+// © 2016 and later: Unicode, Inc. and others.
+// License & terms of use: http://www.unicode.org/copyright.html
/*
***************************************************************************
-* Copyright (C) 1999-2004 International Business Machines Corporation *
-* and others. All rights reserved. *
+* Copyright (C) 1999-2016 International Business Machines Corporation
+* and others. All rights reserved.
***************************************************************************
*/
//
-// file: rbbi.c Contains the implementation of the rule based break iterator
+// file: rbbi.cpp Contains the implementation of the rule based break iterator
// runtime engine and the API implementation for
// class RuleBasedBreakIterator
//
+#include "utypeinfo.h" // for 'typeid' to work
+
#include "unicode/utypes.h"
#if !UCONFIG_NO_BREAK_ITERATION
#include "unicode/rbbi.h"
#include "unicode/schriter.h"
-#include "unicode/udata.h"
+#include "unicode/uchriter.h"
#include "unicode/uclean.h"
-#include "rbbidata.h"
-#include "rbbirb.h"
+#include "unicode/udata.h"
+
+#include "brkeng.h"
+#include "ucln_cmn.h"
#include "cmemory.h"
#include "cstring.h"
-
+#include "rbbidata.h"
+#include "rbbi_cache.h"
+#include "rbbirb.h"
#include "uassert.h"
+#include "umutex.h"
+#include "uvectr32.h"
+
+// if U_LOCAL_SERVICE_HOOK is defined, then localsvc.cpp is expected to be included.
+#if U_LOCAL_SERVICE_HOOK
+#include "localsvc.h"
+#endif
+
+// Apple specific
+//#include <os/log.h>
+
+#ifdef RBBI_DEBUG
+static UBool gTrace = FALSE;
+#endif
U_NAMESPACE_BEGIN
+// The state number of the starting state
+constexpr int32_t START_STATE = 1;
-static const int16_t START_STATE = 1; // The state number of the starting state
-static const int16_t STOP_STATE = 0; // The state-transition value indicating "stop"
+// The state-transition value indicating "stop"
+constexpr int32_t STOP_STATE = 0;
UOBJECT_DEFINE_RTTI_IMPLEMENTATION(RuleBasedBreakIterator)
* tables object that is passed in as a parameter.
*/
RuleBasedBreakIterator::RuleBasedBreakIterator(RBBIDataHeader* data, UErrorCode &status)
+ : fSCharIter(UnicodeString())
{
- init();
+ init(status);
fData = new RBBIDataWrapper(data, status); // status checked in constructor
if (U_FAILURE(status)) {return;}
if(fData == 0) {
}
}
+//
+// Construct from precompiled binary rules (tables). This constructor is public API,
+// taking the rules as a (const uint8_t *) to match the type produced by getBinaryRules().
+//
+RuleBasedBreakIterator::RuleBasedBreakIterator(const uint8_t *compiledRules,
+ uint32_t ruleLength,
+ UErrorCode &status)
+ : fSCharIter(UnicodeString())
+{
+ init(status);
+ if (U_FAILURE(status)) {
+ return;
+ }
+ if (compiledRules == NULL || ruleLength < sizeof(RBBIDataHeader)) {
+ status = U_ILLEGAL_ARGUMENT_ERROR;
+ return;
+ }
+ const RBBIDataHeader *data = (const RBBIDataHeader *)compiledRules;
+ if (data->fLength > ruleLength) {
+ status = U_ILLEGAL_ARGUMENT_ERROR;
+ return;
+ }
+ fData = new RBBIDataWrapper(data, RBBIDataWrapper::kDontAdopt, status);
+ if (U_FAILURE(status)) {return;}
+ if(fData == 0) {
+ status = U_MEMORY_ALLOCATION_ERROR;
+ return;
+ }
+}
+
+
//-------------------------------------------------------------------------------
//
// Constructor from a UDataMemory handle to precompiled break rules
//
//-------------------------------------------------------------------------------
RuleBasedBreakIterator::RuleBasedBreakIterator(UDataMemory* udm, UErrorCode &status)
+ : fSCharIter(UnicodeString())
{
- init();
+ init(status);
fData = new RBBIDataWrapper(udm, status); // status checked in constructor
if (U_FAILURE(status)) {return;}
if(fData == 0) {
RuleBasedBreakIterator::RuleBasedBreakIterator( const UnicodeString &rules,
UParseError &parseError,
UErrorCode &status)
+ : fSCharIter(UnicodeString())
{
- u_init(&status); // Just in case ICU is not yet initialized
- init();
+ init(status);
if (U_FAILURE(status)) {return;}
RuleBasedBreakIterator *bi = (RuleBasedBreakIterator *)
- RBBIRuleBuilder::createRuleBasedBreakIterator(rules, parseError, status);
+ RBBIRuleBuilder::createRuleBasedBreakIterator(rules, &parseError, status);
// Note: This is a bit awkward. The RBBI ruleBuilder has a factory method that
// creates and returns a complete RBBI. From here, in a constructor, we
// can't just return the object created by the builder factory, hence
// Used when creating a RuleBasedBreakIterator from a set
// of rules.
//-------------------------------------------------------------------------------
-RuleBasedBreakIterator::RuleBasedBreakIterator() {
- init();
+RuleBasedBreakIterator::RuleBasedBreakIterator()
+ : fSCharIter(UnicodeString())
+{
+ UErrorCode status = U_ZERO_ERROR;
+ init(status);
}
//
//-------------------------------------------------------------------------------
RuleBasedBreakIterator::RuleBasedBreakIterator(const RuleBasedBreakIterator& other)
-: BreakIterator(other)
+: BreakIterator(other),
+ fSCharIter(UnicodeString())
{
- this->init();
+ UErrorCode status = U_ZERO_ERROR;
+ this->init(status);
*this = other;
}
* Destructor
*/
RuleBasedBreakIterator::~RuleBasedBreakIterator() {
- delete fText;
- fText = NULL;
+ if (fCharIter != &fSCharIter) {
+ // fCharIter was adopted from the outside.
+ delete fCharIter;
+ }
+ fCharIter = NULL;
+
+ utext_close(&fText);
+
if (fData != NULL) {
fData->removeReference();
fData = NULL;
}
+ delete fBreakCache;
+ fBreakCache = NULL;
+
+ delete fDictionaryCache;
+ fDictionaryCache = NULL;
+
+ delete fLanguageBreakEngines;
+ fLanguageBreakEngines = NULL;
+
+ delete fUnhandledBreakEngine;
+ fUnhandledBreakEngine = NULL;
+
+ delete [] fLatin1Cat;
+ fLatin1Cat = NULL;
}
/**
if (this == &that) {
return *this;
}
- delete fText;
- fText = NULL;
- if (that.fText != NULL) {
- fText = that.fText->clone();
+ BreakIterator::operator=(that);
+ fLineWordOpts = that.fLineWordOpts;
+
+ if (fLanguageBreakEngines != NULL) {
+ delete fLanguageBreakEngines;
+ fLanguageBreakEngines = NULL; // Just rebuild for now
+ }
+ // TODO: clone fLanguageBreakEngines from "that"
+ UErrorCode status = U_ZERO_ERROR;
+ utext_clone(&fText, &that.fText, FALSE, TRUE, &status);
+
+ if (fCharIter != &fSCharIter) {
+ delete fCharIter;
+ }
+ fCharIter = &fSCharIter;
+
+ if (that.fCharIter != NULL && that.fCharIter != &that.fSCharIter) {
+ // This is a little bit tricky - it will intially appear that
+ // this->fCharIter is adopted, even if that->fCharIter was
+ // not adopted. That's ok.
+ fCharIter = that.fCharIter->clone();
+ }
+ fSCharIter = that.fSCharIter;
+ if (fCharIter == NULL) {
+ fCharIter = &fSCharIter;
}
if (fData != NULL) {
if (that.fData != NULL) {
fData = that.fData->addReference();
}
- fTrace = that.fTrace;
+
+ delete [] fLatin1Cat;
+ fLatin1Cat = NULL;
+
+ fPosition = that.fPosition;
+ fRuleStatusIndex = that.fRuleStatusIndex;
+ fDone = that.fDone;
+
+ // TODO: both the dictionary and the main cache need to be copied.
+ // Current position could be within a dictionary range. Trying to continue
+ // the iteration without the caches present would go to the rules, with
+ // the assumption that the current position is on a rule boundary.
+ fBreakCache->reset(fPosition, fRuleStatusIndex);
+ fDictionaryCache->reset();
return *this;
}
// Initializes all fields, leaving the object in a consistent state.
//
//-----------------------------------------------------------------------------
-UBool RuleBasedBreakIterator::fTrace = FALSE;
-void RuleBasedBreakIterator::init() {
-
- fText = NULL;
+void RuleBasedBreakIterator::init(UErrorCode &status) {
+ fCharIter = NULL;
fData = NULL;
- fLastRuleStatusIndex = 0;
- fLastStatusIndexValid = TRUE;
+ fLatin1Cat = NULL;
+ fPosition = 0;
+ fRuleStatusIndex = 0;
+ fDone = false;
fDictionaryCharCount = 0;
+ fLanguageBreakEngines = NULL;
+ fUnhandledBreakEngine = NULL;
+ fBreakCache = NULL;
+ fDictionaryCache = NULL;
+
+ // Note: IBM xlC is unable to assign or initialize member fText from UTEXT_INITIALIZER.
