1 /*
2 *******************************************************************************
3 *
4 * Copyright (C) 2009-2010, International Business Machines
5 * Corporation and others. All Rights Reserved.
6 *
7 *******************************************************************************
8 * file name: normalizer2impl.h
9 * encoding: US-ASCII
10 * tab size: 8 (not used)
11 * indentation:4
12 *
13 * created on: 2009nov22
14 * created by: Markus W. Scherer
15 */
16
17 #ifndef __NORMALIZER2IMPL_H__
18 #define __NORMALIZER2IMPL_H__
19
20 #include "unicode/utypes.h"
21
22 #if !UCONFIG_NO_NORMALIZATION
23
24 #include "unicode/normalizer2.h"
25 #include "unicode/udata.h"
26 #include "unicode/unistr.h"
27 #include "unicode/unorm.h"
28 #include "mutex.h"
29 #include "uset_imp.h"
30 #include "utrie2.h"
31
32 U_NAMESPACE_BEGIN
33
34 class CanonIterData;
35
36 class Hangul {
37 public:
38 /* Korean Hangul and Jamo constants */
39 enum {
40 JAMO_L_BASE=0x1100, /* "lead" jamo */
41 JAMO_V_BASE=0x1161, /* "vowel" jamo */
42 JAMO_T_BASE=0x11a7, /* "trail" jamo */
43
44 HANGUL_BASE=0xac00,
45
46 JAMO_L_COUNT=19,
47 JAMO_V_COUNT=21,
48 JAMO_T_COUNT=28,
49
50 JAMO_VT_COUNT=JAMO_V_COUNT*JAMO_T_COUNT,
51
52 HANGUL_COUNT=JAMO_L_COUNT*JAMO_V_COUNT*JAMO_T_COUNT,
53 HANGUL_LIMIT=HANGUL_BASE+HANGUL_COUNT
54 };
55
isHangul(UChar32 c)56 static inline UBool isHangul(UChar32 c) {
57 return HANGUL_BASE<=c && c<HANGUL_LIMIT;
58 }
59 static inline UBool
isHangulWithoutJamoT(UChar c)60 isHangulWithoutJamoT(UChar c) {
61 c-=HANGUL_BASE;
62 return c<HANGUL_COUNT && c%JAMO_T_COUNT==0;
63 }
isJamoL(UChar32 c)64 static inline UBool isJamoL(UChar32 c) {
65 return (uint32_t)(c-JAMO_L_BASE)<JAMO_L_COUNT;
66 }
isJamoV(UChar32 c)67 static inline UBool isJamoV(UChar32 c) {
68 return (uint32_t)(c-JAMO_V_BASE)<JAMO_V_COUNT;
69 }
70
71 /**
72 * Decomposes c, which must be a Hangul syllable, into buffer
73 * and returns the length of the decomposition (2 or 3).
74 */
decompose(UChar32 c,UChar buffer[3])75 static inline int32_t decompose(UChar32 c, UChar buffer[3]) {
76 c-=HANGUL_BASE;
77 UChar32 c2=c%JAMO_T_COUNT;
78 c/=JAMO_T_COUNT;
79 buffer[0]=(UChar)(JAMO_L_BASE+c/JAMO_V_COUNT);
80 buffer[1]=(UChar)(JAMO_V_BASE+c%JAMO_V_COUNT);
81 if(c2==0) {
82 return 2;
83 } else {
84 buffer[2]=(UChar)(JAMO_T_BASE+c2);
85 return 3;
86 }
87 }
88 private:
89 Hangul(); // no instantiation
90 };
91
92 class Normalizer2Impl;
93
94 class ReorderingBuffer : public UMemory {
95 public:
ReorderingBuffer(const Normalizer2Impl & ni,UnicodeString & dest)96 ReorderingBuffer(const Normalizer2Impl &ni, UnicodeString &dest) :
97 impl(ni), str(dest),
98 start(NULL), reorderStart(NULL), limit(NULL),
99 remainingCapacity(0), lastCC(0) {}
~ReorderingBuffer()100 ~ReorderingBuffer() {
101 if(start!=NULL) {
102 str.releaseBuffer((int32_t)(limit-start));
103 }
104 }
105 UBool init(int32_t destCapacity, UErrorCode &errorCode);
106
isEmpty()107 UBool isEmpty() const { return start==limit; }
length()108 int32_t length() const { return (int32_t)(limit-start); }
getStart()109 UChar *getStart() { return start; }
getLimit()110 UChar *getLimit() { return limit; }
getLastCC()111 uint8_t getLastCC() const { return lastCC; }
112
113 UBool equals(const UChar *start, const UChar *limit) const;
114
115 // For Hangul composition, replacing the Leading consonant Jamo with the syllable.
setLastChar(UChar c)116 void setLastChar(UChar c) {
117 *(limit-1)=c;
118 }
119
append(UChar32 c,uint8_t cc,UErrorCode & errorCode)120 UBool append(UChar32 c, uint8_t cc, UErrorCode &errorCode) {
121 return (c<=0xffff) ?
