1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/intl/icu/source/common/unormcmp.cpp Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,646 @@
1.4 +/*
1.5 +*******************************************************************************
1.6 +*
1.7 +* Copyright (C) 2001-2011, International Business Machines
1.8 +* Corporation and others. All Rights Reserved.
1.9 +*
1.10 +*******************************************************************************
1.11 +* file name: unormcmp.cpp
1.12 +* encoding: US-ASCII
1.13 +* tab size: 8 (not used)
1.14 +* indentation:4
1.15 +*
1.16 +* created on: 2004sep13
1.17 +* created by: Markus W. Scherer
1.18 +*
1.19 +* unorm_compare() function moved here from unorm.cpp for better modularization.
1.20 +* Depends on both normalization and case folding.
1.21 +* Allows unorm.cpp to not depend on any character properties code.
1.22 +*/
1.23 +
1.24 +#include "unicode/utypes.h"
1.25 +
1.26 +#if !UCONFIG_NO_NORMALIZATION
1.27 +
1.28 +#include "unicode/unorm.h"
1.29 +#include "unicode/ustring.h"
1.30 +#include "cmemory.h"
1.31 +#include "normalizer2impl.h"
1.32 +#include "ucase.h"
1.33 +#include "uprops.h"
1.34 +#include "ustr_imp.h"
1.35 +
1.36 +U_NAMESPACE_USE
1.37 +
1.38 +#define LENGTHOF(array) (int32_t)(sizeof(array)/sizeof((array)[0]))
1.39 +
1.40 +/* compare canonically equivalent ------------------------------------------- */
1.41 +
1.42 +/*
1.43 + * Compare two strings for canonical equivalence.
1.44 + * Further options include case-insensitive comparison and
1.45 + * code point order (as opposed to code unit order).
1.46 + *
1.47 + * In this function, canonical equivalence is optional as well.
1.48 + * If canonical equivalence is tested, then both strings must fulfill
1.49 + * the FCD check.
1.50 + *
1.51 + * Semantically, this is equivalent to
1.52 + * strcmp[CodePointOrder](NFD(foldCase(s1)), NFD(foldCase(s2)))
1.53 + * where code point order, NFD and foldCase are all optional.
1.54 + *
1.55 + * String comparisons almost always yield results before processing both strings
1.56 + * completely.
1.57 + * They are generally more efficient working incrementally instead of
1.58 + * performing the sub-processing (strlen, normalization, case-folding)
1.59 + * on the entire strings first.
1.60 + *
1.61 + * It is also unnecessary to not normalize identical characters.
1.62 + *
1.63 + * This function works in principle as follows:
1.64 + *
1.65 + * loop {
1.66 + * get one code unit c1 from s1 (-1 if end of source)
1.67 + * get one code unit c2 from s2 (-1 if end of source)
1.68 + *
1.69 + * if(either string finished) {
1.70 + * return result;
1.71 + * }
1.72 + * if(c1==c2) {
1.73 + * continue;
1.74 + * }
1.75 + *
1.76 + * // c1!=c2
1.77 + * try to decompose/case-fold c1/c2, and continue if one does;
1.78 + *
1.79 + * // still c1!=c2 and neither decomposes/case-folds, return result
1.80 + * return c1-c2;
1.81 + * }
1.82 + *
1.83 + * When a character decomposes, then the pointer for that source changes to
1.84 + * the decomposition, pushing the previous pointer onto a stack.
1.85 + * When the end of the decomposition is reached, then the code unit reader
1.86 + * pops the previous source from the stack.
1.87 + * (Same for case-folding.)
1.88 + *
1.89 + * This is complicated further by operating on variable-width UTF-16.
1.90 + * The top part of the loop works on code units, while lookups for decomposition
1.91 + * and case-folding need code points.
1.92 + * Code points are assembled after the equality/end-of-source part.
1.93 + * The source pointer is only advanced beyond all code units when the code point
1.94 + * actually decomposes/case-folds.
1.95 + *
1.96 + * If we were on a trail surrogate unit when assembling a code point,
1.97 + * and the code point decomposes/case-folds, then the decomposition/folding
1.98 + * result must be compared with the part of the other string that corresponds to
1.99 + * this string's lead surrogate.