+ // fText = UTEXT_INITIALIZER;
+ static const UText initializedUText = UTEXT_INITIALIZER;
+ uprv_memcpy(&fText, &initializedUText, sizeof(UText));
+
+ if (U_FAILURE(status)) {
+ return;
+ }
+
+ utext_openUChars(&fText, NULL, 0, &status);
+ fDictionaryCache = new DictionaryCache(this, status);
+ fBreakCache = new BreakCache(this, status);
+ if (U_SUCCESS(status) && (fDictionaryCache == NULL || fBreakCache == NULL)) {
+ status = U_MEMORY_ALLOCATION_ERROR;
+ }
#ifdef RBBI_DEBUG
static UBool debugInitDone = FALSE;
if (debugInitDone == FALSE) {
char *debugEnv = getenv("U_RBBIDEBUG");
if (debugEnv && uprv_strstr(debugEnv, "trace")) {
- fTrace = TRUE;
+ gTrace = TRUE;
}
debugInitDone = TRUE;
}
}
+void RuleBasedBreakIterator::initLatin1Cat(void) {
+ fLatin1Cat = new uint16_t[256];
+ for (UChar32 c = 0; c < 256; ++c) {
+ fLatin1Cat[c] = UTRIE2_GET16(fData->fTrie, c);
+ }
+}
//-----------------------------------------------------------------------------
//
*/
UBool
RuleBasedBreakIterator::operator==(const BreakIterator& that) const {
- UBool r = FALSE;
- if (that.getDynamicClassID() != getDynamicClassID()) {
- return r;
+ if (typeid(*this) != typeid(that)) {
+ return FALSE;
+ }
+ if (this == &that) {
+ return TRUE;
}
+ // The base class BreakIterator carries no state that participates in equality,
+ // and does not implement an equality function that would otherwise be
+ // checked at this point.
+
const RuleBasedBreakIterator& that2 = (const RuleBasedBreakIterator&) that;
- if (fText == that2.fText ||
- (fText != NULL && that2.fText != NULL && *that2.fText == *fText)) {
- if (that2.fData == fData ||
- (fData != NULL && that2.fData != NULL && *that2.fData == *fData)) {
- r = TRUE;
- }
+ if (that2.fLineWordOpts != fLineWordOpts) {
+ return FALSE;
+ }
+
+ if (!utext_equals(&fText, &that2.fText)) {
+ // The two break iterators are operating on different text,
+ // or have a different iteration position.
+ // Note that fText's position is always the same as the break iterator's position.
+ return FALSE;
+ };
+
+ if (!(fPosition == that2.fPosition &&
+ fRuleStatusIndex == that2.fRuleStatusIndex &&
+ fDone == that2.fDone)) {
+ return FALSE;
}
- return r;
+
+ if (that2.fData == fData ||
+ (fData != NULL && that2.fData != NULL && *that2.fData == *fData)) {
+ // The two break iterators are using the same rules.
+ return TRUE;
+ }
+ return FALSE;
}
/**
return hash;
}
-/**
- * Returns the description used to create this iterator
- */
-const UnicodeString&
-RuleBasedBreakIterator::getRules() const {
- if (fData != NULL) {
- return fData->getRuleSourceString();
- } else {
- static const UnicodeString *s;
- if (s == NULL) {
- // TODO: something more elegant here.
- // perhaps API should return the string by value.
- // Note: thread unsafe init & leak are semi-ok, better than
- // what was before. Sould be cleaned up, though.
- s = new UnicodeString;
- }
- return *s;
+
+void RuleBasedBreakIterator::setText(UText *ut, UErrorCode &status) {
+ if (U_FAILURE(status)) {
+ return;
}
+ fBreakCache->reset();
+ fDictionaryCache->reset();
+ utext_clone(&fText, ut, FALSE, TRUE, &status);
+
+ // Set up a dummy CharacterIterator to be returned if anyone
+ // calls getText(). With input from UText, there is no reasonable
+ // way to return a characterIterator over the actual input text.
+ // Return one over an empty string instead - this is the closest
+ // we can come to signaling a failure.
+ // (GetText() is obsolete, this failure is sort of OK)
+ fSCharIter.setText(UnicodeString());
+
+ if (fCharIter != &fSCharIter) {
+ // existing fCharIter was adopted from the outside. Delete it now.
+ delete fCharIter;
+ }
+ fCharIter = &fSCharIter;
+
+ this->first();
}
+
+UText *RuleBasedBreakIterator::getUText(UText *fillIn, UErrorCode &status) const {
+ UText *result = utext_clone(fillIn, &fText, FALSE, TRUE, &status);
+ return result;
+}
+
+
//=======================================================================
// BreakIterator overrides
//=======================================================================
/**
- * Return a CharacterIterator over the text being analyzed. This version
- * of this method returns the actual CharacterIterator we're using internally.
- * Changing the state of this iterator can have undefined consequences. If
- * you need to change it, clone it first.
- * @return An iterator over the text being analyzed.
+ * Return a CharacterIterator over the text being analyzed.
*/
-const CharacterIterator&
+CharacterIterator&
RuleBasedBreakIterator::getText() const {
- RuleBasedBreakIterator* nonConstThis = (RuleBasedBreakIterator*)this;
-
- // The iterator is initialized pointing to no text at all, so if this
- // function is called while we're in that state, we have to fudge an
- // an iterator to return.
- if (nonConstThis->fText == NULL) {
- nonConstThis->fText = new StringCharacterIterator(UnicodeString());
- }
- return *nonConstThis->fText;
+ return *fCharIter;
}
/**
*/
void
RuleBasedBreakIterator::adoptText(CharacterIterator* newText) {
- reset();
- delete fText;
- fText = newText;
+ // If we are holding a CharacterIterator adopted from a
+ // previous call to this function, delete it now.
+ if (fCharIter != &fSCharIter) {
+ delete fCharIter;
+ }
+
+ fCharIter = newText;
+ UErrorCode status = U_ZERO_ERROR;
+ fBreakCache->reset();
+ fDictionaryCache->reset();
+ if (newText==NULL || newText->startIndex() != 0) {
+ // startIndex !=0 wants to be an error, but there's no way to report it.
+ // Make the iterator text be an empty string.
+ utext_openUChars(&fText, NULL, 0, &status);
+ } else {
+ utext_openCharacterIterator(&fText, newText, &status);
+ }
this->first();
}
*/
void
RuleBasedBreakIterator::setText(const UnicodeString& newText) {
- reset();
- if (fText != NULL && fText->getDynamicClassID()
- == StringCharacterIterator::getStaticClassID()) {
- ((StringCharacterIterator*)fText)->setText(newText);
- }
- else {
- delete fText;
- fText = new StringCharacterIterator(newText);
+ UErrorCode status = U_ZERO_ERROR;
+ fBreakCache->reset();
+ fDictionaryCache->reset();
+ utext_openConstUnicodeString(&fText, &newText, &status);
+
+ // Set up a character iterator on the string.
+ // Needed in case someone calls getText().
+ // Can not, unfortunately, do this lazily on the (probably never)
+ // call to getText(), because getText is const.
+ fSCharIter.setText(newText);
+
+ if (fCharIter != &fSCharIter) {
+ // old fCharIter was adopted from the outside. Delete it.
+ delete fCharIter;
}
+ fCharIter = &fSCharIter;
+
this->first();
}
+/**
+ * Provide a new UText for the input text. Must reference text with contents identical
+ * to the original.
+ * Intended for use with text data originating in Java (garbage collected) environments
+ * where the data may be moved in memory at arbitrary times.
+ */
+RuleBasedBreakIterator &RuleBasedBreakIterator::refreshInputText(UText *input, UErrorCode &status) {
+ if (U_FAILURE(status)) {
+ return *this;
+ }
+ if (input == NULL) {
+ status = U_ILLEGAL_ARGUMENT_ERROR;
+ return *this;
+ }
+ int64_t pos = utext_getNativeIndex(&fText);
+ // Shallow read-only clone of the new UText into the existing input UText
+ utext_clone(&fText, input, FALSE, TRUE, &status);
+ if (U_FAILURE(status)) {
+ return *this;
+ }
+ utext_setNativeIndex(&fText, pos);
+ if (utext_getNativeIndex(&fText) != pos) {
+ // Sanity check. The new input utext is supposed to have the exact same
+ // contents as the old. If we can't set to the same position, it doesn't.
+ // The contents underlying the old utext might be invalid at this point,
+ // so it's not safe to check directly.
+ status = U_ILLEGAL_ARGUMENT_ERROR;
+ }
+ return *this;
+}
+
/**
- * Sets the current iteration position to the beginning of the text.
- * (i.e., the CharacterIterator's starting offset).
- * @return The offset of the beginning of the text.
+ * Sets the current iteration position to the beginning of the text, position zero.