122 appendBMP((UChar)c, cc, errorCode) :
123 appendSupplementary(c, cc, errorCode);
124 }
125 // s must be in NFD, otherwise change the implementation.
126 UBool append(const UChar *s, int32_t length,
127 uint8_t leadCC, uint8_t trailCC,
128 UErrorCode &errorCode);
appendBMP(UChar c,uint8_t cc,UErrorCode & errorCode)129 UBool appendBMP(UChar c, uint8_t cc, UErrorCode &errorCode) {
130 if(remainingCapacity==0 && !resize(1, errorCode)) {
131 return FALSE;
132 }
133 if(lastCC<=cc || cc==0) {
134 *limit++=c;
135 lastCC=cc;
136 if(cc<=1) {
137 reorderStart=limit;
138 }
139 } else {
140 insert(c, cc);
141 }
142 --remainingCapacity;
143 return TRUE;
144 }
145 UBool appendZeroCC(UChar32 c, UErrorCode &errorCode);
146 UBool appendZeroCC(const UChar *s, const UChar *sLimit, UErrorCode &errorCode);
147 void remove();
148 void removeSuffix(int32_t suffixLength);
setReorderingLimit(UChar * newLimit)149 void setReorderingLimit(UChar *newLimit) {
150 remainingCapacity+=(int32_t)(limit-newLimit);
151 reorderStart=limit=newLimit;
152 lastCC=0;
153 }
154 private:
155 /*
156 * TODO: Revisit whether it makes sense to track reorderStart.
157 * It is set to after the last known character with cc<=1,
158 * which stops previousCC() before it reads that character and looks up its cc.
159 * previousCC() is normally only called from insert().
160 * In other words, reorderStart speeds up the insertion of a combining mark
161 * into a multi-combining mark sequence where it does not belong at the end.
162 * This might not be worth the trouble.
163 * On the other hand, it's not a huge amount of trouble.
164 *
165 * We probably need it for UNORM_SIMPLE_APPEND.
166 */
167
168 UBool appendSupplementary(UChar32 c, uint8_t cc, UErrorCode &errorCode);
169 void insert(UChar32 c, uint8_t cc);
writeCodePoint(UChar * p,UChar32 c)170 static void writeCodePoint(UChar *p, UChar32 c) {
171 if(c<=0xffff) {
172 *p=(UChar)c;
173 } else {
174 p[0]=U16_LEAD(c);
175 p[1]=U16_TRAIL(c);
176 }
177 }
178 UBool resize(int32_t appendLength, UErrorCode &errorCode);
179
180 const Normalizer2Impl &impl;
181 UnicodeString &str;
182 UChar *start, *reorderStart, *limit;
183 int32_t remainingCapacity;
184 uint8_t lastCC;
185
186 // private backward iterator
setIterator()187 void setIterator() { codePointStart=limit; }
188 void skipPrevious(); // Requires start<codePointStart.
189 uint8_t previousCC(); // Returns 0 if there is no previous character.
