• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2 *******************************************************************************
3 *
4 *   Copyright (C) 2001-2010, International Business Machines
5 *   Corporation and others.  All Rights Reserved.
6 *
7 *******************************************************************************
8 *   file name:  unormcmp.cpp
9 *   encoding:   US-ASCII
10 *   tab size:   8 (not used)
11 *   indentation:4
12 *
13 *   created on: 2004sep13
14 *   created by: Markus W. Scherer
15 *
16 *   unorm_compare() function moved here from unorm.cpp for better modularization.
17 *   Depends on both normalization and case folding.
18 *   Allows unorm.cpp to not depend on any character properties code.
19 */
20 
21 #include "unicode/utypes.h"
22 
23 #if !UCONFIG_NO_NORMALIZATION
24 
25 #include "unicode/unorm.h"
26 #include "unicode/ustring.h"
27 #include "cmemory.h"
28 #include "normalizer2impl.h"
29 #include "ucase.h"
30 #include "uprops.h"
31 #include "ustr_imp.h"
32 
33 U_NAMESPACE_USE
34 
35 #define LENGTHOF(array) (int32_t)(sizeof(array)/sizeof((array)[0]))
36 
37 /* compare canonically equivalent ------------------------------------------- */
38 
39 /*
40  * Compare two strings for canonical equivalence.
41  * Further options include case-insensitive comparison and
42  * code point order (as opposed to code unit order).
43  *
44  * In this function, canonical equivalence is optional as well.
45  * If canonical equivalence is tested, then both strings must fulfill
46  * the FCD check.
47  *
48  * Semantically, this is equivalent to
49  *   strcmp[CodePointOrder](NFD(foldCase(s1)), NFD(foldCase(s2)))
50  * where code point order, NFD and foldCase are all optional.
51  *
52  * String comparisons almost always yield results before processing both strings
53  * completely.
54  * They are generally more efficient working incrementally instead of
55  * performing the sub-processing (strlen, normalization, case-folding)
56  * on the entire strings first.
57  *
58  * It is also unnecessary to not normalize identical characters.
59  *
60  * This function works in principle as follows:
61  *
62  * loop {
63  *   get one code unit c1 from s1 (-1 if end of source)
64  *   get one code unit c2 from s2 (-1 if end of source)
65  *
66  *   if(either string finished) {
67  *     return result;
68  *   }
69  *   if(c1==c2) {
70  *     continue;
71  *   }
72  *
73  *   // c1!=c2
74  *   try to decompose/case-fold c1/c2, and continue if one does;
75  *
76  *   // still c1!=c2 and neither decomposes/case-folds, return result
77  *   return c1-c2;
78  * }
79  *
80  * When a character decomposes, then the pointer for that source changes to
81  * the decomposition, pushing the previous pointer onto a stack.
82  * When the end of the decomposition is reached, then the code unit reader
83  * pops the previous source from the stack.
84  * (Same for case-folding.)
85  *
86  * This is complicated further by operating on variable-width UTF-16.
87  * The top part of the loop works on code units, while lookups for decomposition
88  * and case-folding need code points.
89  * Code points are assembled after the equality/end-of-source part.
90  * The source pointer is only advanced beyond all code units when the code point
91  * actually decomposes/case-folds.
92  *
93  * If we were on a trail surrogate unit when assembling a code point,
94  * and the code point decomposes/case-folds, then the decomposition/folding
95  * result must be compared with the part of the other string that corresponds to
96  * this string's lead surrogate.
97  * Since we only assemble a code point when hitting a trail unit when the
98  * preceding lead units were identical, we back up the other string by one unit
99  * in such a case.
100  *
101  * The optional code point order comparison at the end works with
102  * the same fix-up as the other code point order comparison functions.
103  * See ustring.c and the comment near the end of this function.
104  *
105  * Assumption: A decomposition or case-folding result string never contains
106  * a single surrogate. This is a safe assumption in the Unicode Standard.
107  * Therefore, we do not need to check for surrogate pairs across
108  * decomposition/case-folding boundaries.