1.100 + * Since we only assemble a code point when hitting a trail unit when the
1.101 + * preceding lead units were identical, we back up the other string by one unit
1.102 + * in such a case.
1.103 + *
1.104 + * The optional code point order comparison at the end works with
1.105 + * the same fix-up as the other code point order comparison functions.
1.106 + * See ustring.c and the comment near the end of this function.
1.107 + *
1.108 + * Assumption: A decomposition or case-folding result string never contains
1.109 + * a single surrogate. This is a safe assumption in the Unicode Standard.
1.110 + * Therefore, we do not need to check for surrogate pairs across
1.111 + * decomposition/case-folding boundaries.
1.112 + *
1.113 + * Further assumptions (see verifications tstnorm.cpp):
1.114 + * The API function checks for FCD first, while the core function
1.115 + * first case-folds and then decomposes. This requires that case-folding does not
1.116 + * un-FCD any strings.
1.117 + *
1.118 + * The API function may also NFD the input and turn off decomposition.
1.119 + * This requires that case-folding does not un-NFD strings either.
1.120 + *
1.121 + * TODO If any of the above two assumptions is violated,
1.122 + * then this entire code must be re-thought.
1.123 + * If this happens, then a simple solution is to case-fold both strings up front
1.124 + * and to turn off UNORM_INPUT_IS_FCD.
1.125 + * We already do this when not both strings are in FCD because makeFCD
1.126 + * would be a partial NFD before the case folding, which does not work.
1.127 + * Note that all of this is only a problem when case-folding _and_
1.128 + * canonical equivalence come together.
1.129 + * (Comments in unorm_compare() are more up to date than this TODO.)
1.130 + */
1.131 +
1.132 +/* stack element for previous-level source/decomposition pointers */
1.133 +struct CmpEquivLevel {
1.134 + const UChar *start, *s, *limit;
1.135 +};
1.136 +typedef struct CmpEquivLevel CmpEquivLevel;
1.137 +
1.138 +/**
1.139 + * Internal option for unorm_cmpEquivFold() for decomposing.
1.140 + * If not set, just do strcasecmp().
1.141 + */
1.142 +#define _COMPARE_EQUIV 0x80000
1.143 +
1.144 +/* internal function */
1.145 +static int32_t
1.146 +unorm_cmpEquivFold(const UChar *s1, int32_t length1,
1.147 + const UChar *s2, int32_t length2,
1.148 + uint32_t options,
1.149 + UErrorCode *pErrorCode) {
1.150 + const Normalizer2Impl *nfcImpl;
1.151 + const UCaseProps *csp;
1.152 +
1.153 + /* current-level start/limit - s1/s2 as current */
1.154 + const UChar *start1, *start2, *limit1, *limit2;
1.155 +
1.156 + /* decomposition and case folding variables */
1.157 + const UChar *p;
1.158 + int32_t length;
1.159 +
1.160 + /* stacks of previous-level start/current/limit */
1.161 + CmpEquivLevel stack1[2], stack2[2];
1.162 +
1.163 + /* buffers for algorithmic decompositions */
1.164 + UChar decomp1[4], decomp2[4];
1.165 +
1.166 + /* case folding buffers, only use current-level start/limit */
1.167 + UChar fold1[UCASE_MAX_STRING_LENGTH+1], fold2[UCASE_MAX_STRING_LENGTH+1];
1.168 +
1.169 + /* track which is the current level per string */