190
191 UChar *codePointStart, *codePointLimit;
192 };
193
194 class U_COMMON_API Normalizer2Impl : public UMemory {
195 public:
Normalizer2Impl()196 Normalizer2Impl() : memory(NULL), normTrie(NULL) {
197 fcdTrieSingleton.fInstance=NULL;
198 canonIterDataSingleton.fInstance=NULL;
199 }
200 ~Normalizer2Impl();
201
202 void load(const char *packageName, const char *name, UErrorCode &errorCode);
203
204 void addPropertyStarts(const USetAdder *sa, UErrorCode &errorCode) const;
205 void addCanonIterPropertyStarts(const USetAdder *sa, UErrorCode &errorCode) const;
206
207 // low-level properties ------------------------------------------------ ***
208
getNormTrie()209 const UTrie2 *getNormTrie() const { return normTrie; }
210 const UTrie2 *getFCDTrie(UErrorCode &errorCode) const ;
211
212 UBool ensureCanonIterData(UErrorCode &errorCode) const;
213
getNorm16(UChar32 c)214 uint16_t getNorm16(UChar32 c) const { return UTRIE2_GET16(normTrie, c); }
215
getCompQuickCheck(uint16_t norm16)216 UNormalizationCheckResult getCompQuickCheck(uint16_t norm16) const {
217 if(norm16<minNoNo || MIN_YES_YES_WITH_CC<=norm16) {
218 return UNORM_YES;
219 } else if(minMaybeYes<=norm16) {
220 return UNORM_MAYBE;
221 } else {
222 return UNORM_NO;
223 }
224 }
isCompNo(uint16_t norm16)225 UBool isCompNo(uint16_t norm16) const { return minNoNo<=norm16 && norm16<minMaybeYes; }
isDecompYes(uint16_t norm16)226 UBool isDecompYes(uint16_t norm16) const { return norm16<minYesNo || minMaybeYes<=norm16; }
227
getCC(uint16_t norm16)228 uint8_t getCC(uint16_t norm16) const {
229 if(norm16>=MIN_NORMAL_MAYBE_YES) {
230 return (uint8_t)norm16;
231 }
232 if(norm16<minNoNo || limitNoNo<=norm16) {
233 return 0;
234 }
235 return getCCFromNoNo(norm16);
236 }
getCCFromYesOrMaybe(uint16_t norm16)237 static uint8_t getCCFromYesOrMaybe(uint16_t norm16) {
238 return norm16>=MIN_NORMAL_MAYBE_YES ? (uint8_t)norm16 : 0;
239 }
240
getFCD16(UChar32 c)241 uint16_t getFCD16(UChar32 c) const { return UTRIE2_GET16(fcdTrie(), c); }
getFCD16FromSingleLead(UChar c)242 uint16_t getFCD16FromSingleLead(UChar c) const {
243 return UTRIE2_GET16_FROM_U16_SINGLE_LEAD(fcdTrie(), c);
244 }
getFCD16FromSupplementary(UChar32 c)245 uint16_t getFCD16FromSupplementary(UChar32 c) const {
246 return UTRIE2_GET16_FROM_SUPP(fcdTrie(), c);
247 }
getFCD16FromSurrogatePair(UChar c,UChar c2)248 uint16_t getFCD16FromSurrogatePair(UChar c, UChar c2) const {
249 return getFCD16FromSupplementary(U16_GET_SUPPLEMENTARY(c, c2));
250 }
251
252 void setFCD16FromNorm16(UChar32 start, UChar32 end, uint16_t norm16,
253 UTrie2 *newFCDTrie, UErrorCode &errorCode) const;
254
255 void makeCanonIterDataFromNorm16(UChar32 start, UChar32 end, uint16_t norm16,
256 CanonIterData &newData, UErrorCode &errorCode) const;
257
258 /**
259 * Get the decomposition for one code point.
260 * @param c code point
261 * @param buffer out-only buffer for algorithmic decompositions
262 * @param length out-only, takes the length of the decomposition, if any
263 * @return pointer to the decomposition, or NULL if none
264 */
265 const UChar *getDecomposition(UChar32 c, UChar buffer[4], int32_t &length) const;
266
267 UBool isCanonSegmentStarter(UChar32 c) const;
268 UBool getCanonStartSet(UChar32 c, UnicodeSet &set) const;
269
270 enum {
271 MIN_CCC_LCCC_CP=0x300
272 };
273
274 enum {
275 MIN_YES_YES_WITH_CC=0xff01,
276 JAMO_VT=0xff00,
277 MIN_NORMAL_MAYBE_YES=0xfe00,
278 JAMO_L=1,
279 MAX_DELTA=0x40
280 };
281
282 enum {
283 // Byte offsets from the start of the data, after the generic header.
284 IX_NORM_TRIE_OFFSET,
285 IX_EXTRA_DATA_OFFSET,
286 IX_RESERVED2_OFFSET,
287 IX_RESERVED3_OFFSET,
288 IX_RESERVED4_OFFSET,
289 IX_RESERVED5_OFFSET,
290 IX_RESERVED6_OFFSET,
291 IX_TOTAL_SIZE,
292
293 // Code point thresholds for quick check codes.
294 IX_MIN_DECOMP_NO_CP,
295 IX_MIN_COMP_NO_MAYBE_CP,
296
297 // Norm16 value thresholds for quick check combinations and types of extra data.
