61607c2773
X-SVN-Rev: 38848
532 lines
17 KiB
C++
532 lines
17 KiB
C++
// Copyright (C) 2016 and later: Unicode, Inc. and others.
|
|
// License & terms of use: http://www.unicode.org/copyright.html
|
|
/*
|
|
*******************************************************************************
|
|
* Copyright (C) 2012-2014, International Business Machines
|
|
* Corporation and others. All Rights Reserved.
|
|
*******************************************************************************
|
|
* utf8collationiterator.cpp
|
|
*
|
|
* created on: 2012nov12 (from utf16collationiterator.cpp & uitercollationiterator.cpp)
|
|
* created by: Markus W. Scherer
|
|
*/
|
|
|
|
#include "unicode/utypes.h"
|
|
|
|
#if !UCONFIG_NO_COLLATION
|
|
|
|
#include "unicode/utf8.h"
|
|
#include "charstr.h"
|
|
#include "cmemory.h"
|
|
#include "collation.h"
|
|
#include "collationdata.h"
|
|
#include "collationfcd.h"
|
|
#include "collationiterator.h"
|
|
#include "normalizer2impl.h"
|
|
#include "uassert.h"
|
|
#include "utf8collationiterator.h"
|
|
|
|
U_NAMESPACE_BEGIN
|
|
|
|
UTF8CollationIterator::~UTF8CollationIterator() {}
|
|
|
|
void
|
|
UTF8CollationIterator::resetToOffset(int32_t newOffset) {
|
|
reset();
|
|
pos = newOffset;
|
|
}
|
|
|
|
int32_t
|
|
UTF8CollationIterator::getOffset() const {
|
|
return pos;
|
|
}
|
|
|
|
uint32_t
|
|
UTF8CollationIterator::handleNextCE32(UChar32 &c, UErrorCode & /*errorCode*/) {
|
|
if(pos == length) {
|
|
c = U_SENTINEL;
|
|
return Collation::FALLBACK_CE32;
|
|
}
|
|
// Optimized combination of U8_NEXT_OR_FFFD() and UTRIE2_U8_NEXT32().
|
|
c = u8[pos++];
|
|
if(c < 0xc0) {
|
|
// ASCII 00..7F; trail bytes 80..BF map to error values.
|
|
return trie->data32[c];
|
|
}
|
|
uint8_t t1, t2;
|
|
if(c < 0xe0 && pos != length && (t1 = (u8[pos] - 0x80)) <= 0x3f) {
|
|
// U+0080..U+07FF; 00..7F map to error values.
|
|
uint32_t ce32 = trie->data32[trie->index[(UTRIE2_UTF8_2B_INDEX_2_OFFSET - 0xc0) + c] + t1];
|
|
c = ((c & 0x1f) << 6) | t1;
|
|
++pos;
|
|
return ce32;
|
|
} else if(c <= 0xef &&
|
|
((pos + 1) < length || length < 0) &&
|
|
(t1 = (u8[pos] - 0x80)) <= 0x3f && (c != 0xe0 || t1 >= 0x20) &&
|
|
(t2 = (u8[pos + 1] - 0x80)) <= 0x3f
|
|
) {
|
|
// U+0800..U+FFFF; caller maps surrogates to error values.
|
|
c = (UChar)((c << 12) | (t1 << 6) | t2);
|
|
pos += 2;
|
|
return UTRIE2_GET32_FROM_U16_SINGLE_LEAD(trie, c);
|
|
} else {
|
|
// Function call for supplementary code points and error cases.
|
|
// Illegal byte sequences yield U+FFFD.
