RuleBasedBreakIterator.javaAPI DocJava SE 5 API41572Fri Aug 26 14:57:20 BST 2005java.text


public class RuleBasedBreakIterator extends BreakIterator

A subclass of BreakIterator whose behavior is specified using a list of rules.

There are two kinds of rules, which are separated by semicolons: substitutions and regular expressions.

A substitution rule defines a name that can be used in place of an expression. It consists of a name, which is a string of characters contained in angle brackets, an equals sign, and an expression. (There can be no whitespace on either side of the equals sign.) To keep its syntactic meaning intact, the expression must be enclosed in parentheses or square brackets. A substitution is visible after its definition, and is filled in using simple textual substitution. Substitution definitions can contain other substitutions, as long as those substitutions have been defined first. Substitutions are generally used to make the regular expressions (which can get quite complex) shorted and easier to read. They typically define either character categories or commonly-used subexpressions.

There is one special substitution.  If the description defines a substitution called "<ignore>", the expression must be a [] expression, and the expression defines a set of characters (the "ignore characters") that will be transparent to the BreakIterator.  A sequence of characters will break the same way it would if any ignore characters it contains are taken out.  Break positions never occur befoer ignore characters.

A regular expression uses a subset of the normal Unix regular-expression syntax, and defines a sequence of characters to be kept together. With one significant exception, the iterator uses a longest-possible-match algorithm when matching text to regular expressions. The iterator also treats descriptions containing multiple regular expressions as if they were ORed together (i.e., as if they were separated by |).

The special characters recognized by the regular-expression parser are as follows:

* Specifies that the expression preceding the asterisk may occur any number of times (including not at all).
{} Encloses a sequence of characters that is optional.
() Encloses a sequence of characters.  If followed by *, the sequence repeats.  Otherwise, the parentheses are just a grouping device and a way to delimit the ends of expressions containing |.
| Separates two alternative sequences of characters.  Either one sequence or the other, but not both, matches this expression.  The | character can only occur inside ().
. Matches any character.
*? Specifies a non-greedy asterisk.  *? works the same way as *, except when there is overlap between the last group of characters in the expression preceding the * and the first group of characters following the *.  When there is this kind of overlap, * will match the longest sequence of characters that match the expression before the *, and *? will match the shortest sequence of characters matching the expression before the *?.  For example, if you have "xxyxyyyxyxyxxyxyxyy" in the text, "x[xy]*x" will match through to the last x (i.e., "xxyxyyyxyxyxxyxyxyy", but "x[xy]*?x" will only match the first two xes ("xxyxyyyxyxyxxyxyxyy").
[] Specifies a group of alternative characters.  A [] expression will match any single character that is specified in the [] expression.  For more on the syntax of [] expressions, see below.
/ Specifies where the break position should go if text matches this expression.  (e.g., "[a-z]*/[:Zs:]*[1-0]" will match if the iterator sees a run of letters, followed by a run of whitespace, followed by a digit, but the break position will actually go before the whitespace).  Expressions that don't contain / put the break position at the end of the matching text.
\ Escape character.  The \ itself is ignored, but causes the next character to be treated as literal character.  This has no effect for many characters, but for the characters listed above, this deprives them of their special meaning.  (There are no special escape sequences for Unicode characters, or tabs and newlines; these are all handled by a higher-level protocol.  In a Java string, "\n" will be converted to a literal newline character by the time the regular-expression parser sees it.  Of course, this means that \ sequences that are visible to the regexp parser must be written as \\ when inside a Java string.)  All characters in the ASCII range except for letters, digits, and control characters are reserved characters to the parser and must be preceded by \ even if they currently don't mean anything.
! If ! appears at the beginning of a regular expression, it tells the regexp parser that this expression specifies the backwards-iteration behavior of the iterator, and not its normal iteration behavior.  This is generally only used in situations where the automatically-generated backwards-iteration brhavior doesn't produce satisfactory results and must be supplemented with extra client-specified rules.
(all others) All other characters are treated as literal characters, which must match the corresponding character(s) in the text exactly.

