michael@0: /*
michael@0: *******************************************************************************
michael@0: *
michael@0: *   Copyright (C) 2001-2011, International Business Machines
michael@0: *   Corporation and others.  All Rights Reserved.
michael@0: *
michael@0: *******************************************************************************
michael@0: *   file name:  unormcmp.cpp
michael@0: *   encoding:   US-ASCII
michael@0: *   tab size:   8 (not used)
michael@0: *   indentation:4
michael@0: *
michael@0: *   created on: 2004sep13
michael@0: *   created by: Markus W. Scherer
michael@0: *
michael@0: *   unorm_compare() function moved here from unorm.cpp for better modularization.
michael@0: *   Depends on both normalization and case folding.
michael@0: *   Allows unorm.cpp to not depend on any character properties code.
michael@0: */
michael@0: 
michael@0: #include "unicode/utypes.h"
michael@0: 
michael@0: #if !UCONFIG_NO_NORMALIZATION
michael@0: 
michael@0: #include "unicode/unorm.h"
michael@0: #include "unicode/ustring.h"
michael@0: #include "cmemory.h"
michael@0: #include "normalizer2impl.h"
michael@0: #include "ucase.h"
michael@0: #include "uprops.h"
michael@0: #include "ustr_imp.h"
michael@0: 
michael@0: U_NAMESPACE_USE
michael@0: 
michael@0: #define LENGTHOF(array) (int32_t)(sizeof(array)/sizeof((array)[0]))
michael@0: 
michael@0: /* compare canonically equivalent ------------------------------------------- */
michael@0: 
michael@0: /*
michael@0:  * Compare two strings for canonical equivalence.
michael@0:  * Further options include case-insensitive comparison and
michael@0:  * code point order (as opposed to code unit order).
michael@0:  *
michael@0:  * In this function, canonical equivalence is optional as well.
michael@0:  * If canonical equivalence is tested, then both strings must fulfill
michael@0:  * the FCD check.
michael@0:  *
michael@0:  * Semantically, this is equivalent to
michael@0:  *   strcmp[CodePointOrder](NFD(foldCase(s1)), NFD(foldCase(s2)))
michael@0:  * where code point order, NFD and foldCase are all optional.
michael@0:  *
michael@0:  * String comparisons almost always yield results before processing both strings
michael@0:  * completely.
michael@0:  * They are generally more efficient working incrementally instead of
michael@0:  * performing the sub-processing (strlen, normalization, case-folding)
michael@0:  * on the entire strings first.
michael@0:  *
michael@0:  * It is also unnecessary to not normalize identical characters.
michael@0:  *
michael@0:  * This function works in principle as follows:
michael@0:  *
michael@0:  * loop {
michael@0:  *   get one code unit c1 from s1 (-1 if end of source)
michael@0:  *   get one code unit c2 from s2 (-1 if end of source)
michael@0:  *
michael@0:  *   if(either string finished) {
michael@0:  *     return result;
michael@0:  *   }
michael@0:  *   if(c1==c2) {
michael@0:  *     continue;
michael@0:  *   }
michael@0:  *
michael@0:  *   // c1!=c2
michael@0:  *   try to decompose/case-fold c1/c2, and continue if one does;
michael@0:  *
michael@0:  *   // still c1!=c2 and neither decomposes/case-folds, return result
michael@0:  *   return c1-c2;
michael@0:  * }
michael@0:  *
michael@0:  * When a character decomposes, then the pointer for that source changes to
michael@0:  * the decomposition, pushing the previous pointer onto a stack.
michael@0:  * When the end of the decomposition is reached, then the code unit reader
michael@0:  * pops the previous source from the stack.
michael@0:  * (Same for case-folding.)
michael@0:  *
michael@0:  * This is complicated further by operating on variable-width UTF-16.
michael@0:  * The top part of the loop works on code units, while lookups for decomposition
michael@0:  * and case-folding need code points.
michael@0:  * Code points are assembled after the equality/end-of-source part.
michael@0:  * The source pointer is only advanced beyond all code units when the code point
michael@0:  * actually decomposes/case-folds.
michael@0:  *
michael@0:  * If we were on a trail surrogate unit when assembling a code point,
michael@0:  * and the code point decomposes/case-folds, then the decomposition/folding
michael@0:  * result must be compared with the part of the other string that corresponds to
michael@0:  * this string's lead surrogate.