109  *
110  * Further assumptions (see verifications tstnorm.cpp):
111  * The API function checks for FCD first, while the core function
112  * first case-folds and then decomposes. This requires that case-folding does not
113  * un-FCD any strings.
114  *
115  * The API function may also NFD the input and turn off decomposition.
116  * This requires that case-folding does not un-NFD strings either.
117  *
118  * TODO If any of the above two assumptions is violated,
119  * then this entire code must be re-thought.
120  * If this happens, then a simple solution is to case-fold both strings up front
121  * and to turn off UNORM_INPUT_IS_FCD.
122  * We already do this when not both strings are in FCD because makeFCD
123  * would be a partial NFD before the case folding, which does not work.
124  * Note that all of this is only a problem when case-folding _and_
125  * canonical equivalence come together.
126  * (Comments in unorm_compare() are more up to date than this TODO.)
127  */
128 
129 /* stack element for previous-level source/decomposition pointers */
130 struct CmpEquivLevel {
131     const UChar *start, *s, *limit;
132 };
133 typedef struct CmpEquivLevel CmpEquivLevel;
134 
135 /**
136  * Internal option for unorm_cmpEquivFold() for decomposing.
137  * If not set, just do strcasecmp().
138  */
139 #define _COMPARE_EQUIV 0x80000
140 
141 /* internal function */
142 static int32_t
unorm_cmpEquivFold(const UChar * s1,int32_t length1,const UChar * s2,int32_t length2,uint32_t options,UErrorCode * pErrorCode)143 unorm_cmpEquivFold(const UChar *s1, int32_t length1,
144                    const UChar *s2, int32_t length2,
145                    uint32_t options,
146                    UErrorCode *pErrorCode) {
147     const Normalizer2Impl *nfcImpl;
148     const UCaseProps *csp;
149 
150     /* current-level start/limit - s1/s2 as current */
151     const UChar *start1, *start2, *limit1, *limit2;
152 
153     /* decomposition and case folding variables */
154     const UChar *p;
155     int32_t length;
156 
157     /* stacks of previous-level start/current/limit */
158     CmpEquivLevel stack1[2], stack2[2];
159 
160     /* buffers for algorithmic decompositions */
161     UChar decomp1[4], decomp2[4];
162 
163     /* case folding buffers, only use current-level start/limit */
164     UChar fold1[UCASE_MAX_STRING_LENGTH+1], fold2[UCASE_MAX_STRING_LENGTH+1];
165 
166     /* track which is the current level per string */
167     int32_t level1, level2;
168 
169     /* current code units, and code points for lookups */
170     UChar32 c1, c2, cp1, cp2;
171 
172     /* no argument error checking because this itself is not an API */
173 
174     /*
175      * assume that at least one of the options _COMPARE_EQUIV and U_COMPARE_IGNORE_CASE is set
176      * otherwise this function must behave exactly as uprv_strCompare()
177      * not checking for that here makes testing this function easier
178      */
179 
180     /* normalization/properties data loaded? */
181     if((options&_COMPARE_EQUIV)!=0) {
182         nfcImpl=Normalizer2Factory::getNFCImpl(*pErrorCode);
183     } else {
184         nfcImpl=NULL;
185     }
186     if((options&U_COMPARE_IGNORE_CASE)!=0) {
187         csp=ucase_getSingleton();
188     } else {
189         csp=NULL;
190     }
191     if(U_FAILURE(*pErrorCode)) {
192         return 0;
193     }
194 
195     /* initialize */
196     start1=s1;
197     if(length1==-1) {
198         limit1=NULL;
199     } else {
200         limit1=s1+length1;
201     }
202 
203     start2=s2;
204     if(length2==-1) {
205         limit2=NULL;
206     } else {
207         limit2=s2+length2;
208     }
209 
210     level1=level2=0;
211     c1=c2=-1;
212 
213     /* comparison loop */
214     for(;;) {
215         /*
216          * here a code unit value of -1 means "get another code unit"
217          * below it will mean "this source is finished"
218          */
219 
220         if(c1<0) {
221             /* get next code unit from string 1, post-increment */
222             for(;;) {
223                 if(s1==limit1 || ((c1=*s1)==0 && (limit1==NULL || (options&_STRNCMP_STYLE)))) {
224                     if(level1==0) {
225                         c1=-1;
226                         break;
227                     }
228                 } else {
229                     ++s1;
230                     break;
231                 }
232 
233                 /* reached end of level buffer, pop one level */
234                 do {
235                     --level1;