298 IX_MIN_YES_NO,
299 IX_MIN_NO_NO,
300 IX_LIMIT_NO_NO,
301 IX_MIN_MAYBE_YES,
302
303 IX_RESERVED14,
304 IX_RESERVED15,
305 IX_COUNT
306 };
307
308 enum {
309 MAPPING_HAS_CCC_LCCC_WORD=0x80,
310 MAPPING_PLUS_COMPOSITION_LIST=0x40,
311 MAPPING_NO_COMP_BOUNDARY_AFTER=0x20,
312 MAPPING_LENGTH_MASK=0x1f
313 };
314
315 enum {
316 COMP_1_LAST_TUPLE=0x8000,
317 COMP_1_TRIPLE=1,
318 COMP_1_TRAIL_LIMIT=0x3400,
319 COMP_1_TRAIL_MASK=0x7ffe,
320 COMP_1_TRAIL_SHIFT=9, // 10-1 for the "triple" bit
321 COMP_2_TRAIL_SHIFT=6,
322 COMP_2_TRAIL_MASK=0xffc0
323 };
324
325 // higher-level functionality ------------------------------------------ ***
326
327 const UChar *decompose(const UChar *src, const UChar *limit,
328 ReorderingBuffer *buffer, UErrorCode &errorCode) const;
329 void decomposeAndAppend(const UChar *src, const UChar *limit,
330 UBool doDecompose,
331 ReorderingBuffer &buffer,
332 UErrorCode &errorCode) const;
333 UBool compose(const UChar *src, const UChar *limit,
334 UBool onlyContiguous,
335 UBool doCompose,
336 ReorderingBuffer &buffer,
337 UErrorCode &errorCode) const;
338 const UChar *composeQuickCheck(const UChar *src, const UChar *limit,
339 UBool onlyContiguous,
340 UNormalizationCheckResult *pQCResult) const;
341 void composeAndAppend(const UChar *src, const UChar *limit,
342 UBool doCompose,
343 UBool onlyContiguous,
344 ReorderingBuffer &buffer,
345 UErrorCode &errorCode) const;
346 const UChar *makeFCD(const UChar *src, const UChar *limit,
347 ReorderingBuffer *buffer, UErrorCode &errorCode) const;
348 void makeFCDAndAppend(const UChar *src, const UChar *limit,
349 UBool doMakeFCD,
350 ReorderingBuffer &buffer,
351 UErrorCode &errorCode) const;
352
353 UBool hasDecompBoundary(UChar32 c, UBool before) const;
isDecompInert(UChar32 c)354 UBool isDecompInert(UChar32 c) const { return isDecompYesAndZeroCC(getNorm16(c)); }
355
hasCompBoundaryBefore(UChar32 c)356 UBool hasCompBoundaryBefore(UChar32 c) const {
357 return c<minCompNoMaybeCP || hasCompBoundaryBefore(c, getNorm16(c));
358 }
359 UBool hasCompBoundaryAfter(UChar32 c, UBool onlyContiguous, UBool testInert) const;
360
hasFCDBoundaryBefore(UChar32 c)361 UBool hasFCDBoundaryBefore(UChar32 c) const { return c<MIN_CCC_LCCC_CP || getFCD16(c)<=0xff; }
hasFCDBoundaryAfter(UChar32 c)362 UBool hasFCDBoundaryAfter(UChar32 c) const {
363 uint16_t fcd16=getFCD16(c);
364 return fcd16<=1 || (fcd16&0xff)==0;
365 }
isFCDInert(UChar32 c)366 UBool isFCDInert(UChar32 c) const { return getFCD16(c)<=1; }
367 private:
368 static UBool U_CALLCONV
369 isAcceptable(void *context, const char *type, const char *name, const UDataInfo *pInfo);
370
isMaybe(uint16_t norm16)371 UBool isMaybe(uint16_t norm16) const { return minMaybeYes<=norm16 && norm16<=JAMO_VT; }
isMaybeOrNonZeroCC(uint16_t norm16)372 UBool isMaybeOrNonZeroCC(uint16_t norm16) const { return norm16>=minMaybeYes; }
isInert(uint16_t norm16)373 static UBool isInert(uint16_t norm16) { return norm16==0; }
374 // static UBool isJamoL(uint16_t norm16) const { return norm16==1; }
isJamoVT(uint16_t norm16)375 static UBool isJamoVT(uint16_t norm16) { return norm16==JAMO_VT; }
isHangul(uint16_t norm16)376 UBool isHangul(uint16_t norm16) const { return norm16==minYesNo; }
isCompYesAndZeroCC(uint16_t norm16)377 UBool isCompYesAndZeroCC(uint16_t norm16) const { return norm16<minNoNo; }
378 // UBool isCompYes(uint16_t norm16) const {
379 // return norm16>=MIN_YES_YES_WITH_CC || norm16<minNoNo;
380 // }
381 // UBool isCompYesOrMaybe(uint16_t norm16) const {
382 // return norm16<minNoNo || minMaybeYes<=norm16;
383 // }
384 // UBool hasZeroCCFromDecompYes(uint16_t norm16) const {
385 // return norm16<=MIN_NORMAL_MAYBE_YES || norm16==JAMO_VT;
386 // }
isDecompYesAndZeroCC(uint16_t norm16)387 UBool isDecompYesAndZeroCC(uint16_t norm16) const {
388 return norm16<minYesNo ||
389 norm16==JAMO_VT ||
390 (minMaybeYes<=norm16 && norm16<=MIN_NORMAL_MAYBE_YES);
391 }
392 /**
393 * A little faster and simpler than isDecompYesAndZeroCC() but does not include
394 * the MaybeYes which combine-forward and have ccc=0.