michael@0:  * Since we only assemble a code point when hitting a trail unit when the
michael@0:  * preceding lead units were identical, we back up the other string by one unit
michael@0:  * in such a case.
michael@0:  *
michael@0:  * The optional code point order comparison at the end works with
michael@0:  * the same fix-up as the other code point order comparison functions.
michael@0:  * See ustring.c and the comment near the end of this function.
michael@0:  *
michael@0:  * Assumption: A decomposition or case-folding result string never contains
michael@0:  * a single surrogate. This is a safe assumption in the Unicode Standard.
michael@0:  * Therefore, we do not need to check for surrogate pairs across
michael@0:  * decomposition/case-folding boundaries.
michael@0:  *
michael@0:  * Further assumptions (see verifications tstnorm.cpp):
michael@0:  * The API function checks for FCD first, while the core function
michael@0:  * first case-folds and then decomposes. This requires that case-folding does not
michael@0:  * un-FCD any strings.
michael@0:  *
michael@0:  * The API function may also NFD the input and turn off decomposition.
michael@0:  * This requires that case-folding does not un-NFD strings either.
michael@0:  *
michael@0:  * TODO If any of the above two assumptions is violated,
michael@0:  * then this entire code must be re-thought.
michael@0:  * If this happens, then a simple solution is to case-fold both strings up front
michael@0:  * and to turn off UNORM_INPUT_IS_FCD.
michael@0:  * We already do this when not both strings are in FCD because makeFCD
michael@0:  * would be a partial NFD before the case folding, which does not work.
michael@0:  * Note that all of this is only a problem when case-folding _and_
michael@0:  * canonical equivalence come together.
michael@0:  * (Comments in unorm_compare() are more up to date than this TODO.)
michael@0:  */
michael@0: 
michael@0: /* stack element for previous-level source/decomposition pointers */
michael@0: struct CmpEquivLevel {
michael@0:     const UChar *start, *s, *limit;
michael@0: };
michael@0: typedef struct CmpEquivLevel CmpEquivLevel;
michael@0: 
michael@0: /**
michael@0:  * Internal option for unorm_cmpEquivFold() for decomposing.
michael@0:  * If not set, just do strcasecmp().
michael@0:  */
michael@0: #define _COMPARE_EQUIV 0x80000
michael@0: 
michael@0: /* internal function */
michael@0: static int32_t
michael@0: unorm_cmpEquivFold(const UChar *s1, int32_t length1,
michael@0:                    const UChar *s2, int32_t length2,
michael@0:                    uint32_t options,
michael@0:                    UErrorCode *pErrorCode) {
michael@0:     const Normalizer2Impl *nfcImpl;
michael@0:     const UCaseProps *csp;
michael@0: 
michael@0:     /* current-level start/limit - s1/s2 as current */
michael@0:     const UChar *start1, *start2, *limit1, *limit2;
michael@0: 
michael@0:     /* decomposition and case folding variables */
michael@0:     const UChar *p;
michael@0:     int32_t length;
michael@0: 
michael@0:     /* stacks of previous-level start/current/limit */
michael@0:     CmpEquivLevel stack1[2], stack2[2];
michael@0: 
michael@0:     /* buffers for algorithmic decompositions */
michael@0:     UChar decomp1[4], decomp2[4];
michael@0: 
michael@0:     /* case folding buffers, only use current-level start/limit */
michael@0:     UChar fold1[UCASE_MAX_STRING_LENGTH+1], fold2[UCASE_MAX_STRING_LENGTH+1];
michael@0: 
michael@0:     /* track which is the current level per string */
michael@0:     int32_t level1, level2;
michael@0: 
michael@0:     /* current code units, and code points for lookups */
michael@0:     UChar32 c1, c2, cp1, cp2;
michael@0: 
michael@0:     /* no argument error checking because this itself is not an API */
michael@0: 
michael@0:     /*
michael@0:      * assume that at least one of the options _COMPARE_EQUIV and U_COMPARE_IGNORE_CASE is set
michael@0:      * otherwise this function must behave exactly as uprv_strCompare()
michael@0:      * not checking for that here makes testing this function easier
michael@0:      */
michael@0: 
michael@0:     /* normalization/properties data loaded? */
michael@0:     if((options&_COMPARE_EQUIV)!=0) {
michael@0:         nfcImpl=Normalizer2Factory::getNFCImpl(*pErrorCode);
michael@0:     } else {
michael@0:         nfcImpl=NULL;
