limit) {
|
|
/* increment the hexadecimal number on a character-basis */
|
|
s=end;
|
|
for (;;) {
|
|
c=*--s;
|
|
if(('0'<=c && c<'9') || ('A'<=c && c<'F')) {
|
|
*s=(char)(c+1);
|
|
break;
|
|
} else if(c=='9') {
|
|
*s='A';
|
|
break;
|
|
} else if(c=='F') {
|
|
*s='0';
|
|
}
|
|
}
|
|
|
|
if(!fn(context, start, nameChoice, buffer, length)) {
|
|
return false;
|
|
}
|
|
}
|
|
break;
|
|
}
|
|
case 1: {
|
|
uint16_t indexes[8];
|
|
const char *elementBases[8], *elements[8];
|
|
const uint16_t *factors=(const uint16_t *)(range+1);
|
|
uint16_t count=range->variant;
|
|
const char *s=(const char *)(factors+count);
|
|
char *suffix, *t;
|
|
uint16_t prefixLength, i, idx;
|
|
|
|
char c;
|
|
|
|
/* name = prefix factorized-elements */
|
|
|
|
/* copy prefix */
|
|
suffix=buffer;
|
|
prefixLength=0;
|
|
while((c=*s++)!=0) {
|
|
*suffix++=c;
|
|
++prefixLength;
|
|
}
|
|
|
|
/* append the suffix of the start character */
|
|
length=(uint16_t)(prefixLength+writeFactorSuffix(factors, count,
|
|
s, (uint32_t)start-range->start,
|
|
indexes, elementBases, elements,
|
|
suffix, (uint16_t)(sizeof(buffer)-prefixLength)));
|
|
|
|
/* call the enumerator function with this first character */
|
|
if(!fn(context, start, nameChoice, buffer, length)) {
|
|
return false;
|
|
}
|
|
|
|
/* enumerate the rest of the names */
|
|
while(++start<limit) {
|
|
/* increment the indexes in lexical order bound by the factors */
|
|
i=count;
|
|
for (;;) {
|
|
idx=(uint16_t)(indexes[--i]+1);
|
|
if(idx<factors[i]) {
|
|
/* skip one index and its element string */
|
|
indexes[i]=idx;
|
|
s=elements[i];
|
|
while(*s++!=0) {
|
|
}
|
|
elements[i]=s;
|
|
break;
|
|
} else {
|
|
/* reset this index to 0 and its element string to the first one */
|
|
indexes[i]=0;
|
|
elements[i]=elementBases[i];
|
|
}
|
|
}
|
|
|
|
/* to make matters a little easier, just append all elements to the suffix */
|
|
t=suffix;
|
|
length=prefixLength;
|
|
for(i=0; i<count; ++i) {
|
|
s=elements[i];
|
|
while((c=*s++)!=0) {
|
|
*t++=c;
|
|
++length;
|
|
}
|
|
}
|
|
/* zero-terminate */
|
|
*t=0;
|
|
|
|
if(!fn(context, start, nameChoice, buffer, length)) {
|
|
return false;
|
|
}
|
|
}
|
|
break;
|
|
}
|
|
default:
|
|
/* undefined type */
|
|
break;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
/*
|
|
* findAlgName() is almost the same as enumAlgNames() except that it
|
|
* returns the code point for a name if it fits into the range.
|
|
* It returns 0xffff otherwise.
|
|
*/
|
|
static UChar32
|
|
findAlgName(AlgorithmicRange *range, UCharNameChoice nameChoice, const char *otherName) {
|
|
UChar32 code;
|
|
|
|
if(nameChoice!=U_UNICODE_CHAR_NAME && nameChoice!=U_EXTENDED_CHAR_NAME) {
|
|
return 0xffff;
|
|
}
|
|
|
|
switch(range->type) {
|
|
case 0: {
|
|
/* name = prefix hex-digits */
|
|
const char *s=(const char *)(range+1);
|
|
char c;
|
|
|
|
uint16_t i, count;
|
|
|
|
/* compare prefix */
|
|
while((c=*s++)!=0) {
|
|
if((char)c!=*otherName++) {
|
|
return 0xffff;
|
|
}
|
|
}
|
|
|
|
/* read hexadecimal code point value */
|
|
count=range->variant;
|
|
code=0;
|
|
for(i=0; i<count; ++i) {
|
|
c=*otherName++;
|
|
if('0'<=c && c<='9') {
|
|
code=(code<<4)|(c-'0');
|
|
} else if('A'<=c && c<='F') {
|
|
code=(code<<4)|(c-'A'+10);
|
|
} else {
|
|
return 0xffff;
|
|
}
|
|
}
|
|
|
|
/* does it fit into the range? */
|
|
if(*otherName==0 && range->start<=(uint32_t)code && (uint32_t)code<=range->end) {
|
|
return code;
|
|
}
|
|
break;
|
|
}
|
|
case 1: {
|
|
char buffer[64];
|
|
uint16_t indexes[8];
|
|
const char *elementBases[8], *elements[8];
|
|
const uint16_t *factors=(const uint16_t *)(range+1);
|
|
uint16_t count=range->variant;
|
|
const char *s=(const char *)(factors+count), *t;
|
|
UChar32 start, limit;
|
|
uint16_t i, idx;
|
|
|
|
char c;
|
|
|
|
/* name = prefix factorized-elements */
|
|
|
|
/* compare prefix */
|
|
while((c=*s++)!=0) {
|
|
if((char)c!=*otherName++) {
|
|