395 * (Standard Unicode 5.2 normalization does not have such characters.)
396 */
isMostDecompYesAndZeroCC(uint16_t norm16)397 UBool isMostDecompYesAndZeroCC(uint16_t norm16) const {
398 return norm16<minYesNo || norm16==MIN_NORMAL_MAYBE_YES || norm16==JAMO_VT;
399 }
isDecompNoAlgorithmic(uint16_t norm16)400 UBool isDecompNoAlgorithmic(uint16_t norm16) const { return norm16>=limitNoNo; }
401
402 // For use with isCompYes().
403 // Perhaps the compiler can combine the two tests for MIN_YES_YES_WITH_CC.
404 // static uint8_t getCCFromYes(uint16_t norm16) {
405 // return norm16>=MIN_YES_YES_WITH_CC ? (uint8_t)norm16 : 0;
406 // }
getCCFromNoNo(uint16_t norm16)407 uint8_t getCCFromNoNo(uint16_t norm16) const {
408 const uint16_t *mapping=getMapping(norm16);
409 if(*mapping&MAPPING_HAS_CCC_LCCC_WORD) {
410 return (uint8_t)mapping[1];
411 } else {
412 return 0;
413 }
414 }
415 // requires that the [cpStart..cpLimit[ character passes isCompYesAndZeroCC()
416 uint8_t getTrailCCFromCompYesAndZeroCC(const UChar *cpStart, const UChar *cpLimit) const;
417
418 // Requires algorithmic-NoNo.
mapAlgorithmic(UChar32 c,uint16_t norm16)419 UChar32 mapAlgorithmic(UChar32 c, uint16_t norm16) const {
420 return c+norm16-(minMaybeYes-MAX_DELTA-1);
421 }
422
423 // Requires minYesNo<norm16<limitNoNo.
getMapping(uint16_t norm16)424 const uint16_t *getMapping(uint16_t norm16) const { return extraData+norm16; }
getCompositionsListForDecompYes(uint16_t norm16)425 const uint16_t *getCompositionsListForDecompYes(uint16_t norm16) const {
426 if(norm16==0 || MIN_NORMAL_MAYBE_YES<=norm16) {
427 return NULL;
428 } else if(norm16<minMaybeYes) {
429 return extraData+norm16; // for yesYes; if Jamo L: harmless empty list
430 } else {
431 return maybeYesCompositions+norm16-minMaybeYes;
432 }
433 }
getCompositionsListForComposite(uint16_t norm16)434 const uint16_t *getCompositionsListForComposite(uint16_t norm16) const {
435 const uint16_t *list=extraData+norm16; // composite has both mapping & compositions list
436 return list+ // mapping pointer
437 1+ // +1 to skip the first unit with the mapping lenth
438 (*list&MAPPING_LENGTH_MASK)+ // + mapping length
439 ((*list>>7)&1); // +1 if MAPPING_HAS_CCC_LCCC_WORD
440 }
441 /**
442 * @param c code point must have compositions
443 * @return compositions list pointer
444 */
getCompositionsList(uint16_t norm16)445 const uint16_t *getCompositionsList(uint16_t norm16) const {
446 return isDecompYes(norm16) ?