michael@0:     }
michael@0:     if((options&U_COMPARE_IGNORE_CASE)!=0) {
michael@0:         csp=ucase_getSingleton();
michael@0:     } else {
michael@0:         csp=NULL;
michael@0:     }
michael@0:     if(U_FAILURE(*pErrorCode)) {
michael@0:         return 0;
michael@0:     }
michael@0: 
michael@0:     /* initialize */
michael@0:     start1=s1;
michael@0:     if(length1==-1) {
michael@0:         limit1=NULL;
michael@0:     } else {
michael@0:         limit1=s1+length1;
michael@0:     }
michael@0: 
michael@0:     start2=s2;
michael@0:     if(length2==-1) {
michael@0:         limit2=NULL;
michael@0:     } else {
michael@0:         limit2=s2+length2;
michael@0:     }
michael@0: 
michael@0:     level1=level2=0;
michael@0:     c1=c2=-1;
michael@0: 
michael@0:     /* comparison loop */
michael@0:     for(;;) {
michael@0:         /*
michael@0:          * here a code unit value of -1 means "get another code unit"
michael@0:          * below it will mean "this source is finished"
michael@0:          */
michael@0: 
michael@0:         if(c1<0) {
michael@0:             /* get next code unit from string 1, post-increment */
michael@0:             for(;;) {
michael@0:                 if(s1==limit1 || ((c1=*s1)==0 && (limit1==NULL || (options&_STRNCMP_STYLE)))) {
michael@0:                     if(level1==0) {
michael@0:                         c1=-1;
michael@0:                         break;
michael@0:                     }
michael@0:                 } else {
michael@0:                     ++s1;
michael@0:                     break;
michael@0:                 }
michael@0: 
michael@0:                 /* reached end of level buffer, pop one level */
michael@0:                 do {
michael@0:                     --level1;
michael@0:                     start1=stack1[level1].start;    /*Not uninitialized*/
michael@0:                 } while(start1==NULL);
michael@0:                 s1=stack1[level1].s;                /*Not uninitialized*/
michael@0:                 limit1=stack1[level1].limit;        /*Not uninitialized*/
michael@0:             }
michael@0:         }
michael@0: 
michael@0:         if(c2<0) {
michael@0:             /* get next code unit from string 2, post-increment */
michael@0:             for(;;) {
michael@0:                 if(s2==limit2 || ((c2=*s2)==0 && (limit2==NULL || (options&_STRNCMP_STYLE)))) {
michael@0:                     if(level2==0) {
michael@0:                         c2=-1;
michael@0:                         break;
michael@0:                     }
michael@0:                 } else {
michael@0:                     ++s2;
michael@0:                     break;
michael@0:                 }
michael@0: 
michael@0:                 /* reached end of level buffer, pop one level */
michael@0:                 do {
michael@0:                     --level2;
michael@0:                     start2=stack2[level2].start;    /*Not uninitialized*/
michael@0:                 } while(start2==NULL);
michael@0:                 s2=stack2[level2].s;                /*Not uninitialized*/
michael@0:                 limit2=stack2[level2].limit;        /*Not uninitialized*/
michael@0:             }
michael@0:         }
michael@0: 
michael@0:         /*
michael@0:          * compare c1 and c2
michael@0:          * either variable c1, c2 is -1 only if the corresponding string is finished
michael@0:          */
michael@0:         if(c1==c2) {
michael@0:             if(c1<0) {
michael@0:                 return 0;   /* c1==c2==-1 indicating end of strings */
michael@0:             }
michael@0:             c1=c2=-1;       /* make us fetch new code units */
michael@0:             continue;
michael@0:         } else if(c1<0) {
michael@0:             return -1;      /* string 1 ends before string 2 */
michael@0:         } else if(c2<0) {
michael@0:             return 1;       /* string 2 ends before string 1 */
michael@0:         }
michael@0:         /* c1!=c2 && c1>=0 && c2>=0 */
michael@0: 
michael@0:         /* get complete code points for c1, c2 for lookups if either is a surrogate */
michael@0:         cp1=c1;
michael@0:         if(U_IS_SURROGATE(c1)) {
michael@0:             UChar c;
michael@0: 
michael@0:             if(U_IS_SURROGATE_LEAD(c1)) {
michael@0:                 if(s1!=limit1 && U16_IS_TRAIL(c=*s1)) {
michael@0:                     /* advance ++s1; only below if cp1 decomposes/case-folds */