1.170 + int32_t level1, level2;
1.171 +
1.172 + /* current code units, and code points for lookups */
1.173 + UChar32 c1, c2, cp1, cp2;
1.174 +
1.175 + /* no argument error checking because this itself is not an API */
1.176 +
1.177 + /*
1.178 + * assume that at least one of the options _COMPARE_EQUIV and U_COMPARE_IGNORE_CASE is set
1.179 + * otherwise this function must behave exactly as uprv_strCompare()
1.180 + * not checking for that here makes testing this function easier
1.181 + */
1.182 +
1.183 + /* normalization/properties data loaded? */
1.184 + if((options&_COMPARE_EQUIV)!=0) {
1.185 + nfcImpl=Normalizer2Factory::getNFCImpl(*pErrorCode);
1.186 + } else {
1.187 + nfcImpl=NULL;
1.188 + }
1.189 + if((options&U_COMPARE_IGNORE_CASE)!=0) {
1.190 + csp=ucase_getSingleton();
1.191 + } else {
1.192 + csp=NULL;
1.193 + }
1.194 + if(U_FAILURE(*pErrorCode)) {
1.195 + return 0;
1.196 + }
1.197 +
1.198 + /* initialize */
1.199 + start1=s1;
1.200 + if(length1==-1) {
1.201 + limit1=NULL;
1.202 + } else {
1.203 + limit1=s1+length1;
1.204 + }
1.205 +
1.206 + start2=s2;
1.207 + if(length2==-1) {
1.208 + limit2=NULL;
1.209 + } else {
1.210 + limit2=s2+length2;
1.211 + }
1.212 +
1.213 + level1=level2=0;
1.214 + c1=c2=-1;
1.215 +
1.216 + /* comparison loop */
1.217 + for(;;) {
1.218 + /*
1.219 + * here a code unit value of -1 means "get another code unit"
1.220 + * below it will mean "this source is finished"
1.221 + */
1.222 +
1.223 + if(c1<0) {
1.224 + /* get next code unit from string 1, post-increment */
1.225 + for(;;) {
1.226 + if(s1==limit1 || ((c1=*s1)==0 && (limit1==NULL || (options&_STRNCMP_STYLE)))) {
1.227 + if(level1==0) {
1.228 + c1=-1;
1.229 + break;
1.230 + }
1.231 + } else {
1.232 + ++s1;
1.233 + break;
1.234 + }
1.235 +
1.236 + /* reached end of level buffer, pop one level */
1.237 + do {
1.238 + --level1;
1.239 + start1=stack1[level1].start; /*Not uninitialized*/
1.240 + } while(start1==NULL);
1.241 + s1=stack1[level1].s; /*Not uninitialized*/
1.242 + limit1=stack1[level1].limit; /*Not uninitialized*/
1.243 + }
1.244 + }
1.245 +
1.246 + if(c2<0) {
1.247 + /* get next code unit from string 2, post-increment */
1.248 + for(;;) {
1.249 + if(s2==limit2 || ((c2=*s2)==0 && (limit2==NULL || (options&_STRNCMP_STYLE)))) {
1.250 + if(level2==0) {
1.251 + c2=-1;
1.252 + break;
1.253 + }
1.254 + } else {
1.255 + ++s2;
1.256 + break;
1.257 + }
1.258 +
1.259 + /* reached end of level buffer, pop one level */
1.260 + do {
1.261 + --level2;
1.262 + start2=stack2[level2].start; /*Not uninitialized*/
1.263 + } while(start2==NULL);
1.264 + s2=stack2[level2].s; /*Not uninitialized*/
1.265 + limit2=stack2[level2].limit; /*Not uninitialized*/
1.266 + }
1.267 + }
1.268 +
1.269 + /*
1.270 + * compare c1 and c2
1.271 + * either variable c1, c2 is -1 only if the corresponding string is finished
1.272 + */
1.273 + if(c1==c2) {
1.274 + if(c1<0) {
1.275 + return 0; /* c1==c2==-1 indicating end of strings */
1.276 + }
1.277 + c1=c2=-1; /* make us fetch new code units */
1.278 + continue;
1.279 + } else if(c1<0) {
1.280 + return -1; /* string 1 ends before string 2 */
1.281 + } else if(c2<0) {