|
|
c = utf8_nextCharSafeBody(u8, &pos, length, c, -3);
|
|
return data->getCE32(c);
|
|
}
|
|
}
|
|
|
|
UBool
|
|
UTF8CollationIterator::foundNULTerminator() {
|
|
if(length < 0) {
|
|
length = --pos;
|
|
return TRUE;
|
|
} else {
|
|
return FALSE;
|
|
}
|
|
}
|
|
|
|
UBool
|
|
UTF8CollationIterator::forbidSurrogateCodePoints() const {
|
|
return TRUE;
|
|
}
|
|
|
|
UChar32
|
|
UTF8CollationIterator::nextCodePoint(UErrorCode & /*errorCode*/) {
|
|
if(pos == length) {
|
|
return U_SENTINEL;
|
|
}
|
|
if(u8[pos] == 0 && length < 0) {
|
|
length = pos;
|
|
return U_SENTINEL;
|
|
}
|
|
UChar32 c;
|
|
U8_NEXT_OR_FFFD(u8, pos, length, c);
|
|
return c;
|
|
}
|
|
|
|
UChar32
|
|
UTF8CollationIterator::previousCodePoint(UErrorCode & /*errorCode*/) {
|
|
if(pos == 0) {
|
|
return U_SENTINEL;
|
|
}
|
|
UChar32 c;
|
|
U8_PREV_OR_FFFD(u8, 0, pos, c);
|
|
return c;
|
|
}
|
|
|
|
void
|
|
UTF8CollationIterator::forwardNumCodePoints(int32_t num, UErrorCode & /*errorCode*/) {
|
|
U8_FWD_N(u8, pos, length, num);
|
|
}
|
|
|
|
void
|
|
UTF8CollationIterator::backwardNumCodePoints(int32_t num, UErrorCode & /*errorCode*/) {
|
|
U8_BACK_N(u8, 0, pos, num);
|
|
}
|
|
|
|
// FCDUTF8CollationIterator ------------------------------------------------ ***
|
|
|
|
FCDUTF8CollationIterator::~FCDUTF8CollationIterator() {}
|
|
|
|
void
|
|
FCDUTF8CollationIterator::resetToOffset(int32_t newOffset) {
|
|
reset();
|
|
start = pos = newOffset;
|
|
state = CHECK_FWD;
|
|
}
|
|
|
|
int32_t
|
|
FCDUTF8CollationIterator::getOffset() const {
|
|
if(state != IN_NORMALIZED) {
|
|
return pos;
|
|
} else if(pos == 0) {
|
|
return start;
|
|
} else {
|
|
return limit;
|
|
}
|
|
}
|
|
|
|
uint32_t
|
|
FCDUTF8CollationIterator::handleNextCE32(UChar32 &c, UErrorCode &errorCode) {
|
|
for(;;) {
|
|
if(state == CHECK_FWD) {
|
|
// Combination of UTF8CollationIterator::handleNextCE32() with FCD check fastpath.
|
|
if(pos == length) {
|
|
c = U_SENTINEL;
|
|
return Collation::FALLBACK_CE32;
|
|
}
|
|
c = u8[pos++];
|
|
if(c < 0xc0) {
|
|
// ASCII 00..7F; trail bytes 80..BF map to error values.
|
|
return trie->data32[c];
|
|
}
|
|
uint8_t t1, t2;
|
|
if(c < 0xe0 && pos != length && (t1 = (u8[pos] - 0x80)) <= 0x3f) {
|
|
// U+0080..U+07FF; 00..7F map to error values.
|
|
uint32_t ce32 = trie->data32[trie->index[(UTRIE2_UTF8_2B_INDEX_2_OFFSET - 0xc0) + c] + t1];
|
|
c = ((c & 0x1f) << 6) | t1;
|
|
++pos;
|
|
if(CollationFCD::hasTccc(c) && pos != length && nextHasLccc()) {
|
|
pos -= 2;
|
|
} else {
|
|
return ce32;
|
|
}
|
|
} else if(c <= 0xef &&
|
|
((pos + 1) < length || length < 0) &&
|
|
(t1 = (u8[pos] - 0x80)) <= 0x3f && (c != 0xe0 || t1 >= 0x20) &&
|
|
(t2 = (u8[pos + 1] - 0x80)) <= 0x3f
|
|
) {
|
|
// U+0800..U+FFFF; caller maps surrogates to error values.