michael@0:                     cp1=U16_GET_SUPPLEMENTARY(c1, c);
michael@0:                 }
michael@0:             } else /* isTrail(c1) */ {
michael@0:                 if(start1<=(s1-2) && U16_IS_LEAD(c=*(s1-2))) {
michael@0:                     cp1=U16_GET_SUPPLEMENTARY(c, c1);
michael@0:                 }
michael@0:             }
michael@0:         }
michael@0: 
michael@0:         cp2=c2;
michael@0:         if(U_IS_SURROGATE(c2)) {
michael@0:             UChar c;
michael@0: 
michael@0:             if(U_IS_SURROGATE_LEAD(c2)) {
michael@0:                 if(s2!=limit2 && U16_IS_TRAIL(c=*s2)) {
michael@0:                     /* advance ++s2; only below if cp2 decomposes/case-folds */
michael@0:                     cp2=U16_GET_SUPPLEMENTARY(c2, c);
michael@0:                 }
michael@0:             } else /* isTrail(c2) */ {
michael@0:                 if(start2<=(s2-2) && U16_IS_LEAD(c=*(s2-2))) {
michael@0:                     cp2=U16_GET_SUPPLEMENTARY(c, c2);
michael@0:                 }
michael@0:             }
michael@0:         }
michael@0: 
michael@0:         /*
michael@0:          * go down one level for each string
michael@0:          * continue with the main loop as soon as there is a real change
michael@0:          */
michael@0: 
michael@0:         if( level1==0 && (options&U_COMPARE_IGNORE_CASE) &&
michael@0:             (length=ucase_toFullFolding(csp, (UChar32)cp1, &p, options))>=0
michael@0:         ) {
michael@0:             /* cp1 case-folds to the code point "length" or to p[length] */
michael@0:             if(U_IS_SURROGATE(c1)) {
michael@0:                 if(U_IS_SURROGATE_LEAD(c1)) {
michael@0:                     /* advance beyond source surrogate pair if it case-folds */
michael@0:                     ++s1;
michael@0:                 } else /* isTrail(c1) */ {
michael@0:                     /*
michael@0:                      * we got a supplementary code point when hitting its trail surrogate,
michael@0:                      * therefore the lead surrogate must have been the same as in the other string;
michael@0:                      * compare this decomposition with the lead surrogate in the other string
michael@0:                      * remember that this simulates bulk text replacement:
michael@0:                      * the decomposition would replace the entire code point
michael@0:                      */
michael@0:                     --s2;
michael@0:                     c2=*(s2-1);
michael@0:                 }
michael@0:             }
michael@0: 
michael@0:             /* push current level pointers */
michael@0:             stack1[0].start=start1;
michael@0:             stack1[0].s=s1;
michael@0:             stack1[0].limit=limit1;
michael@0:             ++level1;
michael@0: 
michael@0:             /* copy the folding result to fold1[] */
michael@0:             if(length<=UCASE_MAX_STRING_LENGTH) {
michael@0:                 u_memcpy(fold1, p, length);
michael@0:             } else {
michael@0:                 int32_t i=0;
michael@0:                 U16_APPEND_UNSAFE(fold1, i, length);
michael@0:                 length=i;
michael@0:             }
michael@0: 
michael@0:             /* set next level pointers to case folding */
michael@0:             start1=s1=fold1;
michael@0:             limit1=fold1+length;
michael@0: 
michael@0:             /* get ready to read from decomposition, continue with loop */
michael@0:             c1=-1;
michael@0:             continue;
michael@0:         }
michael@0: 
michael@0:         if( level2==0 && (options&U_COMPARE_IGNORE_CASE) &&
michael@0:             (length=ucase_toFullFolding(csp, (UChar32)cp2, &p, options))>=0
michael@0:         ) {
michael@0:             /* cp2 case-folds to the code point "length" or to p[length] */
michael@0:             if(U_IS_SURROGATE(c2)) {
michael@0:                 if(U_IS_SURROGATE_LEAD(c2)) {
michael@0:                     /* advance beyond source surrogate pair if it case-folds */
michael@0:                     ++s2;
michael@0:                 } else /* isTrail(c2) */ {
michael@0:                     /*
michael@0:                      * we got a supplementary code point when hitting its trail surrogate,
michael@0:                      * therefore the lead surrogate must have been the same as in the other string;