return 0xffff;
|
|
}
|
|
}
|
|
|
|
start=(UChar32)range->start;
|
|
limit=(UChar32)(range->end+1);
|
|
|
|
/* initialize the suffix elements for enumeration; indexes should all be set to 0 */
|
|
writeFactorSuffix(factors, count, s, 0,
|
|
indexes, elementBases, elements, buffer, sizeof(buffer));
|
|
|
|
/* compare the first suffix */
|
|
if(0==uprv_strcmp(otherName, buffer)) {
|
|
return start;
|
|
}
|
|
|
|
/* enumerate and compare the rest of the suffixes */
|
|
while(++start<limit) {
|
|
/* increment the indexes in lexical order bound by the factors */
|
|
i=count;
|
|
for (;;) {
|
|
idx=(uint16_t)(indexes[--i]+1);
|
|
if(idx<factors[i]) {
|
|
/* skip one index and its element string */
|
|
indexes[i]=idx;
|
|
s=elements[i];
|
|
while(*s++!=0) {}
|
|
elements[i]=s;
|
|
break;
|
|
} else {
|
|
/* reset this index to 0 and its element string to the first one */
|
|
indexes[i]=0;
|
|
elements[i]=elementBases[i];
|
|
}
|
|
}
|
|
|
|
/* to make matters a little easier, just compare all elements of the suffix */
|
|
t=otherName;
|
|
for(i=0; i<count; ++i) {
|
|
s=elements[i];
|
|
while((c=*s++)!=0) {
|
|
if(c!=*t++) {
|
|
s=""; /* does not match */
|
|
i=99;
|
|
}
|
|
}
|
|
}
|
|
if(i<99 && *t==0) {
|
|
return start;
|
|
}
|
|
}
|
|
break;
|
|
}
|
|
default:
|
|
/* undefined type */
|
|
break;
|
|
}
|
|
|
|
return 0xffff;
|
|
}
|
|
|
|
/* sets of name characters, maximum name lengths ---------------------------- */
|
|
|
|
#define SET_ADD(set, c) ((set)[(uint8_t)c>>5]|=((uint32_t)1<<((uint8_t)c&0x1f)))
|
|
#define SET_CONTAINS(set, c) (((set)[(uint8_t)c>>5]&((uint32_t)1<<((uint8_t)c&0x1f)))!=0)
|
|
|
|
static int32_t
|
|
calcStringSetLength(uint32_t set[8], const char *s) {
|
|
int32_t length=0;
|
|
char c;
|
|
|
|
while((c=*s++)!=0) {
|
|
SET_ADD(set, c);
|
|
++length;
|
|
}
|
|
return length;
|
|
}
|
|
|
|
static int32_t
|
|
calcAlgNameSetsLengths(int32_t maxNameLength) {
|
|
AlgorithmicRange *range;
|
|
uint32_t *p;
|
|
uint32_t rangeCount;
|
|
int32_t length;
|
|
|
|
/* enumerate algorithmic ranges */
|
|
p=(uint32_t *)((uint8_t *)uCharNames+uCharNames->algNamesOffset);
|
|
rangeCount=*p;
|
|
range=(AlgorithmicRange *)(p+1);
|
|
while(rangeCount>0) {
|
|
switch(range->type) {
|
|
case 0:
|
|
/* name = prefix + (range->variant times) hex-digits */
|
|
/* prefix */
|
|
length=calcStringSetLength(gNameSet, (const char *)(range+1))+range->variant;
|
|
if(length>maxNameLength) {
|
|
maxNameLength=length;
|
|
}
|
|
break;
|
|
case 1: {
|
|
/* name = prefix factorized-elements */
|
|
const uint16_t *factors=(const uint16_t *)(range+1);
|
|
const char *s;
|
|
int32_t i, count=range->variant, factor, factorLength, maxFactorLength;
|
|
|
|
/* prefix length */
|
|
s=(const char *)(factors+count);
|
|
length=calcStringSetLength(gNameSet, s);
|
|
s+=length+1; /* start of factor suffixes */
|
|
|
|
/* get the set and maximum factor suffix length for each factor */
|
|
for(i=0; i<count; ++i) {
|
|
maxFactorLength=0;
|
|
for(factor=factors[i]; factor>0; --factor) {
|
|
factorLength=calcStringSetLength(gNameSet, s);
|
|
s+=factorLength+1;
|
|
if(factorLength>maxFactorLength) {
|
|
maxFactorLength=factorLength;
|
|
}
|
|
}
|
|
length+=maxFactorLength;
|
|
}
|
|
|
|
if(length>maxNameLength) {
|
|
maxNameLength=length;
|
|
}
|
|
break;
|
|
}
|
|
default:
|
|
/* unknown type */
|
|
break;
|
|
}
|
|
|
|
range=(AlgorithmicRange *)((uint8_t *)range+range->size);
|
|
--rangeCount;
|
|
}
|
|
return maxNameLength;
|
|
}
|
|
|
|
static int32_t
|
|
calcExtNameSetsLengths(int32_t maxNameLength) {
|
|
int32_t i, length;
|
|
|
|
for(i=0; i<UPRV_LENGTHOF(charCatNames); ++i) {
|
|
/*
|
|
* for each category, count the length of the category name
|
|
* plus 9=
|
|
* 2 for <>
|
|
* 1 for -
|
|
* 6 for most hex digits per code point
|
|
*/
|
|
length=9+calcStringSetLength(gNameSet, charCatNames[i]);
|
|
if(length>maxNameLength) {
|
|
maxNameLength=length;
|
|
}
|
|
}
|