+ * @return The new iterator position, which is zero.
*/
int32_t RuleBasedBreakIterator::first(void) {
- reset();
- fLastRuleStatusIndex = 0;
- fLastStatusIndexValid = TRUE;
- if (fText == NULL)
- return BreakIterator::DONE;
-
- fText->first();
- return fText->getIndex();
+ UErrorCode status = U_ZERO_ERROR;
+ if (!fBreakCache->seek(0)) {
+ fBreakCache->populateNear(0, status);
+ }
+ fBreakCache->current();
+ U_ASSERT(fPosition == 0);
+ return 0;
}
/**
* Sets the current iteration position to the end of the text.
- * (i.e., the CharacterIterator's ending offset).
* @return The text's past-the-end offset.
*/
int32_t RuleBasedBreakIterator::last(void) {
- reset();
- if (fText == NULL) {
- fLastRuleStatusIndex = 0;
- fLastStatusIndexValid = TRUE;
- return BreakIterator::DONE;
- }
-
- // I'm not sure why, but t.last() returns the offset of the last character,
- // rather than the past-the-end offset
- //
- // (It's so a loop like for(p=it.last(); p!=DONE; p=it.previous()) ...
- // will work correctly.)
-
-
- fLastStatusIndexValid = FALSE;
- int32_t pos = fText->endIndex();
- fText->setIndex(pos);
-
- return pos;
+ int32_t endPos = (int32_t)utext_nativeLength(&fText);
+ UBool endShouldBeBoundary = isBoundary(endPos); // Has side effect of setting iterator position.
+ (void)endShouldBeBoundary;
+ U_ASSERT(endShouldBeBoundary);
+ U_ASSERT(fPosition == endPos);
+ return endPos;
}
/**
* the current one.
*/
int32_t RuleBasedBreakIterator::next(int32_t n) {
- int32_t result = current();
- while (n > 0) {
- result = handleNext();
- --n;
- }
- while (n < 0) {
- result = previous();
- ++n;
+ int32_t result = 0;
+ if (n > 0) {
+ for (; n > 0 && result != UBRK_DONE; --n) {
+ result = next();
+ }
+ } else if (n < 0) {
+ for (; n < 0 && result != UBRK_DONE; ++n) {
+ result = previous();
+ }
+ } else {
+ result = current();
}
return result;
}
* @return The position of the first boundary after this one.
*/
int32_t RuleBasedBreakIterator::next(void) {
- return handleNext();
+ fBreakCache->next();
+ return fDone ? UBRK_DONE : fPosition;
}
/**
- * Advances the iterator backwards, to the last boundary preceding this one.
- * @return The position of the last boundary position preceding this one.
+ * Move the iterator backwards, to the boundary preceding the current one.
+ *
+ * Starts from the current position within fText.
+ * Starting position need not be on a boundary.
+ *
+ * @return The position of the boundary position immediately preceding the starting position.
*/
int32_t RuleBasedBreakIterator::previous(void) {
- // if we're already sitting at the beginning of the text, return DONE
- if (fText == NULL || current() == fText->startIndex()) {
- fLastRuleStatusIndex = 0;
- fLastStatusIndexValid = TRUE;
- return BreakIterator::DONE;
- }
-
- if (fData->fSafeRevTable != NULL || fData->fSafeFwdTable != NULL) {
- return handlePrevious(fData->fReverseTable);
- }
-
- // old rule syntax
- // set things up. handlePrevious() will back us up to some valid
- // break position before the current position (we back our internal
- // iterator up one step to prevent handlePrevious() from returning
- // the current position), but not necessarily the last one before
- // where we started
-
- int32_t start = current();
-
- fText->previous32();
- int32_t lastResult = handlePrevious();
- int32_t result = lastResult;
- int32_t lastTag = 0;
- UBool breakTagValid = FALSE;
-
- // iterate forward from the known break position until we pass our
- // starting point. The last break position before the starting
- // point is our return value
-
- for (;;) {
- result = handleNext();
- if (result == BreakIterator::DONE || result >= start) {
- break;
- }
- lastResult = result;
- lastTag = fLastRuleStatusIndex;
- breakTagValid = TRUE;
- }
-
- // fLastBreakTag wants to have the value for section of text preceding
- // the result position that we are to return (in lastResult.) If
- // the backwards rules overshot and the above loop had to do two or more
- // handleNext()s to move up to the desired return position, we will have a valid
- // tag value. But, if handlePrevious() took us to exactly the correct result positon,
- // we wont have a tag value for that position, which is only set by handleNext().
-
- // set the current iteration position to be the last break position
- // before where we started, and then return that value
- fText->setIndex(lastResult);
- fLastRuleStatusIndex = lastTag; // for use by getRuleStatus()
- fLastStatusIndexValid = breakTagValid;
- return lastResult;
+ UErrorCode status = U_ZERO_ERROR;
+ fBreakCache->previous(status);
+ return fDone ? UBRK_DONE : fPosition;
}
/**
* Sets the iterator to refer to the first boundary position following
* the specified position.
- * @offset The position from which to begin searching for a break position.
+ * @param startPos The position from which to begin searching for a break position.
* @return The position of the first break after the current position.
*/
-int32_t RuleBasedBreakIterator::following(int32_t offset) {
- // if the offset passed in is already past the end of the text,
- // just return DONE; if it's before the beginning, return the
+int32_t RuleBasedBreakIterator::following(int32_t startPos) {
+ // if the supplied position is before the beginning, return the
// text's starting offset
- fLastRuleStatusIndex = 0;
- fLastStatusIndexValid = TRUE;
- if (fText == NULL || offset >= fText->endIndex()) {
- last();
- return next();
- }
- else if (offset < fText->startIndex()) {
+ if (startPos < 0) {
return first();
}
- // otherwise, set our internal iteration position (temporarily)
- // to the position passed in. If this is the _beginning_ position,
- // then we can just use next() to get our return value
-
- int32_t result = 0;
-
- if (fData->fSafeRevTable != NULL) {
- // new rule syntax
- /// todo synwee
- fText->setIndex(offset);
- // move forward one codepoint to prepare for moving back to a
- // safe point.
- // this handles offset being between a supplementary character
- fText->next32();
- // handlePrevious will move most of the time to < 1 boundary away
- handlePrevious(fData->fSafeRevTable);
- int32_t result = next();
- while (result <= offset) {
- result = next();
- }
- return result;
- }
- if (fData->fSafeFwdTable != NULL) {
- // backup plan if forward safe table is not available
- fText->setIndex(offset);
- fText->previous32();
- // handle next will give result >= offset
- handleNext(fData->fSafeFwdTable);
- // previous will give result 0 or 1 boundary away from offset,
- // most of the time
- // we have to
- int32_t oldresult = previous();
- while (oldresult > offset) {
- int32_t result = previous();
- if (result <= offset) {
- return oldresult;
- }
- oldresult = result;
- }
- int32_t result = next();
- if (result <= offset) {
- return next();
- }
- return result;
- }
- // otherwise, we have to sync up first. Use handlePrevious() to back
- // us up to a known break position before the specified position (if
- // we can determine that the specified position is a break position,
- // we don't back up at all). This may or may not be the last break
- // position at or before our starting position. Advance forward
- // from here until we've passed the starting position. The position
- // we stop on will be the first break position after the specified one.
- // old rule syntax
-
- fText->setIndex(offset);
- if (offset == fText->startIndex()) {
- return handleNext();
- }
- result = previous();
-
- while (result != BreakIterator::DONE && result <= offset) {
- result = next();
- }
+ // Move requested offset to a code point start. It might be on a trail surrogate,
+ // or on a trail byte if the input is UTF-8. Or it may be beyond the end of the text.
+ utext_setNativeIndex(&fText, startPos);
+ startPos = (int32_t)utext_getNativeIndex(&fText);
- return result;
+ UErrorCode status = U_ZERO_ERROR;
+ fBreakCache->following(startPos, status);
+ return fDone ? UBRK_DONE : fPosition;
}
/**
* Sets the iterator to refer to the last boundary position before the
* specified position.
- * @offset The position to begin searching for a break from.
+ * @param offset The position to begin searching for a break from.
* @return The position of the last boundary before the starting position.
*/
int32_t RuleBasedBreakIterator::preceding(int32_t offset) {
- // if the offset passed in is already past the end of the text,
- // just return DONE; if it's before the beginning, return the
-
- // text's starting offset
- if (fText == NULL || offset > fText->endIndex()) {
- // return BreakIterator::DONE;
+ if (offset > utext_nativeLength(&fText)) {
return last();
}
- else if (offset < fText->startIndex()) {
- return first();
- }
- // if we start by updating the current iteration position to the
- // position specified by the caller, we can just use previous()
- // to carry out this operation
-
- if (fData->fSafeFwdTable != NULL) {
- /// todo synwee
- // new rule syntax
- fText->setIndex(offset);
- // move backwards one codepoint to prepare for moving forwards to a
- // safe point.
- // this handles offset being between a supplementary character
- // TODO: would it be better to just check for being in the middle of a surrogate pair,
- // rather than adjusting the position unconditionally?