Within a [] expression, a number of other special characters can be used to specify groups of characters:

- Specifies a range of matching characters.  For example "[a-p]" matches all lowercase Latin letters from a to p (inclusive).  The - sign specifies ranges of continuous Unicode numeric values, not ranges of characters in a language's alphabetical order: "[a-z]" doesn't include capital letters, nor does it include accented letters such as a-umlaut.
:: A pair of colons containing a one- or two-letter code matches all characters in the corresponding Unicode category.  The two-letter codes are the same as the two-letter codes in the Unicode database (for example, "[:Sc::Sm:]" matches all currency symbols and all math symbols).  Specifying a one-letter code is the same as specifying all two-letter codes that begin with that letter (for example, "[:L:]" matches all letters, and is equivalent to "[:Lu::Ll::Lo::Lm::Lt:]").  Anything other than a valid two-letter Unicode category code or a single letter that begins a Unicode category code is illegal within colons.
[] [] expressions can nest.  This has no effect, except when used in conjunction with the ^ token.
^ Excludes the character (or the characters in the [] expression) following it from the group of characters.  For example, "[a-z^p]" matches all Latin lowercase letters except p.  "[:L:^[\u4e00-\u9fff]]" matches all letters except the Han ideographs.
(all others) All other characters are treated as literal characters.  (For example, "[aeiou]" specifies just the letters a, e, i, o, and u.)

For a more complete explanation, see   For examples, see the resource data (which is annotated).

Richard Gillam
$RCSFile$ $Revision: 1.1 $ $Date: 1998/11/05 19:32:04 $

Fields Summary
protected static final byte
A token used as a character-category value to identify ignore characters
private static final short
The state number of the starting state
private static final short
The state-transition value indicating "stop"
static final byte[]
Magic number for the BreakIterator data file format.
static final int
static final byte
Version number of the dictionary that was read in.
private static final int
Header size in byte count
private static final int
An array length of indices for BMP characters
private CompactByteArray
Tables that indexes from character values to character category numbers
private SupplementaryCharacterData
private short[]
The table of state transitions used for forward iteration
private short[]
The table of state transitions used to sync up the iterator with the text in backwards and random-access iteration
private boolean[]
A list of flags indicating which states in the state table are accepting ("end") states
private boolean[]
A list of flags indicating which states in the state table are lookahead states (states which turn lookahead on and off)
private byte[]
A table for additional data. May be used by a subclass of RuleBasedBreakIterator.
private int
The number of character categories (and, thus, the number of columns in the state tables)
private CharacterIterator
The character iterator through which this BreakIterator accesses the text
private long
A CRC32 value of all data in datafile
Constructors Summary
public RuleBasedBreakIterator(String datafile)
Constructs a RuleBasedBreakIterator according to the datafile provided.

    // constructors

Methods Summary
protected static final voidcheckOffset(int offset, java.text.CharacterIterator text)
Throw IllegalArgumentException unless begin <= offset < end.

        if (offset < text.getBeginIndex() || offset >= text.getEndIndex()) {
            throw new IllegalArgumentException("offset out of bounds");
public java.lang.Objectclone()
Clones this iterator.

A newly-constructed RuleBasedBreakIterator with the same behavior as this one.

        RuleBasedBreakIterator result = (RuleBasedBreakIterator) super.clone();
        if (text != null) {
            result.text = (CharacterIterator) text.clone();
        return result;
public intcurrent()
Returns the current iteration position.

The current iteration position.

        return getText().getIndex();
public booleanequals(java.lang.Object that)
Returns true if both BreakIterators are of the same class, have the same rules, and iterate over the same text.

        try {
            if (that == null) {
                return false;

            RuleBasedBreakIterator other = (RuleBasedBreakIterator) that;
            if (checksum != other.checksum) {
                return false;
            if (text == null) {
                return other.text == null;
            } else {
                return text.equals(other.text);
        catch(ClassCastException e) {
            return false;
public intfirst()
Sets the current iteration position to the beginning of the text. (i.e., the CharacterIterator's starting offset).

The offset of the beginning of the text.

        CharacterIterator t = getText();

        return t.getIndex();
public intfollowing(int offset)
Sets the iterator to refer to the first boundary position following the specified position.

The position from which to begin searching for a break position.
The position of the first break after the current position.