|
return maxNameLength;
|
|
}
|
|
|
|
static int32_t
|
|
calcNameSetLength(const uint16_t *tokens, uint16_t tokenCount, const uint8_t *tokenStrings, int8_t *tokenLengths,
|
|
uint32_t set[8],
|
|
const uint8_t **pLine, const uint8_t *lineLimit) {
|
|
const uint8_t *line=*pLine;
|
|
int32_t length=0, tokenLength;
|
|
uint16_t c, token;
|
|
|
|
while(line!=lineLimit && (c=*line++)!=(uint8_t)';') {
|
|
if(c>=tokenCount) {
|
|
/* implicit letter */
|
|
SET_ADD(set, c);
|
|
++length;
|
|
} else {
|
|
token=tokens[c];
|
|
if(token==(uint16_t)(-2)) {
|
|
/* this is a lead byte for a double-byte token */
|
|
c=c<<8|*line++;
|
|
token=tokens[c];
|
|
}
|
|
if(token==(uint16_t)(-1)) {
|
|
/* explicit letter */
|
|
SET_ADD(set, c);
|
|
++length;
|
|
} else {
|
|
/* count token word */
|
|
if(tokenLengths!=NULL) {
|
|
/* use cached token length */
|
|
tokenLength=tokenLengths[c];
|
|
if(tokenLength==0) {
|
|
tokenLength=calcStringSetLength(set, (const char *)tokenStrings+token);
|
|
tokenLengths[c]=(int8_t)tokenLength;
|
|
}
|
|
} else {
|
|
tokenLength=calcStringSetLength(set, (const char *)tokenStrings+token);
|
|
}
|
|
length+=tokenLength;
|
|
}
|
|
}
|
|
}
|
|
|
|
*pLine=line;
|
|
return length;
|
|
}
|
|
|
|
static void
|
|
calcGroupNameSetsLengths(int32_t maxNameLength) {
|
|
uint16_t offsets[LINES_PER_GROUP+2], lengths[LINES_PER_GROUP+2];
|
|
|
|
uint16_t *tokens=(uint16_t *)uCharNames+8;
|
|
uint16_t tokenCount=*tokens++;
|
|
uint8_t *tokenStrings=(uint8_t *)uCharNames+uCharNames->tokenStringOffset;
|
|
|
|
int8_t *tokenLengths;
|
|
|
|
const uint16_t *group;
|
|
const uint8_t *s, *line, *lineLimit;
|
|
|
|
int32_t groupCount, lineNumber, length;
|
|
|
|
tokenLengths=(int8_t *)uprv_malloc(tokenCount);
|
|
if(tokenLengths!=NULL) {
|
|
uprv_memset(tokenLengths, 0, tokenCount);
|
|
}
|
|
|
|
group=GET_GROUPS(uCharNames);
|
|
groupCount=*group++;
|
|
|
|
/* enumerate all groups */
|
|
while(groupCount>0) {
|
|
s=(uint8_t *)uCharNames+uCharNames->groupStringOffset+GET_GROUP_OFFSET(group);
|
|
s=expandGroupLengths(s, offsets, lengths);
|
|
|
|
/* enumerate all lines in each group */
|
|
for(lineNumber=0; lineNumber<LINES_PER_GROUP; ++lineNumber) {
|
|
line=s+offsets[lineNumber];
|
|
length=lengths[lineNumber];
|
|
if(length==0) {
|
|
continue;
|
|
}
|
|
|
|
lineLimit=line+length;
|
|
|
|
/* read regular name */
|
|
length=calcNameSetLength(tokens, tokenCount, tokenStrings, tokenLengths, gNameSet, &line, lineLimit);
|
|
if(length>maxNameLength) {
|
|
maxNameLength=length;
|
|
}
|
|
if(line==lineLimit) {
|
|
continue;
|
|
}
|
|
|
|
/* read Unicode 1.0 name */
|
|
length=calcNameSetLength(tokens, tokenCount, tokenStrings, tokenLengths, gNameSet, &line, lineLimit);
|
|
if(length>maxNameLength) {
|
|
maxNameLength=length;
|
|
}
|
|
if(line==lineLimit) {
|
|
continue;
|
|
}
|
|
|
|
/* read ISO comment */
|
|
/*length=calcNameSetLength(tokens, tokenCount, tokenStrings, tokenLengths, gISOCommentSet, &line, lineLimit);*/
|
|
}
|
|
|
|
group=NEXT_GROUP(group);
|
|
--groupCount;
|
|
}
|
|
|
|
if(tokenLengths!=NULL) {
|
|
uprv_free(tokenLengths);
|
|
}
|
|
|
|
/* set gMax... - name length last for threading */
|
|
gMaxNameLength=maxNameLength;
|
|
}
|
|
|
|
static UBool
|
|
calcNameSetsLengths(UErrorCode *pErrorCode) {
|
|
static const char extChars[]="0123456789ABCDEF<>-";
|
|
int32_t i, maxNameLength;
|
|
|
|
if(gMaxNameLength!=0) {
|
|
return true;
|
|
}
|
|
|
|
if(!isDataLoaded(pErrorCode)) {
|
|
return false;
|
|
}
|
|
|
|
/* set hex digits, used in various names, and <>-, used in extended names */
|
|
for(i=0; i<(int32_t)sizeof(extChars)-1; ++i) {
|
|
SET_ADD(gNameSet, extChars[i]);
|
|
}
|
|
|
|
/* set sets and lengths from algorithmic names */
|
|
maxNameLength=calcAlgNameSetsLengths(0);
|
|
|
|
/* set sets and lengths from extended names */
|
|
maxNameLength=calcExtNameSetsLengths(maxNameLength);
|
|
|
|
/* set sets and lengths from group names, set global maximum values */