michael@0:                      * compare this decomposition with the lead surrogate in the other string
michael@0:                      * remember that this simulates bulk text replacement:
michael@0:                      * the decomposition would replace the entire code point
michael@0:                      */
michael@0:                     --s1;
michael@0:                     c1=*(s1-1);
michael@0:                 }
michael@0:             }
michael@0: 
michael@0:             /* push current level pointers */
michael@0:             stack2[0].start=start2;
michael@0:             stack2[0].s=s2;
michael@0:             stack2[0].limit=limit2;
michael@0:             ++level2;
michael@0: 
michael@0:             /* copy the folding result to fold2[] */
michael@0:             if(length<=UCASE_MAX_STRING_LENGTH) {
michael@0:                 u_memcpy(fold2, p, length);
michael@0:             } else {
michael@0:                 int32_t i=0;
michael@0:                 U16_APPEND_UNSAFE(fold2, i, length);
michael@0:                 length=i;
michael@0:             }
michael@0: 
michael@0:             /* set next level pointers to case folding */
michael@0:             start2=s2=fold2;
michael@0:             limit2=fold2+length;
michael@0: 
michael@0:             /* get ready to read from decomposition, continue with loop */
michael@0:             c2=-1;
michael@0:             continue;
michael@0:         }
michael@0: 
michael@0:         if( level1<2 && (options&_COMPARE_EQUIV) &&
michael@0:             0!=(p=nfcImpl->getDecomposition((UChar32)cp1, decomp1, length))
michael@0:         ) {
michael@0:             /* cp1 decomposes into p[length] */
michael@0:             if(U_IS_SURROGATE(c1)) {
michael@0:                 if(U_IS_SURROGATE_LEAD(c1)) {
michael@0:                     /* advance beyond source surrogate pair if it decomposes */
michael@0:                     ++s1;
michael@0:                 } else /* isTrail(c1) */ {
michael@0:                     /*
michael@0:                      * we got a supplementary code point when hitting its trail surrogate,
michael@0:                      * therefore the lead surrogate must have been the same as in the other string;
michael@0:                      * compare this decomposition with the lead surrogate in the other string
michael@0:                      * remember that this simulates bulk text replacement:
michael@0:                      * the decomposition would replace the entire code point
michael@0:                      */
michael@0:                     --s2;
michael@0:                     c2=*(s2-1);
michael@0:                 }
michael@0:             }
michael@0: 
michael@0:             /* push current level pointers */
michael@0:             stack1[level1].start=start1;
michael@0:             stack1[level1].s=s1;
michael@0:             stack1[level1].limit=limit1;
michael@0:             ++level1;
michael@0: 
michael@0:             /* set empty intermediate level if skipped */
michael@0:             if(level1<2) {
michael@0:                 stack1[level1++].start=NULL;
michael@0:             }
michael@0: 
michael@0:             /* set next level pointers to decomposition */
michael@0:             start1=s1=p;
michael@0:             limit1=p+length;
michael@0: 
michael@0:             /* get ready to read from decomposition, continue with loop */
michael@0:             c1=-1;
michael@0:             continue;
michael@0:         }
michael@0: 
michael@0:         if( level2<2 && (options&_COMPARE_EQUIV) &&
michael@0:             0!=(p=nfcImpl->getDecomposition((UChar32)cp2, decomp2, length))
michael@0:         ) {
michael@0:             /* cp2 decomposes into p[length] */
michael@0:             if(U_IS_SURROGATE(c2)) {
michael@0:                 if(U_IS_SURROGATE_LEAD(c2)) {
michael@0:                     /* advance beyond source surrogate pair if it decomposes */
michael@0:                     ++s2;
michael@0:                 } else /* isTrail(c2) */ {
michael@0:                     /*
michael@0:                      * we got a supplementary code point when hitting its trail surrogate,
michael@0:                      * therefore the lead surrogate must have been the same as in the other string;
michael@0:                      * compare this decomposition with the lead surrogate in the other string
