blob: 173166af622dc131b7f19372e2a035b15887f6df [file] [log] [blame]
/*
*******************************************************************************
* Copyright (C) 1996-2010, International Business Machines Corporation and *
* others. All Rights Reserved. *
*******************************************************************************
*/
/*
* File coleitr.cpp
*
*
*
* Created by: Helena Shih
*
* Modification History:
*
* Date Name Description
*
* 6/23/97 helena Adding comments to make code more readable.
* 08/03/98 erm Synched with 1.2 version of CollationElementIterator.java
* 12/10/99 aliu Ported Thai collation support from Java.
* 01/25/01 swquek Modified to a C++ wrapper calling C APIs (ucoliter.h)
* 02/19/01 swquek Removed CollationElementsIterator() since it is
* private constructor and no calls are made to it
*/
#include "unicode/utypes.h"
#if !UCONFIG_NO_COLLATION
#include "unicode/coleitr.h"
#include "unicode/ustring.h"
#include "ucol_imp.h"
#include "cmemory.h"
/* Constants --------------------------------------------------------------- */
U_NAMESPACE_BEGIN
UOBJECT_DEFINE_RTTI_IMPLEMENTATION(CollationElementIterator)
/* CollationElementIterator public constructor/destructor ------------------ */
CollationElementIterator::CollationElementIterator(
const CollationElementIterator& other)
: UObject(other), isDataOwned_(TRUE)
{
UErrorCode status = U_ZERO_ERROR;
m_data_ = ucol_openElements(other.m_data_->iteratordata_.coll, NULL, 0,
&status);
*this = other;
}
CollationElementIterator::~CollationElementIterator()
{
if (isDataOwned_) {
ucol_closeElements(m_data_);
}
}
/* CollationElementIterator public methods --------------------------------- */
int32_t CollationElementIterator::getOffset() const
{
return ucol_getOffset(m_data_);
}
/**
* Get the ordering priority of the next character in the string.
* @return the next character's ordering. Returns NULLORDER if an error has
* occured or if the end of string has been reached
*/
int32_t CollationElementIterator::next(UErrorCode& status)
{
return ucol_next(m_data_, &status);
}
UBool CollationElementIterator::operator!=(
const CollationElementIterator& other) const
{
return !(*this == other);
}
UBool CollationElementIterator::operator==(
const CollationElementIterator& that) const
{
if (this == &that || m_data_ == that.m_data_) {
return TRUE;
}
// option comparison
if (m_data_->iteratordata_.coll != that.m_data_->iteratordata_.coll)
{
return FALSE;
}
// the constructor and setText always sets a length
// and we only compare the string not the contents of the normalization
// buffer
int thislength = (int)(m_data_->iteratordata_.endp - m_data_->iteratordata_.string);
int thatlength = (int)(that.m_data_->iteratordata_.endp - that.m_data_->iteratordata_.string);
if (thislength != thatlength) {
return FALSE;
}
if (uprv_memcmp(m_data_->iteratordata_.string,
that.m_data_->iteratordata_.string,
thislength * U_SIZEOF_UCHAR) != 0) {
return FALSE;
}
if (getOffset() != that.getOffset()) {
return FALSE;
}
// checking normalization buffer
if ((m_data_->iteratordata_.flags & UCOL_ITER_HASLEN) == 0) {
if ((that.m_data_->iteratordata_.flags & UCOL_ITER_HASLEN) != 0) {
return FALSE;
}
// both are in the normalization buffer
if (m_data_->iteratordata_.pos
- m_data_->iteratordata_.writableBuffer.getBuffer()
!= that.m_data_->iteratordata_.pos
- that.m_data_->iteratordata_.writableBuffer.getBuffer()) {
// not in the same position in the normalization buffer
return FALSE;
}
}
else if ((that.m_data_->iteratordata_.flags & UCOL_ITER_HASLEN) == 0) {
return FALSE;
}
// checking ce position
return (m_data_->iteratordata_.CEpos - m_data_->iteratordata_.CEs)
== (that.m_data_->iteratordata_.CEpos
- that.m_data_->iteratordata_.CEs);
}
/**
* Get the ordering priority of the previous collation element in the string.
* @param status the error code status.
* @return the previous element's ordering. Returns NULLORDER if an error has
* occured or if the start of string has been reached.
*/
int32_t CollationElementIterator::previous(UErrorCode& status)
{
return ucol_previous(m_data_, &status);
}
/**
* Resets the cursor to the beginning of the string.
*/
void CollationElementIterator::reset()
{
ucol_reset(m_data_);
}
void CollationElementIterator::setOffset(int32_t newOffset,
UErrorCode& status)
{
ucol_setOffset(m_data_, newOffset, &status);
}
/**
* Sets the source to the new source string.