|
|
calcGroupNameSetsLengths(maxNameLength);
|
|
|
|
return true;
|
|
}
|
|
|
|
U_NAMESPACE_END
|
|
|
|
/* public API --------------------------------------------------------------- */
|
|
|
|
U_NAMESPACE_USE
|
|
|
|
U_CAPI int32_t U_EXPORT2
|
|
u_charName(UChar32 code, UCharNameChoice nameChoice,
|
|
char *buffer, int32_t bufferLength,
|
|
UErrorCode *pErrorCode) {
|
|
AlgorithmicRange *algRange;
|
|
uint32_t *p;
|
|
uint32_t i;
|
|
int32_t length;
|
|
|
|
/* check the argument values */
|
|
if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
|
|
return 0;
|
|
} else if(nameChoice>=U_CHAR_NAME_CHOICE_COUNT ||
|
|
bufferLength<0 || (bufferLength>0 && buffer==NULL)
|
|
) {
|
|
*pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
|
|
return 0;
|
|
}
|
|
|
|
if((uint32_t)code>UCHAR_MAX_VALUE || !isDataLoaded(pErrorCode)) {
|
|
return u_terminateChars(buffer, bufferLength, 0, pErrorCode);
|
|
}
|
|
|
|
length=0;
|
|
|
|
/* try algorithmic names first */
|
|
p=(uint32_t *)((uint8_t *)uCharNames+uCharNames->algNamesOffset);
|
|
i=*p;
|
|
algRange=(AlgorithmicRange *)(p+1);
|
|
while(i>0) {
|
|
if(algRange->start<=(uint32_t)code && (uint32_t)code<=algRange->end) {
|
|
length=getAlgName(algRange, (uint32_t)code, nameChoice, buffer, (uint16_t)bufferLength);
|
|
break;
|
|
}
|
|
algRange=(AlgorithmicRange *)((uint8_t *)algRange+algRange->size);
|
|
--i;
|
|
}
|
|
|
|
if(i==0) {
|
|
if (nameChoice == U_EXTENDED_CHAR_NAME) {
|
|
length = getName(uCharNames, (uint32_t )code, U_EXTENDED_CHAR_NAME, buffer, (uint16_t) bufferLength);
|
|
if (!length) {
|
|
/* extended character name */
|
|
length = getExtName((uint32_t) code, buffer, (uint16_t) bufferLength);
|
|
}
|
|
} else {
|
|
/* normal character name */
|
|
length=getName(uCharNames, (uint32_t)code, nameChoice, buffer, (uint16_t)bufferLength);
|
|
}
|
|
}
|
|
|
|
return u_terminateChars(buffer, bufferLength, length, pErrorCode);
|
|
}
|
|
|
|
U_CAPI int32_t U_EXPORT2
|
|
u_getISOComment(UChar32 /*c*/,
|
|
char *dest, int32_t destCapacity,
|
|
UErrorCode *pErrorCode) {
|
|
/* check the argument values */
|
|
if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
|
|
return 0;
|
|
} else if(destCapacity<0 || (destCapacity>0 && dest==NULL)) {
|
|
*pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
|
|
return 0;
|
|
}
|
|
|
|
return u_terminateChars(dest, destCapacity, 0, pErrorCode);
|
|
}
|
|
|
|
U_CAPI UChar32 U_EXPORT2
|
|
u_charFromName(UCharNameChoice nameChoice,
|
|
const char *name,
|
|
UErrorCode *pErrorCode) {
|
|
char upper[120] = {0};
|
|
char lower[120] = {0};
|
|
FindName findName;
|
|
AlgorithmicRange *algRange;
|
|
uint32_t *p;
|
|
uint32_t i;
|
|
UChar32 cp = 0;
|
|
char c0;
|
|
static constexpr UChar32 error = 0xffff; /* Undefined, but use this for backwards compatibility. */
|
|
|
|
if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
|
|
return error;
|
|
}
|
|
|
|
if(nameChoice>=U_CHAR_NAME_CHOICE_COUNT || name==NULL || *name==0) {
|
|
*pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
|
|
return error;
|
|
}
|
|
|
|
if(!isDataLoaded(pErrorCode)) {
|
|
return error;
|
|
}
|
|
|
|
/* construct the uppercase and lowercase of the name first */
|
|
for(i=0; i<sizeof(upper); ++i) {
|
|
if((c0=*name++)!=0) {
|
|
upper[i]=uprv_toupper(c0);
|
|
lower[i]=uprv_tolower(c0);
|
|
} else {
|
|
upper[i]=lower[i]=0;
|
|
break;
|
|
}
|
|
}
|
|
if(i==sizeof(upper)) {
|
|
/* name too long, there is no such character */
|
|
*pErrorCode = U_ILLEGAL_CHAR_FOUND;
|
|
return error;
|
|
}
|
|
// i==strlen(name)==strlen(lower)==strlen(upper)
|
|
|
|
/* try extended names first */
|
|
if (lower[0] == '<') {
|
|
if (nameChoice == U_EXTENDED_CHAR_NAME && lower[--i] == '>') {
|
|
// Parse a string like "<category-HHHH>" where HHHH is a hex code point.
|
|
uint32_t limit = i;
|
|
while (i >= 3 && lower[--i] != '-') {}
|
|
|
|
// There should be 1 to 8 hex digits.