- // (Change would interact with safe rules.)
- fText->previous32();
- handleNext(fData->fSafeFwdTable);
- int32_t result = fText->getIndex();
- while (result >= offset) {
- result = previous();
- }
- return result;
- }
- if (fData->fSafeRevTable != NULL) {
- // backup plan if forward safe table is not available
- fText->setIndex(offset);
- fText->next32();
- // handle previous will give result <= offset
- handlePrevious(fData->fSafeRevTable);
-
- // next will give result 0 or 1 boundary away from offset,
- // most of the time
- // we have to
- int32_t oldresult = next();
- while (oldresult < offset) {
- int32_t result = next();
- if (result >= offset) {
- return oldresult;
- }
- oldresult = result;
- }
- int32_t result = previous();
- if (result >= offset) {
- return previous();
- }
- return result;
- }
+ // Move requested offset to a code point start. It might be on a trail surrogate,
+ // or on a trail byte if the input is UTF-8.
- // old rule syntax
- fText->setIndex(offset);
- return previous();
+ utext_setNativeIndex(&fText, offset);
+ int32_t adjustedOffset = utext_getNativeIndex(&fText);
+
+ UErrorCode status = U_ZERO_ERROR;
+ fBreakCache->preceding(adjustedOffset, status);
+ return fDone ? UBRK_DONE : fPosition;
}
/**
* Returns true if the specfied position is a boundary position. As a side
* effect, leaves the iterator pointing to the first boundary position at
* or after "offset".
+ *
* @param offset the offset to check.
* @return True if "offset" is a boundary position.
*/
UBool RuleBasedBreakIterator::isBoundary(int32_t offset) {
- // the beginning index of the iterator is always a boundary position by definition
- if (fText == NULL || offset == fText->startIndex()) {
+ // out-of-range indexes are never boundary positions
+ if (offset < 0) {
first(); // For side effects on current position, tag values.
- return TRUE;
+ return FALSE;
}
- if (offset == fText->endIndex()) {
- last(); // For side effects on current position, tag values.
- return TRUE;
- }
+ // Adjust offset to be on a code point boundary and not beyond the end of the text.
+ // Note that isBoundary() is always false for offsets that are not on code point boundaries.
+ // But we still need the side effect of leaving iteration at the following boundary.
- // out-of-range indexes are never boundary positions
- if (offset < fText->startIndex()) {
- first(); // For side effects on current position, tag values.
- return FALSE;
+ utext_setNativeIndex(&fText, offset);
+ int32_t adjustedOffset = utext_getNativeIndex(&fText);
+
+ bool result = false;
+ UErrorCode status = U_ZERO_ERROR;
+ if (fBreakCache->seek(adjustedOffset) || fBreakCache->populateNear(adjustedOffset, status)) {
+ result = (fBreakCache->current() == offset);
}
- if (offset > fText->endIndex()) {
- last(); // For side effects on current position, tag values.
+ if (result && adjustedOffset < offset && utext_char32At(&fText, offset) == U_SENTINEL) {
+ // Original offset is beyond the end of the text. Return FALSE, it's not a boundary,
+ // but the iteration position remains set to the end of the text, which is a boundary.
return FALSE;
}
-
- // otherwise, we can use following() on the position before the specified
- // one and return true if the position we get back is the one the user
- // specified
- return following(offset - 1) == offset;
+ if (!result) {
+ // Not on a boundary. isBoundary() must leave iterator on the following boundary.
+ // Cache->seek(), above, left us on the preceding boundary, so advance one.
+ next();
+ }
+ return result;
}
+
/**
* Returns the current iteration position.
* @return The current iteration position.
*/
int32_t RuleBasedBreakIterator::current(void) const {
- return (fText != NULL) ? fText->getIndex() : BreakIterator::DONE;
+ return fPosition;
}
+
//=======================================================================
// implementation
//=======================================================================
+//
+// RBBIRunMode - the state machine runs an extra iteration at the beginning and end
+// of user text. A variable with this enum type keeps track of where we
+// are. The state machine only fetches user input while in the RUN mode.
+//
+enum RBBIRunMode {
+ RBBI_START, // state machine processing is before first char of input
+ RBBI_RUN, // state machine processing is in the user text
+ RBBI_END // state machine processing is after end of user text.
+};
+
+
+// Map from look-ahead break states (corresponds to rules) to boundary positions.
+// Allows multiple lookahead break rules to be in flight at the same time.
+//
+// This is a temporary approach for ICU 57. A better fix is to make the look-ahead numbers
+// in the state table be sequential, then we can just index an array. And the
+// table could also tell us in advance how big that array needs to be.
+//
+// Before ICU 57 there was just a single simple variable for a look-ahead match that
+// was in progress. Two rules at once did not work.
+
+static const int32_t kMaxLookaheads = 8;
+struct LookAheadResults {
+ int32_t fUsedSlotLimit;
+ int32_t fPositions[8];
+ int16_t fKeys[8];
+
+ LookAheadResults() : fUsedSlotLimit(0), fPositions(), fKeys() {};
+
+ int32_t getPosition(int16_t key) {
+ for (int32_t i=0; i<fUsedSlotLimit; ++i) {
+ if (fKeys[i] == key) {
+ return fPositions[i];
+ }
+ }
+ // with NLLT source rules, Latn sample and ubrk_next, we see a request for key 79 here
+ // near the end of text, when setPosition has only ever set positions for key 80 or 82.
+ //U_ASSERT(FALSE);
+ return -1;
+ }
+
+ void setPosition(int16_t key, int32_t position) {
+ int32_t i;
+ for (i=0; i<fUsedSlotLimit; ++i) {
+ if (fKeys[i] == key) {
+ fPositions[i] = position;
+ return;
+ }
+ }
+ if (i >= kMaxLookaheads) {
+ U_ASSERT(FALSE);
+ i = kMaxLookaheads - 1;
+ }
+ fKeys[i] = key;
+ fPositions[i] = position;
+ U_ASSERT(fUsedSlotLimit == i);
+ fUsedSlotLimit = i + 1;
+ }
+};
+
//-----------------------------------------------------------------------------------
//
// handleNext()
-// This method is the actual implementation of the next() method. All iteration
-// vectors through here. This method initializes the state machine to state 1
-// and advances through the text character by character until we reach the end
-// of the text or the state machine transitions to state 0. We update our return
-// value every time the state machine passes through an accepting state.
+// Run the state machine to find a boundary
//
//-----------------------------------------------------------------------------------
+// Route handleNext calls through the following to handleNextInternal,
+// in order to handle fLineWordOpts.
int32_t RuleBasedBreakIterator::handleNext() {
- return handleNext(fData->fForwardTable);
+ int32_t result = handleNextInternal();
+ while (fLineWordOpts != UBRK_LINEWORD_NORMAL) {
+ UChar32 prevChr = utext_char32At(&fText, result-1);
+ UChar32 currChr = utext_char32At(&fText, result);
+ if (currChr == U_SENTINEL || prevChr == U_SENTINEL) {
+ break;
+ }
+ if (fLineWordOpts == UBRK_LINEWORD_KEEP_HANGUL) {
+ UErrorCode status = U_ZERO_ERROR;
+ if (uscript_getScript(currChr, &status) != USCRIPT_HANGUL || uscript_getScript(prevChr, &status) != USCRIPT_HANGUL) {
+ break;
+ }
+ } else {
+ if (!u_isalpha(currChr) || !u_isalpha(prevChr)) {
+ break;
+ }
+ }
+ int32_t nextResult = handleNextInternal();
+ if (nextResult <= result) {
+ break;
+ }
+ result = nextResult;
+ }
+ return result;
}
-int32_t RuleBasedBreakIterator::handleNext(const RBBIStateTable *statetable) {
- if (fTrace) {
- RBBIDebugPuts("Handle Next pos char state category");
- }
+int32_t RuleBasedBreakIterator::handleNextInternal() {
+ int32_t state;
+ uint16_t category = 0;
+ RBBIRunMode mode;
+
+ RBBIStateTableRow *row;
+ UChar32 c;
+ LookAheadResults lookAheadMatches;
+ int32_t result = 0;
+ int32_t initialPosition = 0;
+ const RBBIStateTable *statetable = fData->fForwardTable;
+ const char *tableData = statetable->fTableData;
+ uint32_t tableRowLen = statetable->fRowLen;
+ #ifdef RBBI_DEBUG
+ if (gTrace) {
+ RBBIDebugPuts("Handle Next pos char state category");
+ }
+ #endif
+
+ // handleNext alway sets the break tag value.
+ // Set the default for it.
+ fRuleStatusIndex = 0;
- // No matter what, handleNext alway correctly sets the break tag value.
- fLastStatusIndexValid = TRUE;
+ fDictionaryCharCount = 0;
// if we're already at the end of the text, return DONE.