447 getCompositionsListForDecompYes(norm16) :
448 getCompositionsListForComposite(norm16);
449 }
450
451 const UChar *copyLowPrefixFromNulTerminated(const UChar *src,
452 UChar32 minNeedDataCP,
453 ReorderingBuffer *buffer,
454 UErrorCode &errorCode) const;
455 UBool decomposeShort(const UChar *src, const UChar *limit,
456 ReorderingBuffer &buffer, UErrorCode &errorCode) const;
457 UBool decompose(UChar32 c, uint16_t norm16,
458 ReorderingBuffer &buffer, UErrorCode &errorCode) const;
459
460 static int32_t combine(const uint16_t *list, UChar32 trail);
461 void addComposites(const uint16_t *list, UnicodeSet &set) const;
462 void recompose(ReorderingBuffer &buffer, int32_t recomposeStartIndex,
463 UBool onlyContiguous) const;
464
465 UBool hasCompBoundaryBefore(UChar32 c, uint16_t norm16) const;
466 const UChar *findPreviousCompBoundary(const UChar *start, const UChar *p) const;
467 const UChar *findNextCompBoundary(const UChar *p, const UChar *limit) const;
468
fcdTrie()469 const UTrie2 *fcdTrie() const { return (const UTrie2 *)fcdTrieSingleton.fInstance; }
470
471 const UChar *findPreviousFCDBoundary(const UChar *start, const UChar *p) const;
472 const UChar *findNextFCDBoundary(const UChar *p, const UChar *limit) const;
473
474 int32_t getCanonValue(UChar32 c) const;
475 const UnicodeSet &getCanonStartSet(int32_t n) const;
476
477 UDataMemory *memory;
478 UVersionInfo dataVersion;
479
480 // Code point thresholds for quick check codes.
481 UChar32 minDecompNoCP;
482 UChar32 minCompNoMaybeCP;
483
484 // Norm16 value thresholds for quick check combinations and types of extra data.
485 uint16_t minYesNo;
486 uint16_t minNoNo;
487 uint16_t limitNoNo;
488 uint16_t minMaybeYes;
489
490 UTrie2 *normTrie;
491 const uint16_t *maybeYesCompositions;
492 const uint16_t *extraData; // mappings and/or compositions for yesYes, yesNo & noNo characters
493
494 SimpleSingleton fcdTrieSingleton;
495 SimpleSingleton canonIterDataSingleton;
496 };
497
498 // bits in canonIterData
499 #define CANON_NOT_SEGMENT_STARTER 0x80000000
500 #define CANON_HAS_COMPOSITIONS 0x40000000
501 #define CANON_HAS_SET 0x200000
502 #define CANON_VALUE_MASK 0x1fffff
503
504 /**
505 * ICU-internal shortcut for quick access to standard Unicode normalization.
506 */
507 class U_COMMON_API Normalizer2Factory {
508 public:
509 static const Normalizer2 *getNFCInstance(UErrorCode &errorCode);
510 static const Normalizer2 *getNFDInstance(UErrorCode &errorCode);
511 static const Normalizer2 *getFCDInstance(UErrorCode &errorCode);
512 static const Normalizer2 *getFCCInstance(UErrorCode &errorCode);
513 static const Normalizer2 *getNFKCInstance(UErrorCode &errorCode);
514 static const Normalizer2 *getNFKDInstance(UErrorCode &errorCode);
515 static const Normalizer2 *getNFKC_CFInstance(UErrorCode &errorCode);
516 static const Normalizer2 *getNoopInstance(UErrorCode &errorCode);
517
518 static const Normalizer2 *getInstance(UNormalizationMode mode, UErrorCode &errorCode);
519
520 static const Normalizer2Impl *getNFCImpl(UErrorCode &errorCode);
521 static const Normalizer2Impl *getNFKCImpl(UErrorCode &errorCode);
522 static const Normalizer2Impl *getNFKC_CFImpl(UErrorCode &errorCode);
523
524 // Get the Impl instance of the Normalizer2.
525 // Must be used only when it is known that norm2 is a Normalizer2WithImpl instance.
526 static const Normalizer2Impl *getImpl(const Normalizer2 *norm2);
527
528 static const UTrie2 *getFCDTrie(UErrorCode &errorCode);
529 private:
530 Normalizer2Factory(); // No instantiation.
531 };
532
533 U_NAMESPACE_END
534
535 U_CAPI int32_t U_EXPORT2
536 unorm2_swap(const UDataSwapper *ds,
537 const void *inData, int32_t length, void *outData,
538 UErrorCode *pErrorCode);
539
540 /**
541 * Get the NF*_QC property for a code point, for u_getIntPropertyValue().
542 * @internal
543 */
544 U_CFUNC UNormalizationCheckResult U_EXPORT2
545 unorm_getQuickCheck(UChar32 c, UNormalizationMode mode);
546
547 /**
548 * Internal API, used by collation code.
549 * Get access to the internal FCD trie table to be able to perform
550 * incremental, per-code unit, FCD checks in collation.
551 * One pointer is sufficient because the trie index values are offset
552 * by the index size, so that the same pointer is used to access the trie data.
553 * Code points at fcdHighStart and above have a zero FCD value.
554 * @internal
555 */
556 U_CAPI const uint16_t * U_EXPORT2
557 unorm_getFCDTrieIndex(UChar32 &fcdHighStart, UErrorCode *pErrorCode);
558
559 /**
560 * Internal API, used by collation code.