|
|
int32_t hexLength = limit - (i + 1);
|
|
if (i >= 2 && lower[i] == '-' && 1 <= hexLength && hexLength <= 8) {
|
|
uint32_t cIdx;
|
|
|
|
lower[i] = 0;
|
|
|
|
for (++i; i < limit; ++i) {
|
|
if (lower[i] >= '0' && lower[i] <= '9') {
|
|
cp = (cp << 4) + lower[i] - '0';
|
|
} else if (lower[i] >= 'a' && lower[i] <= 'f') {
|
|
cp = (cp << 4) + lower[i] - 'a' + 10;
|
|
} else {
|
|
*pErrorCode = U_ILLEGAL_CHAR_FOUND;
|
|
return error;
|
|
}
|
|
// Prevent signed-integer overflow and out-of-range code points.
|
|
if (cp > UCHAR_MAX_VALUE) {
|
|
*pErrorCode = U_ILLEGAL_CHAR_FOUND;
|
|
return error;
|
|
}
|
|
}
|
|
|
|
/* Now validate the category name.
|
|
We could use a binary search, or a trie, if
|
|
we really wanted to. */
|
|
uint8_t cat = getCharCat(cp);
|
|
for (lower[i] = 0, cIdx = 0; cIdx < UPRV_LENGTHOF(charCatNames); ++cIdx) {
|
|
|
|
if (!uprv_strcmp(lower + 1, charCatNames[cIdx])) {
|
|
if (cat == cIdx) {
|
|
return cp;
|
|
}
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
*pErrorCode = U_ILLEGAL_CHAR_FOUND;
|
|
return error;
|
|
}
|
|
|
|
/* try algorithmic names now */
|
|
p=(uint32_t *)((uint8_t *)uCharNames+uCharNames->algNamesOffset);
|
|
i=*p;
|
|
algRange=(AlgorithmicRange *)(p+1);
|
|
while(i>0) {
|
|
if((cp=findAlgName(algRange, nameChoice, upper))!=0xffff) {
|
|
return cp;
|
|
}
|
|
algRange=(AlgorithmicRange *)((uint8_t *)algRange+algRange->size);
|
|
--i;
|
|
}
|
|
|
|
/* normal character name */
|
|
findName.otherName=upper;
|
|
findName.code=error;
|
|
enumNames(uCharNames, 0, UCHAR_MAX_VALUE + 1, DO_FIND_NAME, &findName, nameChoice);
|
|
if (findName.code == error) {
|
|
*pErrorCode = U_ILLEGAL_CHAR_FOUND;
|
|
}
|
|
return findName.code;
|
|
}
|
|
|
|
U_CAPI void U_EXPORT2
|
|
u_enumCharNames(UChar32 start, UChar32 limit,
|
|
UEnumCharNamesFn *fn,
|
|
void *context,
|
|
UCharNameChoice nameChoice,
|
|
UErrorCode *pErrorCode) {
|
|
AlgorithmicRange *algRange;
|
|
uint32_t *p;
|
|
uint32_t i;
|
|
|
|
if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
|
|
return;
|
|
}
|
|
|
|
if(nameChoice>=U_CHAR_NAME_CHOICE_COUNT || fn==NULL) {
|
|
*pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
|
|
return;
|
|
}
|
|
|
|
if((uint32_t) limit > UCHAR_MAX_VALUE + 1) {
|
|
limit = UCHAR_MAX_VALUE + 1;
|
|
}
|
|
if((uint32_t)start>=(uint32_t)limit) {
|
|
return;
|
|
}
|
|
|
|
if(!isDataLoaded(pErrorCode)) {
|
|
return;
|
|
}
|
|
|
|
/* interleave the data-driven ones with the algorithmic ones */
|
|
/* iterate over all algorithmic ranges; assume that they are in ascending order */
|
|
p=(uint32_t *)((uint8_t *)uCharNames+uCharNames->algNamesOffset);
|
|
i=*p;
|
|
algRange=(AlgorithmicRange *)(p+1);
|
|
while(i>0) {
|
|
/* enumerate the character names before the current algorithmic range */
|
|
/* here: start<limit */
|
|
if((uint32_t)start<algRange->start) {
|
|
if((uint32_t)limit<=algRange->start) {
|
|
enumNames(uCharNames, start, limit, fn, context, nameChoice);
|
|
return;
|
|
}
|
|
if(!enumNames(uCharNames, start, (UChar32)algRange->start, fn, context, nameChoice)) {
|
|
return;
|
|
}
|
|
start=(UChar32)algRange->start;
|
|
}
|
|
/* enumerate the character names in the current algorithmic range */
|
|
/* here: algRange->start<=start<limit */
|
|
if((uint32_t)start<=algRange->end) {
|
|
if((uint32_t)limit<=(algRange->end+1)) {
|
|
enumAlgNames(algRange, start, limit, fn, context, nameChoice);
|
|
return;
|
|
}
|
|
if(!enumAlgNames(algRange, start, (UChar32)algRange->end+1, fn, context, nameChoice)) {
|
|
return;
|
|
}
|
|
start=(UChar32)algRange->end+1;
|
|
}
|
|
/* continue to the next algorithmic range (here: start<limit) */
|
|
algRange=(AlgorithmicRange *)((uint8_t *)algRange+algRange->size);
|
|
--i;
|
|
}
|
|
/* enumerate the character names after the last algorithmic range */
|
|
enumNames(uCharNames, start, limit, fn, context, nameChoice);
|
|
}
|
|
|
|
U_CAPI int32_t U_EXPORT2
|
|
uprv_getMaxCharNameLength() {
|
|
UErrorCode errorCode=U_ZERO_ERROR;
|
|
if(calcNameSetsLengths(&errorCode)) {
|
|
return gMaxNameLength;
|
|
} else {
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Converts the char set cset into a Unicode set uset.