1.282 + return 1; /* string 2 ends before string 1 */
1.283 + }
1.284 + /* c1!=c2 && c1>=0 && c2>=0 */
1.285 +
1.286 + /* get complete code points for c1, c2 for lookups if either is a surrogate */
1.287 + cp1=c1;
1.288 + if(U_IS_SURROGATE(c1)) {
1.289 + UChar c;
1.290 +
1.291 + if(U_IS_SURROGATE_LEAD(c1)) {
1.292 + if(s1!=limit1 && U16_IS_TRAIL(c=*s1)) {
1.293 + /* advance ++s1; only below if cp1 decomposes/case-folds */
1.294 + cp1=U16_GET_SUPPLEMENTARY(c1, c);
1.295 + }
1.296 + } else /* isTrail(c1) */ {
1.297 + if(start1<=(s1-2) && U16_IS_LEAD(c=*(s1-2))) {
1.298 + cp1=U16_GET_SUPPLEMENTARY(c, c1);
1.299 + }
1.300 + }
1.301 + }
1.302 +
1.303 + cp2=c2;
1.304 + if(U_IS_SURROGATE(c2)) {
1.305 + UChar c;
1.306 +
1.307 + if(U_IS_SURROGATE_LEAD(c2)) {
1.308 + if(s2!=limit2 && U16_IS_TRAIL(c=*s2)) {
1.309 + /* advance ++s2; only below if cp2 decomposes/case-folds */
1.310 + cp2=U16_GET_SUPPLEMENTARY(c2, c);
1.311 + }
1.312 + } else /* isTrail(c2) */ {
1.313 + if(start2<=(s2-2) && U16_IS_LEAD(c=*(s2-2))) {
1.314 + cp2=U16_GET_SUPPLEMENTARY(c, c2);
1.315 + }
1.316 + }
1.317 + }
1.318 +
1.319 + /*
1.320 + * go down one level for each string
1.321 + * continue with the main loop as soon as there is a real change
1.322 + */
1.323 +
1.324 + if( level1==0 && (options&U_COMPARE_IGNORE_CASE) &&
1.325 + (length=ucase_toFullFolding(csp, (UChar32)cp1, &p, options))>=0
1.326 + ) {
1.327 + /* cp1 case-folds to the code point "length" or to p[length] */
1.328 + if(U_IS_SURROGATE(c1)) {
1.329 + if(U_IS_SURROGATE_LEAD(c1)) {
1.330 + /* advance beyond source surrogate pair if it case-folds */
1.331 + ++s1;
1.332 + } else /* isTrail(c1) */ {
1.333 + /*
1.334 + * we got a supplementary code point when hitting its trail surrogate,
1.335 + * therefore the lead surrogate must have been the same as in the other string;
1.336 + * compare this decomposition with the lead surrogate in the other string
1.337 + * remember that this simulates bulk text replacement:
1.338 + * the decomposition would replace the entire code point
1.339 + */
1.340 + --s2;
1.341 + c2=*(s2-1);
1.342 + }
1.343 + }
1.344 +
1.345 + /* push current level pointers */
1.346 + stack1[0].start=start1;
1.347 + stack1[0].s=s1;
1.348 + stack1[0].limit=limit1;
1.349 + ++level1;
1.350 +
1.351 + /* copy the folding result to fold1[] */
1.352 + if(length<=UCASE_MAX_STRING_LENGTH) {
1.353 + u_memcpy(fold1, p, length);
1.354 + } else {
1.355 + int32_t i=0;
1.356 + U16_APPEND_UNSAFE(fold1, i, length);
1.357 + length=i;
1.358 + }
1.359 +
1.360 + /* set next level pointers to case folding */
1.361 + start1=s1=fold1;
1.362 + limit1=fold1+length;
1.363 +
1.364 + /* get ready to read from decomposition, continue with loop */
1.365 + c1=-1;
1.366 + continue;
1.367 + }
1.368 +
1.369 + if( level2==0 && (options&U_COMPARE_IGNORE_CASE) &&
1.370 + (length=ucase_toFullFolding(csp, (UChar32)cp2, &p, options))>=0
1.371 + ) {
1.372 + /* cp2 case-folds to the code point "length" or to p[length] */
1.373 + if(U_IS_SURROGATE(c2)) {
1.374 + if(U_IS_SURROGATE_LEAD(c2)) {
1.375 + /* advance beyond source surrogate pair if it case-folds */
1.376 + ++s2;
1.377 + } else /* isTrail(c2) */ {
1.378 + /*