*/
void CollationElementIterator::setText(const UnicodeString& source,
UErrorCode& status)
{
if (U_FAILURE(status)) {
return;
}
int32_t length = source.length();
UChar *string = NULL;
if (m_data_->isWritable && m_data_->iteratordata_.string != NULL) {
uprv_free((UChar *)m_data_->iteratordata_.string);
}
m_data_->isWritable = TRUE;
if (length > 0) {
string = (UChar *)uprv_malloc(U_SIZEOF_UCHAR * length);
/* test for NULL */
if (string == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
u_memcpy(string, source.getBuffer(), length);
}
else {
string = (UChar *)uprv_malloc(U_SIZEOF_UCHAR);
/* test for NULL */
if (string == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
*string = 0;
}
/* Free offsetBuffer before initializing it. */
ucol_freeOffsetBuffer(&(m_data_->iteratordata_));
uprv_init_collIterate(m_data_->iteratordata_.coll, string, length,
&m_data_->iteratordata_, &status);
m_data_->reset_ = TRUE;
}
// Sets the source to the new character iterator.
void CollationElementIterator::setText(CharacterIterator& source,
UErrorCode& status)
{
if (U_FAILURE(status))
return;
int32_t length = source.getLength();
UChar *buffer = NULL;
if (length == 0) {
buffer = (UChar *)uprv_malloc(U_SIZEOF_UCHAR);
/* test for NULL */
if (buffer == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
*buffer = 0;
}
else {
buffer = (UChar *)uprv_malloc(U_SIZEOF_UCHAR * length);
/* test for NULL */
if (buffer == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
/*
Using this constructor will prevent buffer from being removed when
string gets removed
*/
UnicodeString string;
source.getText(string);
u_memcpy(buffer, string.getBuffer(), length);
}
if (m_data_->isWritable && m_data_->iteratordata_.string != NULL) {
uprv_free((UChar *)m_data_->iteratordata_.string);
}
m_data_->isWritable = TRUE;
/* Free offsetBuffer before initializing it. */
ucol_freeOffsetBuffer(&(m_data_->iteratordata_));
uprv_init_collIterate(m_data_->iteratordata_.coll, buffer, length,
&m_data_->iteratordata_, &status);
m_data_->reset_ = TRUE;
}
int32_t CollationElementIterator::strengthOrder(int32_t order) const
{
UCollationStrength s = ucol_getStrength(m_data_->iteratordata_.coll);
// Mask off the unwanted differences.
if (s == UCOL_PRIMARY) {
order &= RuleBasedCollator::PRIMARYDIFFERENCEONLY;
}
else if (s == UCOL_SECONDARY) {
order &= RuleBasedCollator::SECONDARYDIFFERENCEONLY;
}
return order;
}
/* CollationElementIterator private constructors/destructors --------------- */
/**
* This is the "real" constructor for this class; it constructs an iterator
* over the source text using the specified collator
*/
CollationElementIterator::CollationElementIterator(
const UnicodeString& sourceText,
const RuleBasedCollator* order,
UErrorCode& status)
: isDataOwned_(TRUE)
{
if (U_FAILURE(status)) {
return;
}
int32_t length = sourceText.length();
UChar *string = NULL;
if (length > 0) {
string = (UChar *)uprv_malloc(U_SIZEOF_UCHAR * length);
/* test for NULL */
if (string == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
/*
Using this constructor will prevent buffer from being removed when
string gets removed
*/
u_memcpy(string, sourceText.getBuffer(), length);
}
else {
string = (UChar *)uprv_malloc(U_SIZEOF_UCHAR);
/* test for NULL */
if (string == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
*string = 0;
}
m_data_ = ucol_openElements(order->ucollator, string, length, &status);
/* Test for buffer overflows */
if (U_FAILURE(status)) {
return;
}
m_data_->isWritable = TRUE;
}
/**
* This is the "real" constructor for this class; it constructs an iterator over
* the source text using the specified collator
*/
CollationElementIterator::CollationElementIterator(
const CharacterIterator& sourceText,
const RuleBasedCollator* order,
UErrorCode& status)
: isDataOwned_(TRUE)
{
if (U_FAILURE(status))
return;
// **** should I just drop this test? ****
/*
if ( sourceText.endIndex() != 0 )
{
// A CollationElementIterator is really a two-layered beast.
// Internally it uses a Normalizer to munge the source text into a form
// where all "composed" Unicode characters (such as \u00FC) are split into a
// normal character and a combining accent character.
// Afterward, CollationElementIterator does its own processing to handle
// expanding and contracting collation sequences, ignorables, and so on.