|
|
* @param cset Set of 256 bit flags corresponding to a set of chars.
|
|
* @param uset USet to receive characters. Existing contents are deleted.
|
|
*/
|
|
static void
|
|
charSetToUSet(uint32_t cset[8], const USetAdder *sa) {
|
|
UChar us[256];
|
|
char cs[256];
|
|
|
|
int32_t i, length;
|
|
UErrorCode errorCode;
|
|
|
|
errorCode=U_ZERO_ERROR;
|
|
|
|
if(!calcNameSetsLengths(&errorCode)) {
|
|
return;
|
|
}
|
|
|
|
/* build a char string with all chars that are used in character names */
|
|
length=0;
|
|
for(i=0; i<256; ++i) {
|
|
if(SET_CONTAINS(cset, i)) {
|
|
cs[length++]=(char)i;
|
|
}
|
|
}
|
|
|
|
/* convert the char string to a UChar string */
|
|
u_charsToUChars(cs, us, length);
|
|
|
|
/* add each UChar to the USet */
|
|
for(i=0; i<length; ++i) {
|
|
if(us[i]!=0 || cs[i]==0) { /* non-invariant chars become (UChar)0 */
|
|
sa->add(sa->set, us[i]);
|
|
}
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Fills set with characters that are used in Unicode character names.
|
|
* @param set USet to receive characters.
|
|
*/
|
|
U_CAPI void U_EXPORT2
|
|
uprv_getCharNameCharacters(const USetAdder *sa) {
|
|
charSetToUSet(gNameSet, sa);
|
|
}
|
|
|
|
/* data swapping ------------------------------------------------------------ */
|
|
|
|
/*
|
|
* The token table contains non-negative entries for token bytes,
|
|
* and -1 for bytes that represent themselves in the data file's charset.
|
|
* -2 entries are used for lead bytes.
|
|
*
|
|
* Direct bytes (-1 entries) must be translated from the input charset family
|
|
* to the output charset family.
|
|
* makeTokenMap() writes a permutation mapping for this.
|
|
* Use it once for single-/lead-byte tokens and once more for all trail byte
|
|
* tokens. (';' is an unused trail byte marked with -1.)
|
|
*/
|
|
static void
|
|
makeTokenMap(const UDataSwapper *ds,
|
|
int16_t tokens[], uint16_t tokenCount,
|
|
uint8_t map[256],
|
|
UErrorCode *pErrorCode) {
|
|
UBool usedOutChar[256];
|
|
uint16_t i, j;
|
|
uint8_t c1, c2;
|
|
|
|
if(U_FAILURE(*pErrorCode)) {
|
|
return;
|
|
}
|
|
|
|
if(ds->inCharset==ds->outCharset) {
|
|
/* Same charset family: identity permutation */
|
|
for(i=0; i<256; ++i) {
|
|
map[i]=(uint8_t)i;
|
|
}
|
|
} else {
|
|
uprv_memset(map, 0, 256);
|
|
uprv_memset(usedOutChar, 0, 256);
|
|
|
|
if(tokenCount>256) {
|
|
tokenCount=256;
|
|
}
|
|
|
|
/* set the direct bytes (byte 0 always maps to itself) */
|
|
for(i=1; i<tokenCount; ++i) {
|
|
if(tokens[i]==-1) {
|
|
/* convert the direct byte character */
|
|
c1=(uint8_t)i;
|
|
ds->swapInvChars(ds, &c1, 1, &c2, pErrorCode);
|
|
if(U_FAILURE(*pErrorCode)) {
|
|
udata_printError(ds, "unames/makeTokenMap() finds variant character 0x%02x used (input charset family %d)\n",
|
|
i, ds->inCharset);
|
|
return;
|
|
}
|
|
|
|
/* enter the converted character into the map and mark it used */
|
|
map[c1]=c2;
|
|
usedOutChar[c2]=true;
|
|
}
|
|
}
|
|
|
|
/* set the mappings for the rest of the permutation */
|
|
for(i=j=1; i<tokenCount; ++i) {
|
|
/* set mappings that were not set for direct bytes */
|
|
if(map[i]==0) {
|
|
/* set an output byte value that was not used as an output byte above */
|
|
while(usedOutChar[j]) {
|
|
++j;
|
|
}
|
|
map[i]=(uint8_t)j++;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* leave mappings at tokenCount and above unset if tokenCount<256
|
|
* because they won't be used
|
|
*/
|
|
}
|
|
}
|
|
|
|
U_CAPI int32_t U_EXPORT2
|
|
uchar_swapNames(const UDataSwapper *ds,
|
|
const void *inData, int32_t length, void *outData,
|
|
UErrorCode *pErrorCode) {
|
|
const UDataInfo *pInfo;
|
|
int32_t headerSize;
|
|
|
|
const uint8_t *inBytes;
|
|
uint8_t *outBytes;
|
|
|
|
uint32_t tokenStringOffset, groupsOffset, groupStringOffset, algNamesOffset,
|
|
offset, i, count, stringsCount;
|
|
|
|
const AlgorithmicRange *inRange;
|
|
AlgorithmicRange *outRange;
|
|
|
|
/* udata_swapDataHeader checks the arguments */
|
|