561 * Get the FCD value for a code unit, with
562 * bits 15..8 lead combining class
563 * bits 7..0 trail combining class
564 *
565 * If c is a lead surrogate and the value is not 0,
566 * then some of c's associated supplementary code points have a non-zero FCD value.
567 *
568 * @internal
569 */
570 static inline uint16_t
unorm_getFCD16(const uint16_t * fcdTrieIndex,UChar c)571 unorm_getFCD16(const uint16_t *fcdTrieIndex, UChar c) {
572 return fcdTrieIndex[_UTRIE2_INDEX_FROM_U16_SINGLE_LEAD(fcdTrieIndex, c)];
573 }
574
575 /**
576 * Internal API, used by collation code.
577 * Get the FCD value of the next code point (post-increment), with
578 * bits 15..8 lead combining class
579 * bits 7..0 trail combining class
580 *
581 * @internal
582 */
583 static inline uint16_t
unorm_nextFCD16(const uint16_t * fcdTrieIndex,UChar32 fcdHighStart,const UChar * & s,const UChar * limit)584 unorm_nextFCD16(const uint16_t *fcdTrieIndex, UChar32 fcdHighStart,
585 const UChar *&s, const UChar *limit) {
586 UChar32 c=*s++;
587 uint16_t fcd=fcdTrieIndex[_UTRIE2_INDEX_FROM_U16_SINGLE_LEAD(fcdTrieIndex, c)];
588 if(fcd!=0 && U16_IS_LEAD(c)) {
589 UChar c2;
590 if(s!=limit && U16_IS_TRAIL(c2=*s)) {
591 ++s;
592 c=U16_GET_SUPPLEMENTARY(c, c2);
593 if(c<fcdHighStart) {
594 fcd=fcdTrieIndex[_UTRIE2_INDEX_FROM_SUPP(fcdTrieIndex, c)];
595 } else {
596 fcd=0;
597 }
598 } else /* unpaired lead surrogate */ {
599 fcd=0;
600 }
601 }
602 return fcd;
603 }
604
605 /**
606 * Internal API, used by collation code.
607 * Get the FCD value of the previous code point (pre-decrement), with
608 * bits 15..8 lead combining class
609 * bits 7..0 trail combining class
610 *
611 * @internal
612 */
613 static inline uint16_t
unorm_prevFCD16(const uint16_t * fcdTrieIndex,UChar32 fcdHighStart,const UChar * start,const UChar * & s)614 unorm_prevFCD16(const uint16_t *fcdTrieIndex, UChar32 fcdHighStart,
615 const UChar *start, const UChar *&s) {
616 UChar32 c=*--s;
617 uint16_t fcd;
618 if(!U16_IS_SURROGATE(c)) {
619 fcd=fcdTrieIndex[_UTRIE2_INDEX_FROM_U16_SINGLE_LEAD(fcdTrieIndex, c)];
620 } else {
621 UChar c2;
622 if(U16_IS_SURROGATE_TRAIL(c) && s!=start && U16_IS_LEAD(c2=*(s-1))) {
623 --s;
624 c=U16_GET_SUPPLEMENTARY(c2, c);
625 if(c<fcdHighStart) {
626 fcd=fcdTrieIndex[_UTRIE2_INDEX_FROM_SUPP(fcdTrieIndex, c)];
627 } else {
628 fcd=0;
629 }
630 } else /* unpaired surrogate */ {
631 fcd=0;
632 }
633 }
634 return fcd;
635 }
636
637 /**
638 * Format of Normalizer2 .nrm data files.
639 * Format version 1.0.
640 *
641 * Normalizer2 .nrm data files provide data for the Unicode Normalization algorithms.
642 * ICU ships with data files for standard Unicode Normalization Forms
643 * NFC and NFD (nfc.nrm), NFKC and NFKD (nfkc.nrm) and NFKC_Casefold (nfkc_cf.nrm).
644 * Custom (application-specific) data can be built into additional .nrm files
645 * with the gennorm2 build tool.
646 *
647 * Normalizer2.getInstance() causes a .nrm file to be loaded, unless it has been
648 * cached already. Internally, Normalizer2Impl.load() reads the .nrm file.
649 *
650 * A .nrm file begins with a standard ICU data file header
651 * (DataHeader, see ucmndata.h and unicode/udata.h).
652 * The UDataInfo.dataVersion field usually contains the Unicode version
653 * for which the data was generated.
654 *
655 * After the header, the file contains the following parts.
656 * Constants are defined as enum values of the Normalizer2Impl class.