michael@0:                      * remember that this simulates bulk text replacement:
michael@0:                      * the decomposition would replace the entire code point
michael@0:                      */
michael@0:                     --s1;
michael@0:                     c1=*(s1-1);
michael@0:                 }
michael@0:             }
michael@0: 
michael@0:             /* push current level pointers */
michael@0:             stack2[level2].start=start2;
michael@0:             stack2[level2].s=s2;
michael@0:             stack2[level2].limit=limit2;
michael@0:             ++level2;
michael@0: 
michael@0:             /* set empty intermediate level if skipped */
michael@0:             if(level2<2) {
michael@0:                 stack2[level2++].start=NULL;
michael@0:             }
michael@0: 
michael@0:             /* set next level pointers to decomposition */
michael@0:             start2=s2=p;
michael@0:             limit2=p+length;
michael@0: 
michael@0:             /* get ready to read from decomposition, continue with loop */
michael@0:             c2=-1;
michael@0:             continue;
michael@0:         }
michael@0: 
michael@0:         /*
michael@0:          * no decomposition/case folding, max level for both sides:
michael@0:          * return difference result
michael@0:          *
michael@0:          * code point order comparison must not just return cp1-cp2
michael@0:          * because when single surrogates are present then the surrogate pairs
michael@0:          * that formed cp1 and cp2 may be from different string indexes
michael@0:          *
michael@0:          * example: { d800 d800 dc01 } vs. { d800 dc00 }, compare at second code units
michael@0:          * c1=d800 cp1=10001 c2=dc00 cp2=10000
michael@0:          * cp1-cp2>0 but c1-c2<0 and in fact in UTF-32 it is { d800 10001 } < { 10000 }
michael@0:          *
michael@0:          * therefore, use same fix-up as in ustring.c/uprv_strCompare()
michael@0:          * except: uprv_strCompare() fetches c=*s while this functions fetches c=*s++
michael@0:          * so we have slightly different pointer/start/limit comparisons here
michael@0:          */
michael@0: 
michael@0:         if(c1>=0xd800 && c2>=0xd800 && (options&U_COMPARE_CODE_POINT_ORDER)) {
michael@0:             /* subtract 0x2800 from BMP code points to make them smaller than supplementary ones */
michael@0:             if(
michael@0:                 (c1<=0xdbff && s1!=limit1 && U16_IS_TRAIL(*s1)) ||
michael@0:                 (U16_IS_TRAIL(c1) && start1!=(s1-1) && U16_IS_LEAD(*(s1-2)))
michael@0:             ) {
michael@0:                 /* part of a surrogate pair, leave >=d800 */
michael@0:             } else {
michael@0:                 /* BMP code point - may be surrogate code point - make <d800 */
michael@0:                 c1-=0x2800;
michael@0:             }
michael@0: 
michael@0:             if(
michael@0:                 (c2<=0xdbff && s2!=limit2 && U16_IS_TRAIL(*s2)) ||
michael@0:                 (U16_IS_TRAIL(c2) && start2!=(s2-1) && U16_IS_LEAD(*(s2-2)))
michael@0:             ) {
michael@0:                 /* part of a surrogate pair, leave >=d800 */
michael@0:             } else {
michael@0:                 /* BMP code point - may be surrogate code point - make <d800 */
michael@0:                 c2-=0x2800;
michael@0:             }
michael@0:         }
michael@0: 
michael@0:         return c1-c2;
michael@0:     }
michael@0: }
michael@0: 
michael@0: static
michael@0: UBool _normalize(const Normalizer2 *n2, const UChar *s, int32_t length,
michael@0:                 UnicodeString &normalized, UErrorCode *pErrorCode) {
michael@0:     UnicodeString str(length<0, s, length);
michael@0: 
michael@0:     // check if s fulfill the conditions
michael@0:     int32_t spanQCYes=n2->spanQuickCheckYes(str, *pErrorCode);
michael@0:     if (U_FAILURE(*pErrorCode)) {
michael@0:         return FALSE;
michael@0:     }
michael@0:     /*
michael@0:      * ICU 2.4 had a further optimization:
michael@0:      * If both strings were not in FCD, then they were both NFD'ed,
michael@0:      * and the _COMPARE_EQUIV option was turned off.
michael@0:      * It is not entirely clear that this is valid with the current
michael@0:      * definition of the canonical caseless match.
michael@0:      * Therefore, ICU 2.6 removes that optimization.