1.379 + * we got a supplementary code point when hitting its trail surrogate,
1.380 + * therefore the lead surrogate must have been the same as in the other string;
1.381 + * compare this decomposition with the lead surrogate in the other string
1.382 + * remember that this simulates bulk text replacement:
1.383 + * the decomposition would replace the entire code point
1.384 + */
1.385 + --s1;
1.386 + c1=*(s1-1);
1.387 + }
1.388 + }
1.389 +
1.390 + /* push current level pointers */
1.391 + stack2[0].start=start2;
1.392 + stack2[0].s=s2;
1.393 + stack2[0].limit=limit2;
1.394 + ++level2;
1.395 +
1.396 + /* copy the folding result to fold2[] */
1.397 + if(length<=UCASE_MAX_STRING_LENGTH) {
1.398 + u_memcpy(fold2, p, length);
1.399 + } else {
1.400 + int32_t i=0;
1.401 + U16_APPEND_UNSAFE(fold2, i, length);
1.402 + length=i;
1.403 + }
1.404 +
1.405 + /* set next level pointers to case folding */
1.406 + start2=s2=fold2;
1.407 + limit2=fold2+length;
1.408 +
1.409 + /* get ready to read from decomposition, continue with loop */
1.410 + c2=-1;
1.411 + continue;
1.412 + }
1.413 +
1.414 + if( level1<2 && (options&_COMPARE_EQUIV) &&
1.415 + 0!=(p=nfcImpl->getDecomposition((UChar32)cp1, decomp1, length))
1.416 + ) {
1.417 + /* cp1 decomposes into p[length] */
1.418 + if(U_IS_SURROGATE(c1)) {
1.419 + if(U_IS_SURROGATE_LEAD(c1)) {
1.420 + /* advance beyond source surrogate pair if it decomposes */
1.421 + ++s1;
1.422 + } else /* isTrail(c1) */ {
1.423 + /*
1.424 + * we got a supplementary code point when hitting its trail surrogate,
1.425 + * therefore the lead surrogate must have been the same as in the other string;
1.426 + * compare this decomposition with the lead surrogate in the other string
1.427 + * remember that this simulates bulk text replacement:
1.428 + * the decomposition would replace the entire code point
1.429 + */
1.430 + --s2;
1.431 + c2=*(s2-1);
1.432 + }
1.433 + }
1.434 +
1.435 + /* push current level pointers */
1.436 + stack1[level1].start=start1;
1.437 + stack1[level1].s=s1;
1.438 + stack1[level1].limit=limit1;
1.439 + ++level1;
1.440 +
1.441 + /* set empty intermediate level if skipped */
1.442 + if(level1<2) {
1.443 + stack1[level1++].start=NULL;
1.444 + }
1.445 +
1.446 + /* set next level pointers to decomposition */
1.447 + start1=s1=p;
1.448 + limit1=p+length;
1.449 +
1.450 + /* get ready to read from decomposition, continue with loop */
1.451 + c1=-1;
1.452 + continue;
1.453 + }
1.454 +
1.455 + if( level2<2 && (options&_COMPARE_EQUIV) &&
1.456 + 0!=(p=nfcImpl->getDecomposition((UChar32)cp2, decomp2, length))
1.457 + ) {
1.458 + /* cp2 decomposes into p[length] */
1.459 + if(U_IS_SURROGATE(c2)) {
1.460 + if(U_IS_SURROGATE_LEAD(c2)) {
1.461 + /* advance beyond source surrogate pair if it decomposes */
1.462 + ++s2;
1.463 + } else /* isTrail(c2) */ {
1.464 + /*
1.465 + * we got a supplementary code point when hitting its trail surrogate,
1.466 + * therefore the lead surrogate must have been the same as in the other string;
1.467 + * compare this decomposition with the lead surrogate in the other string
1.468 + * remember that this simulates bulk text replacement:
1.469 + * the decomposition would replace the entire code point
1.470 + */
1.471 + --s1;
1.472 + c1=*(s1-1);
1.473 + }
1.474 + }
1.475 +
1.476 + /* push current level pointers */
1.477 + stack2[level2].start=start2;
1.478 + stack2[level2].s=s2;
1.479 + stack2[level2].limit=limit2;
1.480 + ++level2;
1.481 +
1.482 + /* set empty intermediate level if skipped */
1.483 + if(level2<2) {
1.484 + stack2[level2++].start=NULL;
1.485 + }
1.486 +
1.487 + /* set next level pointers to decomposition */
1.488 + start2=s2=p;
1.489 + limit2=p+length;
1.490 +
1.491 + /* get ready to read from decomposition, continue with loop */
1.492 + c2=-1;
1.493 + continue;
1.494 + }
1.495 +
1.496 + /*
1.497 + * no decomposition/case folding, max level for both sides:
1.498 + * return difference result
1.499 + *
1.500 + * code point order comparison must not just return cp1-cp2
1.501 + * because when single surrogates are present then the surrogate pairs
1.502 + * that formed cp1 and cp2 may be from different string indexes
1.503 + *
1.504 + * example: { d800 d800 dc01 } vs. { d800 dc00 }, compare at second code units
1.505 + * c1=d800 cp1=10001 c2=dc00 cp2=10000
1.506 + * cp1-cp2>0 but c1-c2<0 and in fact in UTF-32 it is { d800 10001 } < { 10000 }
1.507 + *
1.508 + * therefore, use same fix-up as in ustring.c/uprv_strCompare()
1.509 + * except: uprv_strCompare() fetches c=*s while this functions fetches c=*s++
1.510 + * so we have slightly different pointer/start/limit comparisons here
1.511 + */
1.512 +
1.513 + if(c1>=0xd800 && c2>=0xd800 && (options&U_COMPARE_CODE_POINT_ORDER)) {
1.514 + /* subtract 0x2800 from BMP code points to make them smaller than supplementary ones */
1.515 + if(
1.516 + (c1<=0xdbff && s1!=limit1 && U16_IS_TRAIL(*s1)) ||
1.517 + (U16_IS_TRAIL(c1) && start1!=(s1-1) && U16_IS_LEAD(*(s1-2)))
1.518 + ) {
1.519 + /* part of a surrogate pair, leave >=d800 */
1.520 + } else {
1.521 + /* BMP code point - may be surrogate code point - make <d800 */
1.522 + c1-=0x2800;
1.523 + }
1.524 +
1.525 + if(
1.526 + (c2<=0xdbff && s2!=limit2 && U16_IS_TRAIL(*s2)) ||
1.527 + (U16_IS_TRAIL(c2) && start2!=(s2-1) && U16_IS_LEAD(*(s2-2)))
1.528 + ) {
1.529 + /* part of a surrogate pair, leave >=d800 */
1.530 + } else {
1.531 + /* BMP code point - may be surrogate code point - make <d800 */
1.532 + c2-=0x2800;
1.533 + }
1.534 + }
1.535 +
1.536 + return c1-c2;
1.537 + }
1.538 +}
1.539 +
1.540 +static
1.541 +UBool _normalize(const Normalizer2 *n2, const UChar *s, int32_t length,
1.542 + UnicodeString &normalized, UErrorCode *pErrorCode) {
1.543 + UnicodeString str(length<0, s, length);
1.544 +
1.545 + // check if s fulfill the conditions
1.546 + int32_t spanQCYes=n2->spanQuickCheckYes(str, *pErrorCode);
1.547 + if (U_FAILURE(*pErrorCode)) {
1.548 + return FALSE;
1.549 + }
1.550 + /*
1.551 + * ICU 2.4 had a further optimization:
1.552 + * If both strings were not in FCD, then they were both NFD'ed,
1.553 + * and the _COMPARE_EQUIV option was turned off.
1.554 + * It is not entirely clear that this is valid with the current
1.555 + * definition of the canonical caseless match.
1.556 + * Therefore, ICU 2.6 removes that optimization.