headerSize=udata_swapDataHeader(ds, inData, length, outData, pErrorCode);
|
|
if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
|
|
return 0;
|
|
}
|
|
|
|
/* check data format and format version */
|
|
pInfo=(const UDataInfo *)((const char *)inData+4);
|
|
if(!(
|
|
pInfo->dataFormat[0]==0x75 && /* dataFormat="unam" */
|
|
pInfo->dataFormat[1]==0x6e &&
|
|
pInfo->dataFormat[2]==0x61 &&
|
|
pInfo->dataFormat[3]==0x6d &&
|
|
pInfo->formatVersion[0]==1
|
|
)) {
|
|
udata_printError(ds, "uchar_swapNames(): data format %02x.%02x.%02x.%02x (format version %02x) is not recognized as unames.icu\n",
|
|
pInfo->dataFormat[0], pInfo->dataFormat[1],
|
|
pInfo->dataFormat[2], pInfo->dataFormat[3],
|
|
pInfo->formatVersion[0]);
|
|
*pErrorCode=U_UNSUPPORTED_ERROR;
|
|
return 0;
|
|
}
|
|
|
|
inBytes=(const uint8_t *)inData+headerSize;
|
|
outBytes=(uint8_t *)outData+headerSize;
|
|
if(length<0) {
|
|
algNamesOffset=ds->readUInt32(((const uint32_t *)inBytes)[3]);
|
|
} else {
|
|
length-=headerSize;
|
|
if( length<20 ||
|
|
(uint32_t)length<(algNamesOffset=ds->readUInt32(((const uint32_t *)inBytes)[3]))
|
|
) {
|
|
udata_printError(ds, "uchar_swapNames(): too few bytes (%d after header) for unames.icu\n",
|
|
length);
|
|
*pErrorCode=U_INDEX_OUTOFBOUNDS_ERROR;
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
if(length<0) {
|
|
/* preflighting: iterate through algorithmic ranges */
|
|
offset=algNamesOffset;
|
|
count=ds->readUInt32(*((const uint32_t *)(inBytes+offset)));
|
|
offset+=4;
|
|
|
|
for(i=0; i<count; ++i) {
|
|
inRange=(const AlgorithmicRange *)(inBytes+offset);
|
|
offset+=ds->readUInt16(inRange->size);
|
|
}
|
|
} else {
|
|
/* swap data */
|
|
const uint16_t *p;
|
|
uint16_t *q, *temp;
|
|
|
|
int16_t tokens[512];
|
|
uint16_t tokenCount;
|
|
|
|
uint8_t map[256], trailMap[256];
|
|
|
|
/* copy the data for inaccessible bytes */
|
|
if(inBytes!=outBytes) {
|
|
uprv_memcpy(outBytes, inBytes, length);
|
|
}
|
|
|
|
/* the initial 4 offsets first */
|
|
tokenStringOffset=ds->readUInt32(((const uint32_t *)inBytes)[0]);
|
|
groupsOffset=ds->readUInt32(((const uint32_t *)inBytes)[1]);
|
|
groupStringOffset=ds->readUInt32(((const uint32_t *)inBytes)[2]);
|
|
ds->swapArray32(ds, inBytes, 16, outBytes, pErrorCode);
|
|
|
|
/*
|
|
* now the tokens table
|
|
* it needs to be permutated along with the compressed name strings
|
|
*/
|
|
p=(const uint16_t *)(inBytes+16);
|
|
q=(uint16_t *)(outBytes+16);
|
|
|
|
/* read and swap the tokenCount */
|
|
tokenCount=ds->readUInt16(*p);
|
|
ds->swapArray16(ds, p, 2, q, pErrorCode);
|
|
++p;
|
|
++q;
|
|
|
|
/* read the first 512 tokens and make the token maps */
|
|
if(tokenCount<=512) {
|
|
count=tokenCount;
|
|
} else {
|
|
count=512;
|
|
}
|
|
for(i=0; i<count; ++i) {
|
|
tokens[i]=udata_readInt16(ds, p[i]);
|
|
}
|
|
for(; i<512; ++i) {
|
|
tokens[i]=0; /* fill the rest of the tokens array if tokenCount<512 */
|
|
}
|
|
makeTokenMap(ds, tokens, tokenCount, map, pErrorCode);
|
|
makeTokenMap(ds, tokens+256, (uint16_t)(tokenCount>256 ? tokenCount-256 : 0), trailMap, pErrorCode);
|
|
if(U_FAILURE(*pErrorCode)) {
|
|
return 0;
|
|
}
|
|
|
|
/*
|
|
* swap and permutate the tokens
|
|
* go through a temporary array to support in-place swapping
|
|
*/
|
|
temp=(uint16_t *)uprv_malloc(tokenCount*2);
|
|
if(temp==NULL) {
|
|
udata_printError(ds, "out of memory swapping %u unames.icu tokens\n",
|
|
tokenCount);
|
|
*pErrorCode=U_MEMORY_ALLOCATION_ERROR;
|
|
return 0;
|
|
}
|
|
|
|
/* swap and permutate single-/lead-byte tokens */
|
|
for(i=0; i<tokenCount && i<256; ++i) {
|
|
ds->swapArray16(ds, p+i, 2, temp+map[i], pErrorCode);
|
|
}
|
|
|
|
/* swap and permutate trail-byte tokens */
|
|
for(; i<tokenCount; ++i) {
|
|
ds->swapArray16(ds, p+i, 2, temp+(i&0xffffff00)+trailMap[i&0xff], pErrorCode);
|
|
}
|
|
|
|
/* copy the result into the output and free the temporary array */
|
|
uprv_memcpy(q, temp, tokenCount*2);
|
|
uprv_free(temp);
|
|
|
|
/*
|
|
* swap the token strings but not a possible padding byte after