- if (fText == NULL || fData == NULL || fText->hasNext() == FALSE) {
- fLastRuleStatusIndex = 0;
- return BreakIterator::DONE;
+ initialPosition = fPosition;
+ UTEXT_SETNATIVEINDEX(&fText, initialPosition);
+ result = initialPosition;
+ c = UTEXT_NEXT32(&fText);
+ if (c==U_SENTINEL) {
+ fDone = TRUE;
+ return UBRK_DONE;
}
- int32_t initialPosition = fText->getIndex();
- int32_t result = initialPosition;
- int32_t lookaheadResult = 0;
-
- // Initialize the state machine. Begin in state 1
- int32_t state = START_STATE;
- int16_t category;
- UChar32 c = fText->current32();
- RBBIStateTableRow *row;
- int32_t lookaheadStatus = 0;
- int32_t lookaheadTagIdx = 0;
+ // Set the initial state for the state machine
+ state = START_STATE;
+ row = (RBBIStateTableRow *)
+ //(statetable->fTableData + (statetable->fRowLen * state));
+ (tableData + tableRowLen * state);
- fLastRuleStatusIndex = 0;
- row = (RBBIStateTableRow *) // Point to starting row of state table.
- (statetable->fTableData + (statetable->fRowLen * state));
+ mode = RBBI_RUN;
+ if (statetable->fFlags & RBBI_BOF_REQUIRED) {
+ category = 2;
+ mode = RBBI_START;
+ }
- // Character Category fetch for starting character.
- // See comments on character category code within loop, below.
- UTRIE_GET16(&fData->fTrie, c, category);
- if ((category & 0x4000) != 0) {
- fDictionaryCharCount++;
- category &= ~0x4000;
- }
// loop until we reach the end of the text or transition to state 0
+ //
for (;;) {
- if (c == CharacterIterator::DONE && fText->hasNext()==FALSE) {
+ if (c == U_SENTINEL) {
// Reached end of input string.
- // Note: CharacterIterator::DONE is 0xffff, which is also a legal
- // character value. Check for DONE first, because it's quicker,
- // but also need to check fText->hasNext() to be certain.
-
- if (lookaheadResult > result) {
- // We ran off the end of the string with a pending look-ahead match.
- // Treat this as if the look-ahead condition had been met, and return
- // the match at the / position from the look-ahead rule.
- result = lookaheadResult;
- fLastRuleStatusIndex = lookaheadTagIdx;
- lookaheadStatus = 0;
- } else if (result == initialPosition) {
- // Ran off end, no match found.
- // move forward one
- fText->setIndex(initialPosition);
- fText->next32();
- fText->getIndex();
+ if (mode == RBBI_END) {
+ // We have already run the loop one last time with the
+ // character set to the psueudo {eof} value. Now it is time
+ // to unconditionally bail out.
+ break;
}
- break;
+ // Run the loop one last time with the fake end-of-input character category.
+ mode = RBBI_END;
+ category = 1;
}
- // look up the current character's character category, which tells us
- // which column in the state table to look at.
- // Note: the 16 in UTRIE_GET16 refers to the size of the data being returned,
- // not the size of the character going in, which is a UChar32.
- //
- UTRIE_GET16(&fData->fTrie, c, category);
- // Check the dictionary bit in the character's category.
- // Counter is only used by dictionary based iterators (subclasses).
- // Chars that need to be handled by a dictionary have a flag bit set
- // in their category values.
//
- if ((category & 0x4000) != 0) {
- fDictionaryCharCount++;
- // And off the dictionary flag bit.
- category &= ~0x4000;
+ // Get the char category. An incoming category of 1 or 2 means that
+ // we are preset for doing the beginning or end of input, and
+ // that we shouldn't get a category from an actual text input character.
+ //
+ if (mode == RBBI_RUN) {
+ // look up the current character's character category, which tells us
+ // which column in the state table to look at.
+ // Note: the 16 in UTRIE_GET16 refers to the size of the data being returned,
+ // not the size of the character going in, which is a UChar32.
+ //
+ category = (fLatin1Cat!=NULL && c<0x100)? fLatin1Cat[c]: UTRIE2_GET16(fData->fTrie, c);
+
+ // Check the dictionary bit in the character's category.
+ // Counter is only used by dictionary based iteration.
+ // Chars that need to be handled by a dictionary have a flag bit set
+ // in their category values.
+ //
+ if ((category & 0x4000) != 0) {
+ fDictionaryCharCount++;
+ // And off the dictionary flag bit.
+ category &= ~0x4000;
+ }
}
- #ifdef RBBI_DEBUG
- if (fTrace) {
- RBBIDebugPrintf(" %4d ", fText->getIndex());
+ #ifdef RBBI_DEBUG
+ if (gTrace) {
+ RBBIDebugPrintf(" %4ld ", utext_getNativeIndex(&fText));
if (0x20<=c && c<0x7f) {
RBBIDebugPrintf("\"%c\" ", c);
} else {
}
#endif
- // look up a state transition in the state table
- state = row->fNextState[category];
+ // State Transition - move machine to its next state
+ //
+
+ // fNextState is a variable-length array.
+ U_ASSERT(category<fData->fHeader->fCatCount);
+ state = row->fNextState[category]; /*Not accessing beyond memory*/
row = (RBBIStateTableRow *)
- (statetable->fTableData + (statetable->fRowLen * state));
+ // (statetable->fTableData + (statetable->fRowLen * state));
+ (tableData + tableRowLen * state);
- // Get the next character. Doing it here positions the iterator
- // to the correct position for recording matches in the code that
- // follows.
- c = fText->next32();
if (row->fAccepting == -1) {
- // Match found, common case, could have lookahead so we move on to check it
- result = fText->getIndex();
- /// added
- fLastRuleStatusIndex = row->fTagIdx; // Remember the break status (tag) values.
+ // Match found, common case.
+ if (mode != RBBI_START) {
+ result = (int32_t)UTEXT_GETNATIVEINDEX(&fText);
+ }
+ fRuleStatusIndex = row->fTagIdx; // Remember the break status (tag) values.
}
- if (row->fLookAhead != 0) {
- if (lookaheadStatus != 0
- && row->fAccepting == lookaheadStatus) {
- // Lookahead match is completed. Set the result accordingly, but only
- // if no other rule has matched further in the mean time.
- result = lookaheadResult;
- fLastRuleStatusIndex = lookaheadTagIdx;
- lookaheadStatus = 0;
- /// i think we have to back up to read the lookahead character again
- /// fText->setIndex(lookaheadResult);
- /// TODO: this is a simple hack since reverse rules only have simple
- /// lookahead rules that we can definitely break out from.
- /// we need to make the lookahead rules not chain eventually.
- /// return result;
- /// this is going to be the longest match again
- goto continueOn;
+ int16_t completedRule = row->fAccepting;
+ if (completedRule > 0) {
+ // Lookahead match is completed.
+ int32_t lookaheadResult = lookAheadMatches.getPosition(completedRule);
+ if (lookaheadResult >= 0) {
+ fRuleStatusIndex = row->fTagIdx;
+ fPosition = lookaheadResult;
+ return lookaheadResult;
}
-
- int32_t r = fText->getIndex();
- lookaheadResult = r;
- lookaheadStatus = row->fLookAhead;
- lookaheadTagIdx = row->fTagIdx;
- goto continueOn;
}
-
-
- if (row->fAccepting == 0) {
- // No match, nothing of interest happening, common case.
- goto continueOn;
+ int16_t rule = row->fLookAhead;
+ if (rule != 0) {
+ // At the position of a '/' in a look-ahead match. Record it.
+ int32_t pos = (int32_t)UTEXT_GETNATIVEINDEX(&fText);
+ lookAheadMatches.setPosition(rule, pos);
}
- lookaheadStatus = 0; // clear out any pending look-ahead matches.
-
-continueOn:
if (state == STOP_STATE) {
// This is the normal exit from the lookup state machine.
// We have advanced through the string until it is certain that no
// longer match is possible, no matter what characters follow.
break;
}
+
+ // Advance to the next character.
+ // If this is a beginning-of-input loop iteration, don't advance
+ // the input position. The next iteration will be processing the
+ // first real input character.
+ if (mode == RBBI_RUN) {
+ c = UTEXT_NEXT32(&fText);
+ } else {
+ if (mode == RBBI_START) {
+ mode = RBBI_RUN;
+ }
+ }
}
// The state machine is done. Check whether it found a match...
// (This really indicates a defect in the break rules. They should always match
// at least one character.)
if (result == initialPosition) {
- result = fText->setIndex(initialPosition);
- fText ->next32();
- result = fText->getIndex();
+ utext_setNativeIndex(&fText, initialPosition);
+ utext_next32(&fText);
+ result = (int32_t)utext_getNativeIndex(&fText);
+ fRuleStatusIndex = 0;
}
// Leave the iterator at our result position.