        CharacterIterator text = getText();
        checkOffset(offset, text);

        // 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
        if (offset == text.getBeginIndex()) {
            return handleNext();

        // 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.
        int result = handlePrevious();
        while (result != BreakIterator.DONE && result <= offset) {
            result = handleNext();
        return result;

        return additionalData;
Returns current character

        char c1 = text.current();
        if (Character.isHighSurrogate(c1) &&
            text.getIndex() < text.getEndIndex()) {
            char c2 =;
            if (Character.isLowSurrogate(c2)) {
                return Character.toCodePoint(c1, c2);
        return (int)c1;
private intgetCurrentCodePointCount()
Returns the count of next character.

        char c1 = text.current();
        if (Character.isHighSurrogate(c1) &&
            text.getIndex() < text.getEndIndex()) {
            char c2 =;
            if (Character.isLowSurrogate(c2)) {
                return 2;
        return 1;
Returns next character

        int index = text.getIndex();
        int endIndex = text.getEndIndex();
        if (index == endIndex ||
            (index = index + getCurrentCodePointCount()) >= endIndex) {
            return CharacterIterator.DONE;
        return getCurrent();
private intgetNextIndex()
Returns the position of next character.

        int index = text.getIndex() + getCurrentCodePointCount();
        int endIndex = text.getEndIndex();
        if (index > endIndex) {
            return endIndex;
        } else {
            return index;
private intgetPrevious()
Returns previous character

        char c2 = text.previous();
        if (Character.isLowSurrogate(c2) &&
            text.getIndex() > text.getBeginIndex()) {
            char c1 = text.previous();
            if (Character.isHighSurrogate(c1)) {
                return Character.toCodePoint(c1, c2);
            } else {
        return (int)c2;
public java.text.CharacterIteratorgetText()
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.

An iterator over the text being analyzed.

        // 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
        // iterator to return.
        if (text == null) {
            text = new StringCharacterIterator("");
        return text;
protected inthandleNext()
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 a possible end state.

        // if we're already at the end of the text, return DONE.
        CharacterIterator text = getText();
        if (text.getIndex() == text.getEndIndex()) {
            return BreakIterator.DONE;

        // no matter what, we always advance at least one character forward
        int result = getNextIndex();
        int lookaheadResult = 0;

        // begin in state 1
        int state = START_STATE;
        int category;
        int c = getCurrent();

        // loop until we reach the end of the text or transition to state 0
        while (c != CharacterIterator.DONE && state != STOP_STATE) {

            // look up the current character's character category (which tells us
            // which column in the state table to look at)
            category = lookupCategory(c);

            // if the character isn't an ignore character, look up a state
            // transition in the state table
            if (category != IGNORE) {
                state = lookupState(state, category);

            // if the state we've just transitioned to is a lookahead state,
            // (but not also an end state), save its position.  If it's
            // both a lookahead state and an end state, update the break position
            // to the last saved lookup-state position
            if (lookaheadStates[state]) {
                if (endStates[state]) {
                    result = lookaheadResult;
                else {
                    lookaheadResult = getNextIndex();

            // otherwise, if the state we've just transitioned to is an accepting
            // state, update the break position to be the current iteration position
            else {
                if (endStates[state]) {
                    result = getNextIndex();

            c = getNext();

        // if we've run off the end of the text, and the very last character took us into
        // a lookahead state, advance the break position to the lookahead position
        // (the theory here is that if there are no characters at all after the lookahead
        // position, that always matches the lookahead criteria)
        if (c == CharacterIterator.DONE && lookaheadResult == text.getEndIndex()) {
            result = lookaheadResult;

        return result;
protected inthandlePrevious()
This method backs the iterator back up to a "safe position" in the text. This is a position that we know, without any context, must be a break position. The various calling methods then iterate forward from this safe position to the appropriate position to return. (For more information, see the description of buildBackwardsStateTable() in RuleBasedBreakIterator.Builder.)