|
|
c = (UChar)((c << 12) | (t1 << 6) | t2);
|
|
pos += 2;
|
|
if(CollationFCD::hasTccc(c) &&
|
|
(CollationFCD::maybeTibetanCompositeVowel(c) ||
|
|
(pos != length && nextHasLccc()))) {
|
|
pos -= 3;
|
|
} else {
|
|
break; // return CE32(BMP)
|
|
}
|
|
} else {
|
|
// Function call for supplementary code points and error cases.
|
|
// Illegal byte sequences yield U+FFFD.
|
|
c = utf8_nextCharSafeBody(u8, &pos, length, c, -3);
|
|
if(c == 0xfffd) {
|
|
return Collation::FFFD_CE32;
|
|
} else {
|
|
U_ASSERT(c > 0xffff);
|
|
if(CollationFCD::hasTccc(U16_LEAD(c)) && pos != length && nextHasLccc()) {
|
|
pos -= 4;
|
|
} else {
|
|
return data->getCE32FromSupplementary(c);
|
|
}
|
|
}
|
|
}
|
|
if(!nextSegment(errorCode)) {
|
|
c = U_SENTINEL;
|
|
return Collation::FALLBACK_CE32;
|
|
}
|
|
continue;
|
|
} else if(state == IN_FCD_SEGMENT && pos != limit) {
|
|
return UTF8CollationIterator::handleNextCE32(c, errorCode);
|
|
} else if(state == IN_NORMALIZED && pos != normalized.length()) {
|
|
c = normalized[pos++];
|
|
break;
|
|
} else {
|
|
switchToForward();
|
|
}
|
|
}
|
|
return UTRIE2_GET32_FROM_U16_SINGLE_LEAD(trie, c);
|
|
}
|
|
|
|
UBool
|
|
FCDUTF8CollationIterator::nextHasLccc() const {
|
|
U_ASSERT(state == CHECK_FWD && pos != length);
|
|
// The lowest code point with ccc!=0 is U+0300 which is CC 80 in UTF-8.
|
|
// CJK U+4000..U+DFFF except U+Axxx are also FCD-inert. (Lead bytes E4..ED except EA.)
|
|
UChar32 c = u8[pos];
|
|
if(c < 0xcc || (0xe4 <= c && c <= 0xed && c != 0xea)) { return FALSE; }
|
|
int32_t i = pos;
|
|
U8_NEXT_OR_FFFD(u8, i, length, c);
|
|
if(c > 0xffff) { c = U16_LEAD(c); }
|
|
return CollationFCD::hasLccc(c);
|
|
}
|
|
|
|
UBool
|
|
FCDUTF8CollationIterator::previousHasTccc() const {
|
|
U_ASSERT(state == CHECK_BWD && pos != 0);
|
|
UChar32 c = u8[pos - 1];
|
|
if(c < 0x80) { return FALSE; }
|
|
int32_t i = pos;
|
|
U8_PREV_OR_FFFD(u8, 0, i, c);
|
|
if(c > 0xffff) { c = U16_LEAD(c); }
|
|
return CollationFCD::hasTccc(c);
|
|
}
|
|
|
|
UChar
|
|
FCDUTF8CollationIterator::handleGetTrailSurrogate() {
|
|
if(state != IN_NORMALIZED) { return 0; }
|
|
U_ASSERT(pos < normalized.length());
|
|
UChar trail;
|
|
if(U16_IS_TRAIL(trail = normalized[pos])) { ++pos; }
|
|
return trail;
|
|
}
|
|
|
|
UBool
|
|
FCDUTF8CollationIterator::foundNULTerminator() {
|
|
if(state == CHECK_FWD && length < 0) {
|
|
length = --pos;
|
|
return TRUE;
|
|
} else {
|
|
return FALSE;
|
|
}
|
|
}
|
|
|
|
UChar32
|
|
FCDUTF8CollationIterator::nextCodePoint(UErrorCode &errorCode) {
|
|
UChar32 c;
|
|
for(;;) {
|
|
if(state == CHECK_FWD) {
|
|
if(pos == length || ((c = u8[pos]) == 0 && length < 0)) {
|
|
return U_SENTINEL;
|
|
}
|
|
if(c < 0x80) {
|
|
++pos;
|
|
return c;
|
|
}
|
|
U8_NEXT_OR_FFFD(u8, pos, length, c);
|
|
if(CollationFCD::hasTccc(c <= 0xffff ? c : U16_LEAD(c)) &&
|
|
(CollationFCD::maybeTibetanCompositeVowel(c) ||
|
|
(pos != length && nextHasLccc()))) {
|
|
// c is not FCD-inert, therefore it is not U+FFFD and it has a valid byte sequence
|
|
// and we can use U8_LENGTH() rather than a previous-position variable.