236                     start1=stack1[level1].start;
237                 } while(start1==NULL);
238                 s1=stack1[level1].s;
239                 limit1=stack1[level1].limit;
240             }
241         }
242 
243         if(c2<0) {
244             /* get next code unit from string 2, post-increment */
245             for(;;) {
246                 if(s2==limit2 || ((c2=*s2)==0 && (limit2==NULL || (options&_STRNCMP_STYLE)))) {
247                     if(level2==0) {
248                         c2=-1;
249                         break;
250                     }
251                 } else {
252                     ++s2;
253                     break;
254                 }
255 
256                 /* reached end of level buffer, pop one level */
257                 do {
258                     --level2;
259                     start2=stack2[level2].start;
260                 } while(start2==NULL);
261                 s2=stack2[level2].s;
262                 limit2=stack2[level2].limit;
263             }
264         }
265 
266         /*
267          * compare c1 and c2
268          * either variable c1, c2 is -1 only if the corresponding string is finished
269          */
270         if(c1==c2) {
271             if(c1<0) {
272                 return 0;   /* c1==c2==-1 indicating end of strings */
273             }
274             c1=c2=-1;       /* make us fetch new code units */
275             continue;
276         } else if(c1<0) {
277             return -1;      /* string 1 ends before string 2 */
278         } else if(c2<0) {
279             return 1;       /* string 2 ends before string 1 */
280         }
281         /* c1!=c2 && c1>=0 && c2>=0 */
282 
283         /* get complete code points for c1, c2 for lookups if either is a surrogate */
284         cp1=c1;
285         if(U_IS_SURROGATE(c1)) {
286             UChar c;
287 
288             if(U_IS_SURROGATE_LEAD(c1)) {
289                 if(s1!=limit1 && U16_IS_TRAIL(c=*s1)) {
290                     /* advance ++s1; only below if cp1 decomposes/case-folds */
291                     cp1=U16_GET_SUPPLEMENTARY(c1, c);
292                 }
293             } else /* isTrail(c1) */ {
294                 if(start1<=(s1-2) && U16_IS_LEAD(c=*(s1-2))) {
295                     cp1=U16_GET_SUPPLEMENTARY(c, c1);
296                 }
297             }
298         }
299 
300         cp2=c2;
301         if(U_IS_SURROGATE(c2)) {
302             UChar c;
303 
304             if(U_IS_SURROGATE_LEAD(c2)) {
305                 if(s2!=limit2 && U16_IS_TRAIL(c=*s2)) {
306                     /* advance ++s2; only below if cp2 decomposes/case-folds */
307                     cp2=U16_GET_SUPPLEMENTARY(c2, c);
308                 }
309             } else /* isTrail(c2) */ {
310                 if(start2<=(s2-2) && U16_IS_LEAD(c=*(s2-2))) {
311                     cp2=U16_GET_SUPPLEMENTARY(c, c2);
312                 }
313             }
314         }
315 
316         /*
317          * go down one level for each string
318          * continue with the main loop as soon as there is a real change
319          */
320 
321         if( level1==0 && (options&U_COMPARE_IGNORE_CASE) &&
322             (length=ucase_toFullFolding(csp, (UChar32)cp1, &p, options))>=0
323         ) {
324             /* cp1 case-folds to the code point "length" or to p[length] */
325             if(U_IS_SURROGATE(c1)) {
326                 if(U_IS_SURROGATE_LEAD(c1)) {
327                     /* advance beyond source surrogate pair if it case-folds */
328                     ++s1;
329                 } else /* isTrail(c1) */ {
330                     /*
331                      * we got a supplementary code point when hitting its trail surrogate,
332                      * therefore the lead surrogate must have been the same as in the other string;
333                      * compare this decomposition with the lead surrogate in the other string
334                      * remember that this simulates bulk text replacement:
335                      * the decomposition would replace the entire code point
336                      */
337                     --s2;
338                     c2=*(s2-1);
339                 }
340             }
341 
342             /* push current level pointers */
343             stack1[0].start=start1;
344             stack1[0].s=s1;
345             stack1[0].limit=limit1;
346             ++level1;
347 
348             /* copy the folding result to fold1[] */
349             if(length<=UCASE_MAX_STRING_LENGTH) {