michael@0:      */
michael@0:     if(spanQCYes<str.length()) {
michael@0:         UnicodeString unnormalized=str.tempSubString(spanQCYes);
michael@0:         normalized.setTo(FALSE, str.getBuffer(), spanQCYes);
michael@0:         n2->normalizeSecondAndAppend(normalized, unnormalized, *pErrorCode);
michael@0:         if (U_SUCCESS(*pErrorCode)) {
michael@0:             return TRUE;
michael@0:         }
michael@0:     }
michael@0:     return FALSE;
michael@0: }
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: unorm_compare(const UChar *s1, int32_t length1,
michael@0:               const UChar *s2, int32_t length2,
michael@0:               uint32_t options,
michael@0:               UErrorCode *pErrorCode) {
michael@0:     /* argument checking */
michael@0:     if(U_FAILURE(*pErrorCode)) {
michael@0:         return 0;
michael@0:     }
michael@0:     if(s1==0 || length1<-1 || s2==0 || length2<-1) {
michael@0:         *pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
michael@0:         return 0;
michael@0:     }
michael@0: 
michael@0:     UnicodeString fcd1, fcd2;
michael@0:     int32_t normOptions=(int32_t)(options>>UNORM_COMPARE_NORM_OPTIONS_SHIFT);
michael@0:     options|=_COMPARE_EQUIV;
michael@0: 
michael@0:     /*
michael@0:      * UAX #21 Case Mappings, as fixed for Unicode version 4
michael@0:      * (see Jitterbug 2021), defines a canonical caseless match as
michael@0:      *
michael@0:      * A string X is a canonical caseless match
michael@0:      * for a string Y if and only if
michael@0:      * NFD(toCasefold(NFD(X))) = NFD(toCasefold(NFD(Y)))
michael@0:      *
michael@0:      * For better performance, we check for FCD (or let the caller tell us that
michael@0:      * both strings are in FCD) for the inner normalization.
michael@0:      * BasicNormalizerTest::FindFoldFCDExceptions() makes sure that
michael@0:      * case-folding preserves the FCD-ness of a string.
michael@0:      * The outer normalization is then only performed by unorm_cmpEquivFold()
michael@0:      * when there is a difference.
michael@0:      *
michael@0:      * Exception: When using the Turkic case-folding option, we do perform
michael@0:      * full NFD first. This is because in the Turkic case precomposed characters
michael@0:      * with 0049 capital I or 0069 small i fold differently whether they
michael@0:      * are first decomposed or not, so an FCD check - a check only for
michael@0:      * canonical order - is not sufficient.
michael@0:      */
michael@0:     if(!(options&UNORM_INPUT_IS_FCD) || (options&U_FOLD_CASE_EXCLUDE_SPECIAL_I)) {
michael@0:         const Normalizer2 *n2;
michael@0:         if(options&U_FOLD_CASE_EXCLUDE_SPECIAL_I) {
michael@0:             n2=Normalizer2Factory::getNFDInstance(*pErrorCode);
michael@0:         } else {
michael@0:             n2=Normalizer2Factory::getFCDInstance(*pErrorCode);
michael@0:         }
michael@0:         if (U_FAILURE(*pErrorCode)) {
michael@0:             return 0;
michael@0:         }
michael@0: 
michael@0:         if(normOptions&UNORM_UNICODE_3_2) {
michael@0:             const UnicodeSet *uni32=uniset_getUnicode32Instance(*pErrorCode);
michael@0:             FilteredNormalizer2 fn2(*n2, *uni32);
michael@0:             if(_normalize(&fn2, s1, length1, fcd1, pErrorCode)) {
michael@0:                 s1=fcd1.getBuffer();
michael@0:                 length1=fcd1.length();
michael@0:             }
michael@0:             if(_normalize(&fn2, s2, length2, fcd2, pErrorCode)) {
michael@0:                 s2=fcd2.getBuffer();
michael@0:                 length2=fcd2.length();
michael@0:             }
michael@0:         } else {
michael@0:             if(_normalize(n2, s1, length1, fcd1, pErrorCode)) {
michael@0:                 s1=fcd1.getBuffer();
michael@0:                 length1=fcd1.length();
michael@0:             }
michael@0:             if(_normalize(n2, s2, length2, fcd2, pErrorCode)) {
michael@0:                 s2=fcd2.getBuffer();
michael@0:                 length2=fcd2.length();
michael@0:             }
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     if(U_SUCCESS(*pErrorCode)) {
michael@0:         return unorm_cmpEquivFold(s1, length1, s2, length2, options, pErrorCode);
michael@0:     } else {
michael@0:         return 0;
michael@0:     }
michael@0: }
michael@0: 
michael@0: #endif /* #if !UCONFIG_NO_NORMALIZATION */