|
|
pos -= U8_LENGTH(c);
|
|
if(!nextSegment(errorCode)) {
|
|
return U_SENTINEL;
|
|
}
|
|
continue;
|
|
}
|
|
return c;
|
|
} else if(state == IN_FCD_SEGMENT && pos != limit) {
|
|
U8_NEXT_OR_FFFD(u8, pos, length, c);
|
|
return c;
|
|
} else if(state == IN_NORMALIZED && pos != normalized.length()) {
|
|
c = normalized.char32At(pos);
|
|
pos += U16_LENGTH(c);
|
|
return c;
|
|
} else {
|
|
switchToForward();
|
|
}
|
|
}
|
|
}
|
|
|
|
UChar32
|
|
FCDUTF8CollationIterator::previousCodePoint(UErrorCode &errorCode) {
|
|
UChar32 c;
|
|
for(;;) {
|
|
if(state == CHECK_BWD) {
|
|
if(pos == 0) {
|
|
return U_SENTINEL;
|
|
}
|
|
if((c = u8[pos - 1]) < 0x80) {
|
|
--pos;
|
|
return c;
|
|
}
|
|
U8_PREV_OR_FFFD(u8, 0, pos, c);
|
|
if(CollationFCD::hasLccc(c <= 0xffff ? c : U16_LEAD(c)) &&
|
|
(CollationFCD::maybeTibetanCompositeVowel(c) ||
|
|
(pos != 0 && previousHasTccc()))) {
|
|
// c is not FCD-inert, therefore it is not U+FFFD and it has a valid byte sequence
|
|
// and we can use U8_LENGTH() rather than a previous-position variable.
|
|
pos += U8_LENGTH(c);
|
|
if(!previousSegment(errorCode)) {
|
|
return U_SENTINEL;
|
|
}
|
|
continue;
|
|
}
|
|
return c;
|
|
} else if(state == IN_FCD_SEGMENT && pos != start) {
|
|
U8_PREV_OR_FFFD(u8, 0, pos, c);
|
|
return c;
|
|
} else if(state >= IN_NORMALIZED && pos != 0) {
|
|
c = normalized.char32At(pos - 1);
|
|
pos -= U16_LENGTH(c);
|
|
return c;
|
|
} else {
|
|
switchToBackward();
|
|
}
|
|
}
|
|
}
|
|
|
|
void
|
|
FCDUTF8CollationIterator::forwardNumCodePoints(int32_t num, UErrorCode &errorCode) {
|
|
// Specify the class to avoid a virtual-function indirection.
|
|
// In Java, we would declare this class final.
|
|
while(num > 0 && FCDUTF8CollationIterator::nextCodePoint(errorCode) >= 0) {
|
|
--num;
|
|
}
|
|
}
|
|
|
|
void
|
|
FCDUTF8CollationIterator::backwardNumCodePoints(int32_t num, UErrorCode &errorCode) {
|
|
// Specify the class to avoid a virtual-function indirection.