- fText->setIndex(result);
- if (fTrace) {
- RBBIDebugPrintf("result = %d\n\n", result);
- }
- return result;
-}
-
-
-//----------------------------------------------------------------
-//
-// handlePrevious(void) This is the variant used with old style rules
-// (Overshoot to a safe point, then move forward)
-//
-//----------------------------------------------------------------
-int32_t RuleBasedBreakIterator::handlePrevious(void) {
- if (fText == NULL || fData == NULL) {
- return 0;
- }
- if (fData->fReverseTable == NULL) {
- return fText->setToStart();
- }
-
- int32_t state = START_STATE;
- int32_t category;
- int32_t lastCategory = 0;
- int32_t result = fText->getIndex();
- int32_t lookaheadStatus = 0;
- int32_t lookaheadResult = 0;
- int32_t lookaheadTagIdx = 0;
- UChar32 c = fText->current32();
- RBBIStateTableRow *row;
-
- row = (RBBIStateTableRow *)
- (this->fData->fReverseTable->fTableData + (state * fData->fReverseTable->fRowLen));
- UTRIE_GET16(&fData->fTrie, c, category);
- if ((category & 0x4000) != 0) {
- fDictionaryCharCount++;
- category &= ~0x4000;
- }
-
- if (fTrace) {
- RBBIDebugPuts("Handle Prev pos char state category");
- }
-
- // loop until we reach the beginning of the text or transition to state 0
- for (;;) {
- if (c == CharacterIterator::DONE && fText->hasPrevious()==FALSE) {
- break;
- }
-
- // save the last character's category and look up the current
- // character's category
- lastCategory = category;
- UTRIE_GET16(&fData->fTrie, c, category);
-
- // Check the dictionary bit in the character's category.
- // Counter is only used by dictionary based iterators.
- //
- if ((category & 0x4000) != 0) {
- fDictionaryCharCount++;
- category &= ~0x4000;
- }
-
- #ifdef RBBI_DEBUG
- if (fTrace) {
- RBBIDebugPrintf(" %4d ", fText->getIndex());
- if (0x20<=c && c<0x7f) {
- RBBIDebugPrintf("\"%c\" ", c);
- } else {
- RBBIDebugPrintf("%5x ", c);
- }
- RBBIDebugPrintf("%3d %3d\n", state, category);
- }
- #endif
-
- // look up a state transition in the backwards state table
- state = row->fNextState[category];
- row = (RBBIStateTableRow *)
- (this->fData->fReverseTable->fTableData + (state * fData->fReverseTable->fRowLen));
-
- if (row->fAccepting == 0 && row->fLookAhead == 0) {
- // No match, nothing of interest happening, common case.
- goto continueOn;
+ fPosition = result;
+ #ifdef RBBI_DEBUG
+ if (gTrace) {
+ RBBIDebugPrintf("result = %d\n\n", result);
}
-
- if (row->fAccepting == -1) {
- // Match found, common case, no lookahead involved.
- result = fText->getIndex();
- lookaheadStatus = 0; // clear out any pending look-ahead matches.
- goto continueOn;
- }
-
- if (row->fAccepting == 0 && row->fLookAhead != 0) {
- // Lookahead match point. Remember it, but only if no other rule
- // has unconditionally matched to this point.
- // TODO: handle case where there's a pending match from a different rule
- // where lookaheadStatus != 0 && lookaheadStatus != row->fLookAhead.
- int32_t r = fText->getIndex();
- if (r > result) {
- lookaheadResult = r;
- lookaheadStatus = row->fLookAhead;
- lookaheadTagIdx = row->fTagIdx;
- }
- goto continueOn;
- }
-
- if (row->fAccepting != 0 && row->fLookAhead != 0) {
- // Lookahead match is completed. Set the result accordingly, but only
- // if no other rule has matched further in the mean time.
- if (lookaheadResult > result) {
- U_ASSERT(row->fAccepting == lookaheadStatus); // TODO: handle this case
- // of overlapping lookahead matches.
- result = lookaheadResult;
- fLastRuleStatusIndex = lookaheadTagIdx;
- lookaheadStatus = 0;
- }
- goto continueOn;
- }
-
-continueOn:
- if (state == STOP_STATE) {
- break;
- }
-
- // then advance one character backwards
- c = fText->previous32();
- }
-
- // Note: the result postion isn't what is returned to the user by previous(),
- // but where the implementation of previous() turns around and
- // starts iterating forward again.
- if (c == CharacterIterator::DONE && fText->hasPrevious()==FALSE) {
- result = fText->startIndex();
- }
- fText->setIndex(result);
-
+ #endif
return result;
}
//-----------------------------------------------------------------------------------
//
-// handlePrevious()
+// handleSafePrevious()
//
-// This method backs the iterator back up to a "safe position" in the text.
-// This is a position that we know, without any context, may be any position
-// not more than 2 breaks away. Occasionally, the position may be less than
-// one break away.
-// The various calling methods then iterate forward from this safe position to
-// the appropriate position to return.
-//
-// The logic of this function is very similar to handleNext(), above.
+// Iterate backwards using the safe reverse rules.
+// The logic of this function is similar to handleNext(), but simpler
+// because the safe table does not require as many options.
//
//-----------------------------------------------------------------------------------
-int32_t RuleBasedBreakIterator::handlePrevious(const RBBIStateTable *statetable) {
- if (fText == NULL || statetable == NULL) {
- return 0;
- }
- // break tag is no longer valid after icu switched to exact backwards
- // positioning.
- fLastStatusIndexValid = FALSE;
- if (statetable == NULL) {
- return fText->setToStart();
- }
-
- int32_t state = START_STATE;
- int32_t category;
- int32_t lastCategory = 0;
- UBool hasPassedStartText = !fText->hasPrevious();
- UChar32 c = fText->previous32();
- // previous character
- int32_t result = fText->getIndex();
- int32_t lookaheadStatus = 0;
- int32_t lookaheadResult = 0;
- int32_t lookaheadTagIdx = 0;
- UBool lookAheadHardBreak = (statetable->fFlags & RBBI_LOOKAHEAD_HARD_BREAK) != 0;
-
- RBBIStateTableRow *row;
-
- row = (RBBIStateTableRow *)
- (statetable->fTableData + (state * statetable->fRowLen));
- UTRIE_GET16(&fData->fTrie, c, category);
- if ((category & 0x4000) != 0) {
- fDictionaryCharCount++;
- category &= ~0x4000;
- }
+int32_t RuleBasedBreakIterator::handleSafePrevious(int32_t fromPosition) {
+ int32_t state;
+ uint16_t category = 0;
+ RBBIStateTableRow *row;
+ UChar32 c;
+ int32_t result = 0;
+
+ const RBBIStateTable *stateTable = fData->fReverseTable;
+ UTEXT_SETNATIVEINDEX(&fText, fromPosition);
+ #ifdef RBBI_DEBUG
+ if (gTrace) {
+ RBBIDebugPuts("Handle Previous pos char state category");
+ }
+ #endif
- if (fTrace) {
- RBBIDebugPuts("Handle Prev pos char state category");
+ // if we're already at the start of the text, return DONE.
+ if (fData == NULL || UTEXT_GETNATIVEINDEX(&fText)==0) {
+ return BreakIterator::DONE;
}
- // loop until we reach the beginning of the text or transition to state 0
- for (;;) {
- // if (c == CharacterIterator::DONE && fText->hasPrevious()==FALSE) {
- if (hasPassedStartText) {
- // if we have already considered the start of the text
- if (row->fLookAhead != 0 && lookaheadResult == 0) {
- result = 0;
- }
- break;
- }
+ // Set the initial state for the state machine
+ c = UTEXT_PREVIOUS32(&fText);
+ state = START_STATE;
+ row = (RBBIStateTableRow *)
+ (stateTable->fTableData + (stateTable->fRowLen * state));
- // save the last character's category and look up the current
- // character's category
- lastCategory = category;
- UTRIE_GET16(&fData->fTrie, c, category);
+ // loop until we reach the start of the text or transition to state 0
+ //
+ for (; c != U_SENTINEL; c = UTEXT_PREVIOUS32(&fText)) {
- // Check the dictionary bit in the character's category.
- // Counter is only used by dictionary based iterators.
+ // look up the current character's character category, which tells us
+ // which column in the state table to look at.
+ // Note: the 16 in UTRIE_GET16 refers to the size of the data being returned,
+ // not the size of the character going in, which is a UChar32.
//
- if ((category & 0x4000) != 0) {
- fDictionaryCharCount++;
- category &= ~0x4000;
- }
+ // And off the dictionary flag bit. For reverse iteration it is not used.
+ category = UTRIE2_GET16(fData->fTrie, c);
+ category &= ~0x4000;
#ifdef RBBI_DEBUG
- if (fTrace) {
- RBBIDebugPrintf(" %4d ", fText->getIndex());
+ if (gTrace) {
+ RBBIDebugPrintf(" %4d ", (int32_t)utext_getNativeIndex(&fText));
if (0x20<=c && c<0x7f) {
RBBIDebugPrintf("\"%c\" ", c);
} else {
}
#endif
- // look up a state transition in the backwards state table
- state = row->fNextState[category];
+ // State Transition - move machine to its next state
+ //
+ // fNextState is a variable-length array.
+ U_ASSERT(category<fData->fHeader->fCatCount);
+ state = row->fNextState[category]; /*Not accessing beyond memory*/
row = (RBBIStateTableRow *)
- (statetable->fTableData + (state * statetable->fRowLen));
-
- if (row->fAccepting == -1) {
- // Match found, common case, could have lookahead so we move on to check it
- result = fText->getIndex();
- /// added
- fLastRuleStatusIndex = row->fTagIdx; // Remember the break status (tag) value.
