/* ****************************************************************************** * Copyright (C) 2001, International Business Machines * Corporation and others. All Rights Reserved. ****************************************************************************** * * File ucoleitr.cpp * * Modification History: * * Date Name Description * 02/15/2001 synwee Modified all methods to process its own function * instead of calling the equivalent c++ api (coleitr.h) ******************************************************************************/ #include "unicode/ucoleitr.h" #include "unicode/ustring.h" #include "ucol_imp.h" #include "unicode/sortkey.h" #include "cmemory.h" U_NAMESPACE_USE #define BUFFER_LENGTH 100 typedef struct collIterate collIterator; /* public methods ---------------------------------------------------- */ /** * Since this is going to be deprecated, I'll leave it as it is */ U_CAPI int32_t U_EXPORT2 ucol_keyHashCode(const uint8_t *key, int32_t length) { CollationKey newKey(key, length); return newKey.hashCode(); } U_CAPI UCollationElements* U_EXPORT2 ucol_openElements(const UCollator *coll, const UChar *text, int32_t textLength, UErrorCode *status) { UCollationElements *result; if (U_FAILURE(*status)) { return NULL; } result = (UCollationElements *)uprv_malloc(sizeof(UCollationElements)); result->reset_ = TRUE; result->isWritable = FALSE; if (text == NULL) { textLength = 0; } init_collIterate(coll, text, textLength, &result->iteratordata_); return result; } U_CAPI void U_EXPORT2 ucol_closeElements(UCollationElements *elems) { collIterate *ci = &elems->iteratordata_; if (ci->writableBuffer != ci->stackWritableBuffer) { uprv_free(ci->writableBuffer); } if (elems->isWritable && elems->iteratordata_.string != NULL) { uprv_free(elems->iteratordata_.string); } uprv_free(elems); } U_CAPI void U_EXPORT2 ucol_reset(UCollationElements *elems) { collIterate *ci = &(elems->iteratordata_); elems->reset_ = TRUE; ci->pos = ci->string; if ((ci->flags & UCOL_ITER_HASLEN) == 0 || ci->endp == NULL) { ci->endp = ci->string + u_strlen(ci->string); } ci->CEpos = ci->toReturn = ci->CEs; ci->flags = UCOL_ITER_HASLEN; if (ci->coll->normalizationMode == UCOL_ON) { ci->flags |= UCOL_ITER_NORM; } if (ci->stackWritableBuffer != ci->writableBuffer) { uprv_free(ci->writableBuffer); ci->writableBuffer = ci->stackWritableBuffer; ci->writableBufSize = UCOL_WRITABLE_BUFFER_SIZE; } ci->fcdPosition = NULL; } U_CAPI int32_t U_EXPORT2 ucol_next(UCollationElements *elems, UErrorCode *status) { uint32_t result; if (U_FAILURE(*status)) { return UCOL_NULLORDER; } elems->reset_ = FALSE; result = ucol_getNextCE(elems->iteratordata_.coll, &elems->iteratordata_, status); if (result == UCOL_NO_MORE_CES) { result = UCOL_NULLORDER; } return result; } U_CAPI int32_t U_EXPORT2 ucol_previous(UCollationElements *elems, UErrorCode *status) { if(U_FAILURE(*status)) { return UCOL_NULLORDER; } else { uint32_t result; if (elems->reset_ && (elems->iteratordata_.pos == elems->iteratordata_.string)) { if (elems->iteratordata_.endp == NULL) { elems->iteratordata_.endp = elems->iteratordata_.string + u_strlen(elems->iteratordata_.string); elems->iteratordata_.flags |= UCOL_ITER_HASLEN; } elems->iteratordata_.pos = elems->iteratordata_.endp; elems->iteratordata_.fcdPosition = elems->iteratordata_.endp; } elems->reset_ = FALSE; result = ucol_getPrevCE(elems->iteratordata_.coll, &(elems->iteratordata_), status); if (result == UCOL_NO_MORE_CES) { result = UCOL_NULLORDER; } return result; } } U_CAPI int32_t U_EXPORT2 ucol_getMaxExpansion(const UCollationElements *elems, int32_t order) { uint8_t result; UCOL_GETMAXEXPANSION(elems->iteratordata_.coll, (uint32_t)order, result); return result; } U_CAPI void U_EXPORT2 ucol_setText( UCollationElements *elems, const UChar *text, int32_t textLength, UErrorCode *status) { if (U_FAILURE(*status)) { return; } if (elems->isWritable && elems->iteratordata_.string != NULL) { uprv_free(elems->iteratordata_.string); } if (text == NULL) { textLength = 0; } elems->isWritable = FALSE; init_collIterate(elems->iteratordata_.coll, text, textLength, &elems->iteratordata_); elems->reset_ = TRUE; } U_CAPI int32_t U_EXPORT2 ucol_getOffset(const UCollationElements *elems) { const collIterate *ci = &(elems->iteratordata_); // while processing characters in normalization buffer getOffset will // return the next non-normalized character. // should be inline with the old implementation since the old codes uses // nextDecomp in normalizer which also decomposes the string till the // first base character is found. if (ci->flags & UCOL_ITER_INNORMBUF) { if (ci->fcdPosition == NULL) { return 0; } return (int32_t)(ci->fcdPosition - ci->string); } else { return (int32_t)(ci->pos - ci->string); } } U_CAPI void U_EXPORT2 ucol_setOffset(UCollationElements *elems, int32_t offset, UErrorCode *status) { if (U_FAILURE(*status)) { return; } // this methods will clean up any use of the writable buffer and points to // the original string collIterate *ci = &(elems->iteratordata_); ci->pos = ci->string + offset; ci->CEpos = ci->toReturn = ci->CEs; if (ci->flags & UCOL_ITER_INNORMBUF) { ci->flags = ci->origFlags; } if ((ci->flags & UCOL_ITER_HASLEN) == 0) { ci->endp = ci->string + u_strlen(ci->string); ci->flags |= UCOL_ITER_HASLEN; } ci->fcdPosition = NULL; }