|
|
* the terminating NUL of the last string
|
|
*/
|
|
udata_swapInvStringBlock(ds, inBytes+tokenStringOffset, (int32_t)(groupsOffset-tokenStringOffset),
|
|
outBytes+tokenStringOffset, pErrorCode);
|
|
if(U_FAILURE(*pErrorCode)) {
|
|
udata_printError(ds, "uchar_swapNames(token strings) failed\n");
|
|
return 0;
|
|
}
|
|
|
|
/* swap the group table */
|
|
count=ds->readUInt16(*((const uint16_t *)(inBytes+groupsOffset)));
|
|
ds->swapArray16(ds, inBytes+groupsOffset, (int32_t)((1+count*3)*2),
|
|
outBytes+groupsOffset, pErrorCode);
|
|
|
|
/*
|
|
* swap the group strings
|
|
* swap the string bytes but not the nibble-encoded string lengths
|
|
*/
|
|
if(ds->inCharset!=ds->outCharset) {
|
|
uint16_t offsets[LINES_PER_GROUP+1], lengths[LINES_PER_GROUP+1];
|
|
|
|
const uint8_t *inStrings, *nextInStrings;
|
|
uint8_t *outStrings;
|
|
|
|
uint8_t c;
|
|
|
|
inStrings=inBytes+groupStringOffset;
|
|
outStrings=outBytes+groupStringOffset;
|
|
|
|
stringsCount=algNamesOffset-groupStringOffset;
|
|
|
|
/* iterate through string groups until only a few padding bytes are left */
|
|
while(stringsCount>32) {
|
|
nextInStrings=expandGroupLengths(inStrings, offsets, lengths);
|
|
|
|
/* move past the length bytes */
|
|
stringsCount-=(uint32_t)(nextInStrings-inStrings);
|
|
outStrings+=nextInStrings-inStrings;
|
|
inStrings=nextInStrings;
|
|
|
|
count=offsets[31]+lengths[31]; /* total number of string bytes in this group */
|
|
stringsCount-=count;
|
|
|
|
/* swap the string bytes using map[] and trailMap[] */
|
|
while(count>0) {
|
|
c=*inStrings++;
|
|
*outStrings++=map[c];
|
|
if(tokens[c]!=-2) {
|
|
--count;
|
|
} else {
|
|
/* token lead byte: swap the trail byte, too */
|
|
*outStrings++=trailMap[*inStrings++];
|
|
count-=2;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
/* swap the algorithmic ranges */
|
|
offset=algNamesOffset;
|
|
count=ds->readUInt32(*((const uint32_t *)(inBytes+offset)));
|
|
ds->swapArray32(ds, inBytes+offset, 4, outBytes+offset, pErrorCode);
|
|
offset+=4;
|
|
|
|
for(i=0; i<count; ++i) {
|
|
if(offset>(uint32_t)length) {
|
|
udata_printError(ds, "uchar_swapNames(): too few bytes (%d after header) for unames.icu algorithmic range %u\n",
|
|
length, i);
|
|
*pErrorCode=U_INDEX_OUTOFBOUNDS_ERROR;
|
|
return 0;
|
|
}
|
|
|
|
inRange=(const AlgorithmicRange *)(inBytes+offset);
|
|
outRange=(AlgorithmicRange *)(outBytes+offset);
|
|
offset+=ds->readUInt16(inRange->size);
|
|
|
|
ds->swapArray32(ds, inRange, 8, outRange, pErrorCode);
|
|
ds->swapArray16(ds, &inRange->size, 2, &outRange->size, pErrorCode);
|
|
switch(inRange->type) {
|
|
case 0:
|
|
/* swap prefix string */
|
|
ds->swapInvChars(ds, inRange+1, (int32_t)uprv_strlen((const char *)(inRange+1)),
|
|
outRange+1, pErrorCode);
|
|
if(U_FAILURE(*pErrorCode)) {
|
|
udata_printError(ds, "uchar_swapNames(prefix string of algorithmic range %u) failed\n",
|
|
i);
|
|
return 0;
|
|
}
|
|
break;
|
|
case 1:
|
|
{
|
|
/* swap factors and the prefix and factor strings */
|
|
uint32_t factorsCount;
|
|
|
|
factorsCount=inRange->variant;
|
|
p=(const uint16_t *)(inRange+1);
|
|
q=(uint16_t *)(outRange+1);
|
|
ds->swapArray16(ds, p, (int32_t)(factorsCount*2), q, pErrorCode);
|
|
|
|
/* swap the strings, up to the last terminating NUL */
|
|
p+=factorsCount;
|
|
q+=factorsCount;
|
|
stringsCount=(uint32_t)((inBytes+offset)-(const uint8_t *)p);
|
|
while(stringsCount>0 && ((const uint8_t *)p)[stringsCount-1]!=0) {
|
|
--stringsCount;
|
|
}
|
|
ds->swapInvChars(ds, p, (int32_t)stringsCount, q, pErrorCode);
|
|
}
|
|
break;
|
|
default:
|
|
udata_printError(ds, "uchar_swapNames(): unknown type %u of algorithmic range %u\n",
|
|
inRange->type, i);
|
|
*pErrorCode=U_UNSUPPORTED_ERROR;
|
|
return 0;
|
|
}
|
|
}
|
|
}
|
|
|
|
return headerSize+(int32_t)offset;
|
|
}
|
|
|
|
/*
|
|
* Hey, Emacs, please set the following:
|
|
*
|
|
* Local Variables:
|
|
* indent-tabs-mode: nil
|
|
* End:
|
|
*
|
|
*/
|