1.557 + */
1.558 + if(spanQCYes<str.length()) {
1.559 + UnicodeString unnormalized=str.tempSubString(spanQCYes);
1.560 + normalized.setTo(FALSE, str.getBuffer(), spanQCYes);
1.561 + n2->normalizeSecondAndAppend(normalized, unnormalized, *pErrorCode);
1.562 + if (U_SUCCESS(*pErrorCode)) {
1.563 + return TRUE;
1.564 + }
1.565 + }
1.566 + return FALSE;
1.567 +}
1.568 +
1.569 +U_CAPI int32_t U_EXPORT2
1.570 +unorm_compare(const UChar *s1, int32_t length1,
1.571 + const UChar *s2, int32_t length2,
1.572 + uint32_t options,
1.573 + UErrorCode *pErrorCode) {
1.574 + /* argument checking */
1.575 + if(U_FAILURE(*pErrorCode)) {
1.576 + return 0;
1.577 + }
1.578 + if(s1==0 || length1<-1 || s2==0 || length2<-1) {
1.579 + *pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
1.580 + return 0;
1.581 + }
1.582 +
1.583 + UnicodeString fcd1, fcd2;
1.584 + int32_t normOptions=(int32_t)(options>>UNORM_COMPARE_NORM_OPTIONS_SHIFT);
1.585 + options|=_COMPARE_EQUIV;
1.586 +
1.587 + /*
1.588 + * UAX #21 Case Mappings, as fixed for Unicode version 4
1.589 + * (see Jitterbug 2021), defines a canonical caseless match as
1.590 + *
1.591 + * A string X is a canonical caseless match
1.592 + * for a string Y if and only if
1.593 + * NFD(toCasefold(NFD(X))) = NFD(toCasefold(NFD(Y)))
1.594 + *
1.595 + * For better performance, we check for FCD (or let the caller tell us that
1.596 + * both strings are in FCD) for the inner normalization.
1.597 + * BasicNormalizerTest::FindFoldFCDExceptions() makes sure that
1.598 + * case-folding preserves the FCD-ness of a string.
1.599 + * The outer normalization is then only performed by unorm_cmpEquivFold()
1.600 + * when there is a difference.
1.601 + *
1.602 + * Exception: When using the Turkic case-folding option, we do perform
1.603 + * full NFD first. This is because in the Turkic case precomposed characters
1.604 + * with 0049 capital I or 0069 small i fold differently whether they
1.605 + * are first decomposed or not, so an FCD check - a check only for
1.606 + * canonical order - is not sufficient.
1.607 + */
1.608 + if(!(options&UNORM_INPUT_IS_FCD) || (options&U_FOLD_CASE_EXCLUDE_SPECIAL_I)) {
1.609 + const Normalizer2 *n2;
1.610 + if(options&U_FOLD_CASE_EXCLUDE_SPECIAL_I) {
1.611 + n2=Normalizer2Factory::getNFDInstance(*pErrorCode);
1.612 + } else {
1.613 + n2=Normalizer2Factory::getFCDInstance(*pErrorCode);
1.614 + }
1.615 + if (U_FAILURE(*pErrorCode)) {
1.616 + return 0;
1.617 + }
1.618 +
1.619 + if(normOptions&UNORM_UNICODE_3_2) {
1.620 + const UnicodeSet *uni32=uniset_getUnicode32Instance(*pErrorCode);
1.621 + FilteredNormalizer2 fn2(*n2, *uni32);
1.622 + if(_normalize(&fn2, s1, length1, fcd1, pErrorCode)) {
1.623 + s1=fcd1.getBuffer();
1.624 + length1=fcd1.length();
1.625 + }
1.626 + if(_normalize(&fn2, s2, length2, fcd2, pErrorCode)) {
1.627 + s2=fcd2.getBuffer();
1.628 + length2=fcd2.length();
1.629 + }
1.630 + } else {
1.631 + if(_normalize(n2, s1, length1, fcd1, pErrorCode)) {
1.632 + s1=fcd1.getBuffer();
1.633 + length1=fcd1.length();
1.634 + }
1.635 + if(_normalize(n2, s2, length2, fcd2, pErrorCode)) {
1.636 + s2=fcd2.getBuffer();
1.637 + length2=fcd2.length();
1.638 + }
1.639 + }
1.640 + }
1.641 +
1.642 + if(U_SUCCESS(*pErrorCode)) {
1.643 + return unorm_cmpEquivFold(s1, length1, s2, length2, options, pErrorCode);
1.644 + } else {
1.645 + return 0;
1.646 + }
1.647 +}
1.648 +
1.649 +#endif /* #if !UCONFIG_NO_NORMALIZATION */