- }
-
- if (row->fLookAhead != 0) {
- if (lookaheadStatus != 0
- && row->fAccepting == lookaheadStatus) {
- // Lookahead match is completed. Set the result accordingly, but only
- // if no other rule has matched further in the mean time.
- result = lookaheadResult;
- fLastRuleStatusIndex = lookaheadTagIdx;
- lookaheadStatus = 0;
- /// i think we have to back up to read the lookahead character again
- /// fText->setIndex(lookaheadResult);
- /// TODO: this is a simple hack since reverse rules only have simple
- /// lookahead rules that we can definitely break out from.
- /// we need to make the lookahead rules not chain eventually.
- /// return result;
- /// this is going to be the longest match again
-
- /// syn wee todo hard coded for line breaks stuff
- /// needs to provide a tag in rules to ensure a stop.
-
- if (lookAheadHardBreak) {
- fText->setIndex(result);
- return result;
- }
- category = lastCategory;
- fText->setIndex(result);
-
- goto continueOn;
- }
-
- int32_t r = fText->getIndex();
- lookaheadResult = r;
- lookaheadStatus = row->fLookAhead;
- fLastRuleStatusIndex = row->fTagIdx;
- goto continueOn;
- }
+ (stateTable->fTableData + (stateTable->fRowLen * state));
- // not lookahead
- if (row->fAccepting == 0) {
- // No match, nothing of interest happening, common case.
- goto continueOn;
- }
-
- lookaheadStatus = 0; // clear out any pending look-ahead matches.
-
-continueOn:
if (state == STOP_STATE) {
+ // This is the normal exit from the lookup state machine.
+ // Transistion to state zero means we have found a safe point.
break;
}
-
- // then advance one character backwards
- hasPassedStartText = !fText->hasPrevious();
- c = fText->previous32();
}
- // Note: the result postion isn't what is returned to the user by previous(),
- // but where the implementation of previous() turns around and
- // starts iterating forward again.
- fText->setIndex(result);
-
+ // The state machine is done. Check whether it found a match...
+ result = (int32_t)UTEXT_GETNATIVEINDEX(&fText);
+ #ifdef RBBI_DEBUG
+ if (gTrace) {
+ RBBIDebugPrintf("result = %d\n\n", result);
+ }
+ #endif
return result;
}
-
-void
-RuleBasedBreakIterator::reset()
-{
- // Base-class version of this function is a no-op.
- // Subclasses may override with their own reset behavior.
-}
-
-
-
//-------------------------------------------------------------------------------
//
// getRuleStatus() Return the break rule tag associated with the current
// position by iterating forwards, the value will have been
// cached by the handleNext() function.
//
-// If no cached status value is available, the status is
-// found by doing a previous() followed by a next(), which
-// leaves the iterator where it started, and computes the
-// status while doing the next().
-//
//-------------------------------------------------------------------------------
-void RuleBasedBreakIterator::makeRuleStatusValid() {
- if (fLastStatusIndexValid == FALSE) {
- // No cached status is available.
- if (fText == NULL || current() == fText->startIndex()) {
- // At start of text, or there is no text. Status is always zero.
- fLastRuleStatusIndex = 0;
- fLastStatusIndexValid = TRUE;
- } else {
- // Not at start of text. Find status the tedious way.
- int32_t pa = current();
- previous();
- int32_t pb = next();
- if (pa != pb) {
- // note: the if (pa != pb) test is here only to eliminate warnings for
- // unused local variables on gcc. Logically, it isn't needed.
- U_ASSERT(pa == pb);
- }
- }
- }
- U_ASSERT(fLastStatusIndexValid == TRUE);
- U_ASSERT(fLastRuleStatusIndex >= 0 && fLastRuleStatusIndex < fData->fStatusMaxIdx);
-}
-
int32_t RuleBasedBreakIterator::getRuleStatus() const {
- RuleBasedBreakIterator *nonConstThis = (RuleBasedBreakIterator *)this;
- nonConstThis->makeRuleStatusValid();
// fLastRuleStatusIndex indexes to the start of the appropriate status record
// (the number of status values.)
// This function returns the last (largest) of the array of status values.
- int32_t idx = fLastRuleStatusIndex + fData->fRuleStatusTable[fLastRuleStatusIndex];
+ int32_t idx = fRuleStatusIndex + fData->fRuleStatusTable[fRuleStatusIndex];
int32_t tagVal = fData->fRuleStatusTable[idx];
return tagVal;
}
-
-
int32_t RuleBasedBreakIterator::getRuleStatusVec(
- int32_t *fillInVec, int32_t capacity, UErrorCode &status)
-{
+ int32_t *fillInVec, int32_t capacity, UErrorCode &status) {
if (U_FAILURE(status)) {
return 0;
}
- RuleBasedBreakIterator *nonConstThis = (RuleBasedBreakIterator *)this;
- nonConstThis->makeRuleStatusValid();
- int32_t numVals = fData->fRuleStatusTable[fLastRuleStatusIndex];
+ int32_t numVals = fData->fRuleStatusTable[fRuleStatusIndex];
int32_t numValsToCopy = numVals;
if (numVals > capacity) {
status = U_BUFFER_OVERFLOW_ERROR;
}
int i;
for (i=0; i<numValsToCopy; i++) {
- fillInVec[i] = fData->fRuleStatusTable[fLastRuleStatusIndex + i + 1];
+ fillInVec[i] = fData->fRuleStatusTable[fRuleStatusIndex + i + 1];
}
return numVals;
}
-
+// Apple custom addition
+int32_t RuleBasedBreakIterator::tokenize(int32_t maxTokens, RuleBasedTokenRange *outTokenRanges, unsigned long *outTokenFlags)
+{
+ //os_log(OS_LOG_DEFAULT, "# tokenize 0: maxT %d; txt idx %lld, len %lld", maxTokens, utext_getNativeIndex(fText), utext_nativeLength(fText));
+ if (fDone) {
+ return 0;
+ }
+ RuleBasedTokenRange *outTokenLimit = outTokenRanges + maxTokens;
+ RuleBasedTokenRange *outTokenP = outTokenRanges;
+ int32_t lastOffset = fPosition;
+ //os_log(OS_LOG_DEFAULT, "# tokenize 1");
+ while (outTokenP < outTokenLimit) {
+ // start portion from inlining populateFollowing()
+ int32_t pos = 0;
+ int32_t ruleStatusIdx = 0;
+ int32_t startPos = fPosition;
+
+ if (fDictionaryCache->following(startPos, &pos, &ruleStatusIdx)) {
+ fPosition = pos;
+ fRuleStatusIndex = ruleStatusIdx;
+ } else {
+ pos = handleNextInternal(); // sets fRuleStatusIndex for the pos it returns, updates fPosition
+ if (pos == UBRK_DONE) {
+ // fDone = TRUE; already set by handleNextInternal
+ break;
+ }
+ // Use current result from handleNextInternal(), including fRuleStatusIndex,
+ // unless overridden by dictionary subdivisions
+ fPosition = pos;
+ if (fDictionaryCharCount > 0) {
+ // The text segment obtained from the rules includes dictionary characters.
+ // Subdivide it, with subdivided results going into the dictionary cache.
+ fDictionaryCache->populateDictionary(startPos, pos, fRuleStatusIndex, fRuleStatusIndex);
+ if (fDictionaryCache->following(startPos, &pos, &ruleStatusIdx)) {
+ fPosition = pos;
+ fRuleStatusIndex = ruleStatusIdx;
+ }
+ }
+ }
+ // end portion from inlining populateFollowing()
+ int32_t flagCount = fData->fRuleStatusTable[fRuleStatusIndex];
+ const int32_t* flagPtr = fData->fRuleStatusTable + fRuleStatusIndex + flagCount;
+ int32_t flagSet = *flagPtr; // if -1 then skip token
+ if (flagSet != -1) {
+ outTokenP->location = lastOffset;
+ outTokenP++->length = fPosition - lastOffset;
+ if (outTokenFlags) {
+ // flagSet should be the OR of all flags returned by getRuleStatusVec;
+ // here we collect from high-order to low-order.
+ while (--flagCount > 0) {
+ flagSet |= *--flagPtr;
+ }
+ *outTokenFlags++ = (unsigned long)flagSet;
+ }
+ }
+ lastOffset = fPosition;
+ }
+ return (outTokenP - outTokenRanges);
+}
//-------------------------------------------------------------------------------
//
}
-
-
-//-------------------------------------------------------------------------------
-//
-// BufferClone TODO: In my (Andy) opinion, this function should be deprecated.
-// Saving one heap allocation isn't worth the trouble.
-// Cloning shouldn't be done in tight loops, and
-// making the clone copy involves other heap operations anyway.
-// And the application code for correctly dealing with buffer
-// size problems and the eventual object destruction is ugly.