350                 u_memcpy(fold1, p, length);
351             } else {
352                 int32_t i=0;
353                 U16_APPEND_UNSAFE(fold1, i, length);
354                 length=i;
355             }
356 
357             /* set next level pointers to case folding */
358             start1=s1=fold1;
359             limit1=fold1+length;
360 
361             /* get ready to read from decomposition, continue with loop */
362             c1=-1;
363             continue;
364         }
365 
366         if( level2==0 && (options&U_COMPARE_IGNORE_CASE) &&
367             (length=ucase_toFullFolding(csp, (UChar32)cp2, &p, options))>=0
368         ) {
369             /* cp2 case-folds to the code point "length" or to p[length] */
370             if(U_IS_SURROGATE(c2)) {
371                 if(U_IS_SURROGATE_LEAD(c2)) {
372                     /* advance beyond source surrogate pair if it case-folds */
373                     ++s2;
374                 } else /* isTrail(c2) */ {
375                     /*
376                      * we got a supplementary code point when hitting its trail surrogate,
377                      * therefore the lead surrogate must have been the same as in the other string;
378                      * compare this decomposition with the lead surrogate in the other string
379                      * remember that this simulates bulk text replacement:
380                      * the decomposition would replace the entire code point
381                      */
382                     --s1;
383                     c1=*(s1-1);
384                 }
385             }
386 
387             /* push current level pointers */
388             stack2[0].start=start2;
389             stack2[0].s=s2;
390             stack2[0].limit=limit2;
391             ++level2;
392 
393             /* copy the folding result to fold2[] */
394             if(length<=UCASE_MAX_STRING_LENGTH) {
395                 u_memcpy(fold2, p, length);
396             } else {
397                 int32_t i=0;
398                 U16_APPEND_UNSAFE(fold2, i, length);
399                 length=i;
400             }
401 
402             /* set next level pointers to case folding */
403             start2=s2=fold2;
404             limit2=fold2+length;
405 
406             /* get ready to read from decomposition, continue with loop */
407             c2=-1;
408             continue;
409         }
410 
411         if( level1<2 && (options&_COMPARE_EQUIV) &&
412             0!=(p=nfcImpl->getDecomposition((UChar32)cp1, decomp1, length))
413         ) {
414             /* cp1 decomposes into p[length] */
415             if(U_IS_SURROGATE(c1)) {
416                 if(U_IS_SURROGATE_LEAD(c1)) {
417                     /* advance beyond source surrogate pair if it decomposes */
418                     ++s1;
419                 } else /* isTrail(c1) */ {
420                     /*
421                      * we got a supplementary code point when hitting its trail surrogate,
422                      * therefore the lead surrogate must have been the same as in the other string;
423                      * compare this decomposition with the lead surrogate in the other string
424                      * remember that this simulates bulk text replacement:
425                      * the decomposition would replace the entire code point
426                      */
427                     --s2;
428                     c2=*(s2-1);
429                 }
430             }
431 
432             /* push current level pointers */
433             stack1[level1].start=start1;
434             stack1[level1].s=s1;
435             stack1[level1].limit=limit1;
436             ++level1;
437 
438             /* set empty intermediate level if skipped */
439             if(level1<2) {
440                 stack1[level1++].start=NULL;
441             }
442 
443             /* set next level pointers to decomposition */
444             start1=s1=p;
445             limit1=p+length;
446 
447             /* get ready to read from decomposition, continue with loop */
448             c1=-1;
449             continue;
450         }
451 
452         if( level2<2 && (options&_COMPARE_EQUIV) &&
453             0!=(p=nfcImpl->getDecomposition((UChar32)cp2, decomp2, length))
454         ) {
455             /* cp2 decomposes into p[length] */
456             if(U_IS_SURROGATE(c2)) {
457                 if(U_IS_SURROGATE_LEAD(c2)) {
458                     /* advance beyond source surrogate pair if it decomposes */
459                     ++s2;