Normalizer::EMode decomp = order->getStrength() == Collator::IDENTICAL
? Normalizer::NO_OP : order->getDecomposition();
text = new Normalizer(sourceText, decomp);
if (text == NULL)
status = U_MEMORY_ALLOCATION_ERROR;
}
*/
int32_t length = sourceText.getLength();
UChar *buffer;
if (length > 0) {
buffer = (UChar *)uprv_malloc(U_SIZEOF_UCHAR * length);
/* test for NULL */
if (buffer == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
/*
Using this constructor will prevent buffer from being removed when
string gets removed
*/
UnicodeString string(buffer, length, length);
((CharacterIterator &)sourceText).getText(string);
const UChar *temp = string.getBuffer();
u_memcpy(buffer, temp, length);
}
else {
buffer = (UChar *)uprv_malloc(U_SIZEOF_UCHAR);
/* test for NULL */
if (buffer == NULL) {
status = U_MEMORY_ALLOCATION_ERROR;
return;
}
*buffer = 0;
}
m_data_ = ucol_openElements(order->ucollator, buffer, length, &status);
/* Test for buffer overflows */
if (U_FAILURE(status)) {
return;
}
m_data_->isWritable = TRUE;
}
/* CollationElementIterator protected methods ----------------------------- */
const CollationElementIterator& CollationElementIterator::operator=(
const CollationElementIterator& other)
{
if (this != &other)
{
UCollationElements *ucolelem = this->m_data_;
UCollationElements *otherucolelem = other.m_data_;
collIterate *coliter = &(ucolelem->iteratordata_);
collIterate *othercoliter = &(otherucolelem->iteratordata_);
int length = 0;
// checking only UCOL_ITER_HASLEN is not enough here as we may be in
// the normalization buffer
length = (int)(othercoliter->endp - othercoliter->string);
ucolelem->reset_ = otherucolelem->reset_;
ucolelem->isWritable = TRUE;
/* create a duplicate of string */
if (length > 0) {
coliter->string = (UChar *)uprv_malloc(length * U_SIZEOF_UCHAR);
if(coliter->string != NULL) {
uprv_memcpy((UChar *)coliter->string, othercoliter->string,
length * U_SIZEOF_UCHAR);
} else { // Error: couldn't allocate memory. No copying should be done
length = 0;
}
}
else {
coliter->string = NULL;
}
/* start and end of string */
coliter->endp = coliter->string + length;
/* handle writable buffer here */
if (othercoliter->flags & UCOL_ITER_INNORMBUF) {
coliter->writableBuffer = othercoliter->writableBuffer;
coliter->writableBuffer.getTerminatedBuffer();
}
/* current position */
if (othercoliter->pos >= othercoliter->string &&
othercoliter->pos <= othercoliter->endp)
{
coliter->pos = coliter->string +
(othercoliter->pos - othercoliter->string);
}
else {
coliter->pos = coliter->writableBuffer.getTerminatedBuffer() +
(othercoliter->pos - othercoliter->writableBuffer.getBuffer());
}
/* CE buffer */
int32_t CEsize;
if (coliter->extendCEs) {
uprv_memcpy(coliter->CEs, othercoliter->CEs, sizeof(uint32_t) * UCOL_EXPAND_CE_BUFFER_SIZE);
CEsize = sizeof(othercoliter->extendCEs);
if (CEsize > 0) {
othercoliter->extendCEs = (uint32_t *)uprv_malloc(CEsize);
uprv_memcpy(coliter->extendCEs, othercoliter->extendCEs, CEsize);
}
coliter->toReturn = coliter->extendCEs +
(othercoliter->toReturn - othercoliter->extendCEs);
coliter->CEpos = coliter->extendCEs + CEsize;
} else {
CEsize = (int32_t)(othercoliter->CEpos - othercoliter->CEs);
if (CEsize > 0) {
uprv_memcpy(coliter->CEs, othercoliter->CEs, CEsize);
}
coliter->toReturn = coliter->CEs +
(othercoliter->toReturn - othercoliter->CEs);
coliter->CEpos = coliter->CEs + CEsize;
}
if (othercoliter->fcdPosition != NULL) {
coliter->fcdPosition = coliter->string +
(othercoliter->fcdPosition
- othercoliter->string);
}
else {
coliter->fcdPosition = NULL;
}
coliter->flags = othercoliter->flags/*| UCOL_ITER_HASLEN*/;
coliter->origFlags = othercoliter->origFlags;
coliter->coll = othercoliter->coll;
this->isDataOwned_ = TRUE;
}
return *this;
}
U_NAMESPACE_END
#endif /* #if !UCONFIG_NO_COLLATION */
/* eof */