657 *
658 * Many details of the data structures are described in the design doc
659 * which is at http://site.icu-project.org/design/normalization/custom
660 *
661 * int32_t indexes[indexesLength]; -- indexesLength=indexes[IX_NORM_TRIE_OFFSET]/4;
662 *
663 * The first eight indexes are byte offsets in ascending order.
664 * Each byte offset marks the start of the next part in the data file,
665 * and the end of the previous one.
666 * When two consecutive byte offsets are the same, then the corresponding part is empty.
667 * Byte offsets are offsets from after the header,
668 * that is, from the beginning of the indexes[].
669 * Each part starts at an offset with proper alignment for its data.
670 * If necessary, the previous part may include padding bytes to achieve this alignment.
671 *
672 * minDecompNoCP=indexes[IX_MIN_DECOMP_NO_CP] is the lowest code point
673 * with a decomposition mapping, that is, with NF*D_QC=No.
674 * minCompNoMaybeCP=indexes[IX_MIN_COMP_NO_MAYBE_CP] is the lowest code point
675 * with NF*C_QC=No (has a one-way mapping) or Maybe (combines backward).
676 *
677 * The next four indexes are thresholds of 16-bit trie values for ranges of
678 * values indicating multiple normalization properties.
679 * minYesNo=indexes[IX_MIN_YES_NO];
680 * minNoNo=indexes[IX_MIN_NO_NO];
681 * limitNoNo=indexes[IX_LIMIT_NO_NO];
682 * minMaybeYes=indexes[IX_MIN_MAYBE_YES];
683 * See the normTrie description below and the design doc for details.
684 *
685 * UTrie2 normTrie; -- see utrie2_impl.h and utrie2.h
686 *
687 * The trie holds the main normalization data. Each code point is mapped to a 16-bit value.
688 * Rather than using independent bits in the value (which would require more than 16 bits),
689 * information is extracted primarily via range checks.
690 * For example, a 16-bit value norm16 in the range minYesNo<=norm16<minNoNo
691 * means that the character has NF*C_QC=Yes and NF*D_QC=No properties,
692 * which means it has a two-way (round-trip) decomposition mapping.
693 * Values in the range 2<=norm16<limitNoNo are also directly indexes into the extraData
694 * pointing to mappings, composition lists, or both.
695 * Value norm16==0 means that the character is normalization-inert, that is,
696 * it does not have a mapping, does not participate in composition, has a zero
697 * canonical combining class, and forms a boundary where text before it and after it
698 * can be normalized independently.
699 * For details about how multiple properties are encoded in 16-bit values
700 * see the design doc.
701 * Note that the encoding cannot express all combinations of the properties involved;
702 * it only supports those combinations that are allowed by
703 * the Unicode Normalization algorithms. Details are in the design doc as well.
704 * The gennorm2 tool only builds .nrm files for data that conforms to the limitations.
705 *
706 * The trie has a value for each lead surrogate code unit representing the "worst case"
707 * properties of the 1024 supplementary characters whose UTF-16 form starts with
708 * the lead surrogate. If all of the 1024 supplementary characters are normalization-inert,
709 * then their lead surrogate code unit has the trie value 0.
710 * When the lead surrogate unit's value exceeds the quick check minimum during processing,
711 * the properties for the full supplementary code point need to be looked up.
712 *
713 * uint16_t maybeYesCompositions[MIN_NORMAL_MAYBE_YES-minMaybeYes];
714 * uint16_t extraData[];
715 *
716 * There is only one byte offset for the end of these two arrays.
717 * The split between them is given by the constant and variable mentioned above.
718 *
719 * The maybeYesCompositions array contains composition lists for characters that
720 * combine both forward (as starters in composition pairs)
721 * and backward (as trailing characters in composition pairs).
722 * Such characters do not occur in Unicode 5.2 but are allowed by
723 * the Unicode Normalization algorithms.
724 * If there are no such characters, then minMaybeYes==MIN_NORMAL_MAYBE_YES
725 * and the maybeYesCompositions array is empty.
726 * If there are such characters, then minMaybeYes is subtracted from their norm16 values
727 * to get the index into this array.
728 *
729 * The extraData array contains composition lists for "YesYes" characters,
730 * followed by mappings and optional composition lists for "YesNo" characters,
731 * followed by only mappings for "NoNo" characters.
732 * (Referring to pairs of NFC/NFD quick check values.)
733 * The norm16 values of those characters are directly indexes into the extraData array.
734 *
735 * The data structures for composition lists and mappings are described in the design doc.
736 */
737
738 #endif /* !UCONFIG_NO_NORMALIZATION */
739 #endif /* __NORMALIZER2IMPL_H__ */
740