460                 } else /* isTrail(c2) */ {
461                     /*
462                      * we got a supplementary code point when hitting its trail surrogate,
463                      * therefore the lead surrogate must have been the same as in the other string;
464                      * compare this decomposition with the lead surrogate in the other string
465                      * remember that this simulates bulk text replacement:
466                      * the decomposition would replace the entire code point
467                      */
468                     --s1;
469                     c1=*(s1-1);
470                 }
471             }
472 
473             /* push current level pointers */
474             stack2[level2].start=start2;
475             stack2[level2].s=s2;
476             stack2[level2].limit=limit2;
477             ++level2;
478 
479             /* set empty intermediate level if skipped */
480             if(level2<2) {
481                 stack2[level2++].start=NULL;
482             }
483 
484             /* set next level pointers to decomposition */
485             start2=s2=p;
486             limit2=p+length;
487 
488             /* get ready to read from decomposition, continue with loop */
489             c2=-1;
490             continue;
491         }
492 
493         /*
494          * no decomposition/case folding, max level for both sides:
495          * return difference result
496          *
497          * code point order comparison must not just return cp1-cp2
498          * because when single surrogates are present then the surrogate pairs
499          * that formed cp1 and cp2 may be from different string indexes
500          *
501          * example: { d800 d800 dc01 } vs. { d800 dc00 }, compare at second code units
502          * c1=d800 cp1=10001 c2=dc00 cp2=10000
503          * cp1-cp2>0 but c1-c2<0 and in fact in UTF-32 it is { d800 10001 } < { 10000 }
504          *
505          * therefore, use same fix-up as in ustring.c/uprv_strCompare()
506          * except: uprv_strCompare() fetches c=*s while this functions fetches c=*s++
507          * so we have slightly different pointer/start/limit comparisons here
508          */
509 
510         if(c1>=0xd800 && c2>=0xd800 && (options&U_COMPARE_CODE_POINT_ORDER)) {
511             /* subtract 0x2800 from BMP code points to make them smaller than supplementary ones */
512             if(
513                 (c1<=0xdbff && s1!=limit1 && U16_IS_TRAIL(*s1)) ||
514                 (U16_IS_TRAIL(c1) && start1!=(s1-1) && U16_IS_LEAD(*(s1-2)))
515             ) {
516                 /* part of a surrogate pair, leave >=d800 */
517             } else {
518                 /* BMP code point - may be surrogate code point - make <d800 */
519                 c1-=0x2800;
520             }
521 
522             if(
523                 (c2<=0xdbff && s2!=limit2 && U16_IS_TRAIL(*s2)) ||
524                 (U16_IS_TRAIL(c2) && start2!=(s2-1) && U16_IS_LEAD(*(s2-2)))
525             ) {
526                 /* part of a surrogate pair, leave >=d800 */
527             } else {
528                 /* BMP code point - may be surrogate code point - make <d800 */
529                 c2-=0x2800;
530             }
531         }
532 
533         return c1-c2;
534     }
535 }
536 
537 U_CAPI int32_t U_EXPORT2
unorm_compare(const UChar * s1,int32_t length1,const UChar * s2,int32_t length2,uint32_t options,UErrorCode * pErrorCode)538 unorm_compare(const UChar *s1, int32_t length1,
539               const UChar *s2, int32_t length2,
540               uint32_t options,
541               UErrorCode *pErrorCode) {
542     /* argument checking */
543     if(U_FAILURE(*pErrorCode)) {
544         return 0;
545     }
546     if(s1==0 || length1<-1 || s2==0 || length2<-1) {
547         *pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
548         return 0;
549     }
550 
551     UnicodeString fcd1, fcd2;
552     int32_t normOptions=(int32_t)(options>>UNORM_COMPARE_NORM_OPTIONS_SHIFT);
553     options|=_COMPARE_EQUIV;
554 
555     /*
556      * UAX #21 Case Mappings, as fixed for Unicode version 4
557      * (see Jitterbug 2021), defines a canonical caseless match as
558      *
559      * A string X is a canonical caseless match
560      * for a string Y if and only if
561      * NFD(toCasefold(NFD(X))) = NFD(toCasefold(NFD(Y)))
562      *
563      * For better performance, we check for FCD (or let the caller tell us that
564      * both strings are in FCD) for the inner normalization.