-//
-//-------------------------------------------------------------------------------
-BreakIterator * RuleBasedBreakIterator::createBufferClone(void *stackBuffer,
+BreakIterator * RuleBasedBreakIterator::createBufferClone(void * /*stackBuffer*/,
int32_t &bufferSize,
UErrorCode &status)
{
return NULL;
}
- //
- // If user buffer size is zero this is a preflight operation to
- // obtain the needed buffer size, allowing for worst case misalignment.
- //
if (bufferSize == 0) {
- bufferSize = sizeof(RuleBasedBreakIterator) + U_ALIGNMENT_OFFSET_UP(0);
+ bufferSize = 1; // preflighting for deprecated functionality
return NULL;
}
+ BreakIterator *clonedBI = clone();
+ if (clonedBI == NULL) {
+ status = U_MEMORY_ALLOCATION_ERROR;
+ } else {
+ status = U_SAFECLONE_ALLOCATED_WARNING;
+ }
+ return (RuleBasedBreakIterator *)clonedBI;
+}
- //
- // Check the alignment and size of the user supplied buffer.
- // Allocate heap memory if the user supplied memory is insufficient.
- //
- char *buf = (char *)stackBuffer;
- uint32_t s = bufferSize;
+U_NAMESPACE_END
- if (stackBuffer == NULL) {
- s = 0; // Ignore size, force allocation if user didn't give us a buffer.
- }
- if (U_ALIGNMENT_OFFSET(stackBuffer) != 0) {
- uint32_t offsetUp = (uint32_t)U_ALIGNMENT_OFFSET_UP(buf);
- s -= offsetUp;
- buf += offsetUp;
- }
- if (s < sizeof(RuleBasedBreakIterator)) {
- buf = (char *) new RuleBasedBreakIterator;
- if (buf == 0) {
- status = U_MEMORY_ALLOCATION_ERROR;
- return NULL;
+
+static icu::UStack *gLanguageBreakFactories = nullptr;
+static const icu::UnicodeString *gEmptyString = nullptr;
+static icu::UInitOnce gLanguageBreakFactoriesInitOnce = U_INITONCE_INITIALIZER;
+static icu::UInitOnce gRBBIInitOnce = U_INITONCE_INITIALIZER;
+
+/**
+ * Release all static memory held by breakiterator.
+ */
+U_CDECL_BEGIN
+static UBool U_CALLCONV rbbi_cleanup(void) {
+ delete gLanguageBreakFactories;
+ gLanguageBreakFactories = nullptr;
+ delete gEmptyString;
+ gEmptyString = nullptr;
+ gLanguageBreakFactoriesInitOnce.reset();
+ gRBBIInitOnce.reset();
+ return TRUE;
+}
+U_CDECL_END
+
+U_CDECL_BEGIN
+static void U_CALLCONV _deleteFactory(void *obj) {
+ delete (icu::LanguageBreakFactory *) obj;
+}
+U_CDECL_END
+U_NAMESPACE_BEGIN
+
+static void U_CALLCONV rbbiInit() {
+ gEmptyString = new UnicodeString();
+ ucln_common_registerCleanup(UCLN_COMMON_RBBI, rbbi_cleanup);
+}
+
+static void U_CALLCONV initLanguageFactories() {
+ UErrorCode status = U_ZERO_ERROR;
+ U_ASSERT(gLanguageBreakFactories == NULL);
+ gLanguageBreakFactories = new UStack(_deleteFactory, NULL, status);
+ if (gLanguageBreakFactories != NULL && U_SUCCESS(status)) {
+ ICULanguageBreakFactory *builtIn = new ICULanguageBreakFactory(status);
+ gLanguageBreakFactories->push(builtIn, status);
+#ifdef U_LOCAL_SERVICE_HOOK
+ LanguageBreakFactory *extra = (LanguageBreakFactory *)uprv_svc_hook("languageBreakFactory", &status);
+ if (extra != NULL) {
+ gLanguageBreakFactories->push(extra, status);
}
- status = U_SAFECLONE_ALLOCATED_WARNING;
+#endif
}
+ ucln_common_registerCleanup(UCLN_COMMON_RBBI, rbbi_cleanup);
+}
- //
- // Clone the object.
- // TODO: using an overloaded operator new to directly initialize the
- // copy in the user's buffer would be better, but it doesn't seem
- // to get along with namespaces. Investigate why.
- //
- // The memcpy is only safe with an empty (default constructed)
- // break iterator. Use on others can screw up reference counts
- // to data. memcpy-ing objects is not really a good idea...
- //
- RuleBasedBreakIterator localIter; // Empty break iterator, source for memcpy
- RuleBasedBreakIterator *clone = (RuleBasedBreakIterator *)buf;
- uprv_memcpy(clone, &localIter, sizeof(RuleBasedBreakIterator)); // clone = empty, but initialized, iterator.
- *clone = *this; // clone = the real one we want.
- if (status != U_SAFECLONE_ALLOCATED_WARNING) {
- clone->fBufferClone = TRUE;
+
+static const LanguageBreakEngine*
+getLanguageBreakEngineFromFactory(UChar32 c)
+{
+ umtx_initOnce(gLanguageBreakFactoriesInitOnce, &initLanguageFactories);
+ if (gLanguageBreakFactories == NULL) {
+ return NULL;
}
- return clone;
+ int32_t i = gLanguageBreakFactories->size();
+ const LanguageBreakEngine *lbe = NULL;
+ while (--i >= 0) {
+ LanguageBreakFactory *factory = (LanguageBreakFactory *)(gLanguageBreakFactories->elementAt(i));
+ lbe = factory->getEngineFor(c);
+ if (lbe != NULL) {
+ break;
+ }
+ }
+ return lbe;
}
-
//-------------------------------------------------------------------------------
//
-// isDictionaryChar Return true if the category lookup for this char
-// indicates that it is in the set of dictionary lookup
-// chars.
-//
-// This function is intended for use by dictionary based
-// break iterators.
+// getLanguageBreakEngine Find an appropriate LanguageBreakEngine for the
+// the character c.
//
//-------------------------------------------------------------------------------
-UBool RuleBasedBreakIterator::isDictionaryChar(UChar32 c) {
- if (fData == NULL) {
- return FALSE;
+const LanguageBreakEngine *
+RuleBasedBreakIterator::getLanguageBreakEngine(UChar32 c) {
+ const LanguageBreakEngine *lbe = NULL;
+ UErrorCode status = U_ZERO_ERROR;
+
+ if (fLanguageBreakEngines == NULL) {
+ fLanguageBreakEngines = new UStack(status);
+ if (fLanguageBreakEngines == NULL || U_FAILURE(status)) {
+ delete fLanguageBreakEngines;
+ fLanguageBreakEngines = 0;
+ return NULL;
+ }
}
- uint16_t category;
- UTRIE_GET16(&fData->fTrie, c, category);
- return (category & 0x4000) != 0;
+
+ int32_t i = fLanguageBreakEngines->size();
+ while (--i >= 0) {
+ lbe = (const LanguageBreakEngine *)(fLanguageBreakEngines->elementAt(i));
+ if (lbe->handles(c)) {
+ return lbe;
+ }
+ }
+
+ // No existing dictionary took the character. See if a factory wants to
+ // give us a new LanguageBreakEngine for this character.
+ lbe = getLanguageBreakEngineFromFactory(c);
+
+ // If we got one, use it and push it on our stack.
+ if (lbe != NULL) {
+ fLanguageBreakEngines->push((void *)lbe, status);
+ // Even if we can't remember it, we can keep looking it up, so
+ // return it even if the push fails.
+ return lbe;
+ }
+
+ // No engine is forthcoming for this character. Add it to the
+ // reject set. Create the reject break engine if needed.
+ if (fUnhandledBreakEngine == NULL) {
+ fUnhandledBreakEngine = new UnhandledEngine(status);
+ if (U_SUCCESS(status) && fUnhandledBreakEngine == NULL) {
+ status = U_MEMORY_ALLOCATION_ERROR;
+ return nullptr;
+ }
+ // Put it last so that scripts for which we have an engine get tried
+ // first.
+ fLanguageBreakEngines->insertElementAt(fUnhandledBreakEngine, 0, status);
+ // If we can't insert it, or creation failed, get rid of it
+ if (U_FAILURE(status)) {
+ delete fUnhandledBreakEngine;
+ fUnhandledBreakEngine = 0;
+ return NULL;
+ }
+ }
+
+ // Tell the reject engine about the character; at its discretion, it may
+ // add more than just the one character.
+ fUnhandledBreakEngine->handleCharacter(c);
+
+ return fUnhandledBreakEngine;
}
+void RuleBasedBreakIterator::dumpCache() {
+ fBreakCache->dumpCache();
+}
+void RuleBasedBreakIterator::dumpTables() {
+ fData->printData();
+}
+
+/**
+ * Returns the description used to create this iterator
+ */
+
+const UnicodeString&
+RuleBasedBreakIterator::getRules() const {
+ if (fData != NULL) {
+ return fData->getRuleSourceString();
+ } else {
+ umtx_initOnce(gRBBIInitOnce, &rbbiInit);
+ return *gEmptyString;
+ }
+}
U_NAMESPACE_END
projects / apple / icu.git / blobdiff