|
|
// In Java, we would declare this class final.
|
|
while(num > 0 && FCDUTF8CollationIterator::previousCodePoint(errorCode) >= 0) {
|
|
--num;
|
|
}
|
|
}
|
|
|
|
void
|
|
FCDUTF8CollationIterator::switchToForward() {
|
|
U_ASSERT(state == CHECK_BWD ||
|
|
(state == IN_FCD_SEGMENT && pos == limit) ||
|
|
(state == IN_NORMALIZED && pos == normalized.length()));
|
|
if(state == CHECK_BWD) {
|
|
// Turn around from backward checking.
|
|
start = pos;
|
|
if(pos == limit) {
|
|
state = CHECK_FWD; // Check forward.
|
|
} else { // pos < limit
|
|
state = IN_FCD_SEGMENT; // Stay in FCD segment.
|
|
}
|
|
} else {
|
|
// Reached the end of the FCD segment.
|
|
if(state == IN_FCD_SEGMENT) {
|
|
// The input text segment is FCD, extend it forward.
|
|
} else {
|
|
// The input text segment needed to be normalized.
|
|
// Switch to checking forward from it.
|
|
start = pos = limit;
|
|
}
|
|
state = CHECK_FWD;
|
|
}
|
|
}
|
|
|
|
UBool
|
|
FCDUTF8CollationIterator::nextSegment(UErrorCode &errorCode) {
|
|
if(U_FAILURE(errorCode)) { return FALSE; }
|
|
U_ASSERT(state == CHECK_FWD && pos != length);
|
|
// The input text [start..pos[ passes the FCD check.
|
|
int32_t segmentStart = pos;
|
|
// Collect the characters being checked, in case they need to be normalized.
|
|
UnicodeString s;
|
|
uint8_t prevCC = 0;
|
|
for(;;) {
|
|
// Fetch the next character and its fcd16 value.
|
|
int32_t cpStart = pos;
|
|
UChar32 c;
|
|
U8_NEXT_OR_FFFD(u8, pos, length, c);
|
|
uint16_t fcd16 = nfcImpl.getFCD16(c);
|
|
uint8_t leadCC = (uint8_t)(fcd16 >> 8);
|
|
if(leadCC == 0 && cpStart != segmentStart) {
|
|
// FCD boundary before this character.
|
|
pos = cpStart;
|
|
break;
|
|
}
|
|
s.append(c);
|
|
if(leadCC != 0 && (prevCC > leadCC || CollationFCD::isFCD16OfTibetanCompositeVowel(fcd16))) {
|
|
// Fails FCD check. Find the next FCD boundary and normalize.
|
|
while(pos != length) {
|
|
cpStart = pos;
|
|
U8_NEXT_OR_FFFD(u8, pos, length, c);
|
|
if(nfcImpl.getFCD16(c) <= 0xff) {
|
|
pos = cpStart;
|
|
break;
|
|
}
|
|
s.append(c);
|
|
}
|
|
if(!normalize(s, errorCode)) { return FALSE; }
|
|
start = segmentStart;
|
|
limit = pos;
|
|
state = IN_NORMALIZED;
|
|
pos = 0;
|
|
return TRUE;
|
|
}
|
|
prevCC = (uint8_t)fcd16;
|
|
if(pos == length || prevCC == 0) {
|
|
// FCD boundary after the last character.
|
|
break;
|
|
}
|
|
}
|
|
limit = pos;
|
|
pos = segmentStart;
|
|
U_ASSERT(pos != limit);
|
|
state = IN_FCD_SEGMENT;
|
|
return TRUE;
|
|
}
|
|
|
|
void
|
|
FCDUTF8CollationIterator::switchToBackward() {
|
|
U_ASSERT(state == CHECK_FWD ||
|
|
(state == IN_FCD_SEGMENT && pos == start) ||
|
|
(state >= IN_NORMALIZED && pos == 0));
|
|
if(state == CHECK_FWD) {
|
|
// Turn around from forward checking.