565      * BasicNormalizerTest::FindFoldFCDExceptions() makes sure that
566      * case-folding preserves the FCD-ness of a string.
567      * The outer normalization is then only performed by unorm_cmpEquivFold()
568      * when there is a difference.
569      *
570      * Exception: When using the Turkic case-folding option, we do perform
571      * full NFD first. This is because in the Turkic case precomposed characters
572      * with 0049 capital I or 0069 small i fold differently whether they
573      * are first decomposed or not, so an FCD check - a check only for
574      * canonical order - is not sufficient.
575      */
576     if(!(options&UNORM_INPUT_IS_FCD) || (options&U_FOLD_CASE_EXCLUDE_SPECIAL_I)) {
577         const Normalizer2 *n2;
578         if(options&U_FOLD_CASE_EXCLUDE_SPECIAL_I) {
579             n2=Normalizer2Factory::getNFDInstance(*pErrorCode);
580         } else {
581             n2=Normalizer2Factory::getFCDInstance(*pErrorCode);
582         }
583         if (U_FAILURE(*pErrorCode)) {
584             return 0;
585         }
586 
587         // check if s1 and/or s2 fulfill the FCD conditions
588         const UnicodeSet *uni32;
589         if(normOptions&UNORM_UNICODE_3_2) {
590             uni32=uniset_getUnicode32Instance(*pErrorCode);
591         } else {
592             uni32=NULL;  // unused
593         }
594         FilteredNormalizer2 fn2(*n2, *uni32);
595         if(normOptions&UNORM_UNICODE_3_2) {
596             n2=&fn2;
597         }
598 
599         UnicodeString str1(length1<0, s1, length1);
600         UnicodeString str2(length2<0, s2, length2);
601         int32_t spanQCYes1=n2->spanQuickCheckYes(str1, *pErrorCode);
602         int32_t spanQCYes2=n2->spanQuickCheckYes(str2, *pErrorCode);
603         if(U_FAILURE(*pErrorCode)) {
604             return 0;
605         }
606 
607         /*
608          * ICU 2.4 had a further optimization:
609          * If both strings were not in FCD, then they were both NFD'ed,
610          * and the _COMPARE_EQUIV option was turned off.
611          * It is not entirely clear that this is valid with the current
612          * definition of the canonical caseless match.
613          * Therefore, ICU 2.6 removes that optimization.
614          */
615 
616         if(spanQCYes1<str1.length()) {
617             UnicodeString unnormalized=str1.tempSubString(spanQCYes1);
618             fcd1.setTo(FALSE, str1.getBuffer(), spanQCYes1);
619             n2->normalizeSecondAndAppend(fcd1, unnormalized, *pErrorCode);
620             s1=fcd1.getBuffer();
621             length1=fcd1.length();
622         }
623         if(spanQCYes2<str2.length()) {
624             UnicodeString unnormalized=str2.tempSubString(spanQCYes2);
625             fcd2.setTo(FALSE, str2.getBuffer(), spanQCYes2);
626             n2->normalizeSecondAndAppend(fcd2, unnormalized, *pErrorCode);
627             s2=fcd2.getBuffer();
628             length2=fcd2.length();
629         }
630     }
631 
632     if(U_SUCCESS(*pErrorCode)) {
633         return unorm_cmpEquivFold(s1, length1, s2, length2, options, pErrorCode);
634     } else {
635         return 0;
636     }
637 }
638 
639 #endif /* #if !UCONFIG_NO_NORMALIZATION */
640