|
|
limit = pos;
|
|
if(pos == start) {
|
|
state = CHECK_BWD; // Check backward.
|
|
} else { // pos > start
|
|
state = IN_FCD_SEGMENT; // Stay in FCD segment.
|
|
}
|
|
} else {
|
|
// Reached the start of the FCD segment.
|
|
if(state == IN_FCD_SEGMENT) {
|
|
// The input text segment is FCD, extend it backward.
|
|
} else {
|
|
// The input text segment needed to be normalized.
|
|
// Switch to checking backward from it.
|
|
limit = pos = start;
|
|
}
|
|
state = CHECK_BWD;
|
|
}
|
|
}
|
|
|
|
UBool
|
|
FCDUTF8CollationIterator::previousSegment(UErrorCode &errorCode) {
|
|
if(U_FAILURE(errorCode)) { return FALSE; }
|
|
U_ASSERT(state == CHECK_BWD && pos != 0);
|
|
// The input text [pos..limit[ passes the FCD check.
|
|
int32_t segmentLimit = pos;
|
|
// Collect the characters being checked, in case they need to be normalized.
|
|
UnicodeString s;
|
|
uint8_t nextCC = 0;
|
|
for(;;) {
|
|
// Fetch the previous character and its fcd16 value.
|
|
int32_t cpLimit = pos;
|
|
UChar32 c;
|
|
U8_PREV_OR_FFFD(u8, 0, pos, c);
|
|
uint16_t fcd16 = nfcImpl.getFCD16(c);
|
|
uint8_t trailCC = (uint8_t)fcd16;
|
|
if(trailCC == 0 && cpLimit != segmentLimit) {
|
|
// FCD boundary after this character.
|
|
pos = cpLimit;
|
|
break;
|
|
}
|
|
s.append(c);
|
|
if(trailCC != 0 && ((nextCC != 0 && trailCC > nextCC) ||
|
|
CollationFCD::isFCD16OfTibetanCompositeVowel(fcd16))) {
|
|
// Fails FCD check. Find the previous FCD boundary and normalize.
|
|
while(fcd16 > 0xff && pos != 0) {
|
|
cpLimit = pos;
|
|
U8_PREV_OR_FFFD(u8, 0, pos, c);
|
|
fcd16 = nfcImpl.getFCD16(c);
|
|
if(fcd16 == 0) {
|
|
pos = cpLimit;
|
|
break;
|
|
}
|
|
s.append(c);
|
|
}
|
|
s.reverse();
|
|
if(!normalize(s, errorCode)) { return FALSE; }
|
|
limit = segmentLimit;
|
|
start = pos;
|
|
state = IN_NORMALIZED;
|
|
pos = normalized.length();
|
|
return TRUE;
|
|
}
|
|
nextCC = (uint8_t)(fcd16 >> 8);
|
|
if(pos == 0 || nextCC == 0) {
|
|
// FCD boundary before the following character.
|
|
break;
|
|
}
|
|
}
|
|
start = pos;
|
|
pos = segmentLimit;
|
|
U_ASSERT(pos != start);
|
|
state = IN_FCD_SEGMENT;
|
|
return TRUE;
|
|
}
|
|
|
|
UBool
|
|
FCDUTF8CollationIterator::normalize(const UnicodeString &s, UErrorCode &errorCode) {
|
|
// NFD without argument checking.
|
|
U_ASSERT(U_SUCCESS(errorCode));
|
|
nfcImpl.decompose(s, normalized, errorCode);
|
|
return U_SUCCESS(errorCode);
|
|
}
|
|
|
|
U_NAMESPACE_END
|
|
|
|
#endif // !UCONFIG_NO_COLLATION
|