        CharacterIterator text = getText();
        int state = START_STATE;
        int category = 0;
        int lastCategory = 0;
        int c = getCurrent();

        // loop until we reach the beginning of the text or transition to state 0
        while (c != CharacterIterator.DONE && state != STOP_STATE) {

            // save the last character's category and look up the current
            // character's category
            lastCategory = category;
            category = lookupCategory(c);

            // if the current character isn't an ignore character, look up a
            // state transition in the backwards state table
            if (category != IGNORE) {
                state = lookupBackwardState(state, category);

            // then advance one character backwards
            c = getPrevious();

        // if we didn't march off the beginning of the text, we're either one or two
        // positions away from the real break position.  (One because of the call to
        // previous() at the end of the loop above, and another because the character
        // that takes us into the stop state will always be the character BEFORE
        // the break position.)
        if (c != CharacterIterator.DONE) {
            if (lastCategory != IGNORE) {
            else {
        return text.getIndex();
public inthashCode()
Compute a hashcode for this BreakIterator

A hash code

        return (int)checksum;
public booleanisBoundary(int offset)
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".

offset the offset to check.
True if "offset" is a boundary position.

        CharacterIterator text = getText();
        checkOffset(offset, text);
        if (offset == text.getBeginIndex()) {
            return true;

        // to check whether this is a boundary, 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
        else {
            return following(offset - 1) == offset;
public intlast()
Sets the current iteration position to the end of the text. (i.e., the CharacterIterator's ending offset).

The text's past-the-end offset.

        CharacterIterator t = getText();

        // I'm not sure why, but t.last() returns the offset of the last character,
        // rather than the past-the-end offset
        return t.getIndex();
protected intlookupBackwardState(int state, int category)
Given a current state and a character category, looks up the next state to transition to in the backwards state table.

        return backwardsStateTable[state * numCategories + category];
protected intlookupCategory(int c)
Looks up a character's category (i.e., its category for breaking purposes, not its Unicode category)

        if (c < Character.MIN_SUPPLEMENTARY_CODE_POINT) {
            return charCategoryTable.elementAt((char)c);
        } else {
            return supplementaryCharCategoryTable.getValue(c);
protected intlookupState(int state, int category)
Given a current state and a character category, looks up the next state to transition to in the state table.

        return stateTable[state * numCategories + category];
public intnext(int n)
Advances the iterator either forward or backward the specified number of steps. Negative values move backward, and positive values move forward. This is equivalent to repeatedly calling next() or previous().

n The number of steps to move. The sign indicates the direction (negative is backwards, and positive is forwards).
The character offset of the boundary position n boundaries away from the current one.

        int result = current();
        while (n > 0) {
            result = handleNext();
        while (n < 0) {
            result = previous();
        return result;
public intnext()
Advances the iterator to the next boundary position.

The position of the first boundary after this one.

        return handleNext();
public intpreceding(int offset)
Sets the iterator to refer to the last boundary position before the specified position.

The position to begin searching for a break from.
The position of the last boundary before the starting position.

        // 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
        CharacterIterator text = getText();
        checkOffset(offset, text);
        return previous();
public intprevious()
Advances the iterator backwards, to the last boundary preceding this one.

The position of the last boundary position preceding this one.

        // if we're already sitting at the beginning of the text, return DONE
        CharacterIterator text = getText();
        if (current() == text.getBeginIndex()) {
            return BreakIterator.DONE;

        // 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
        int start = current();
        int lastResult = handlePrevious();
        int result = lastResult;

        // 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
        while (result != BreakIterator.DONE && result < start) {
            lastResult = result;
            result = handleNext();

        // set the current iteration position to be the last break position
        // before where we started, and then return that value
        return lastResult;
protected byte[]readFile(java.lang.String datafile)

        BufferedInputStream is;
        try {
            is = (BufferedInputStream)AccessController.doPrivileged(
                new PrivilegedExceptionAction() {
                    public Object run() throws Exception {
                        return new BufferedInputStream(getClass().getResourceAsStream("/sun/text/resources/" + datafile));
        catch (PrivilegedActionException e) {
            throw new InternalError(e.toString());

        int offset = 0;

        /* First, read magic, version, and header_info. */
        int len = LABEL_LENGTH + 5;
        byte[] buf = new byte[len];
        if ( != len) {
            throw new MissingResourceException("Wrong header length",
                                               datafile, "");

        /* Validate the magic number. */
        for (int i = 0; i < LABEL_LENGTH; i++, offset++) {
            if (buf[offset] != LABEL[offset]) {
                throw new MissingResourceException("Wrong magic number",
                                                   datafile, "");

        /* Validate the version number. */
        if (buf[offset] != supportedVersion) {
            throw new MissingResourceException("Unsupported version(" + buf[offset] + ")",
                                               datafile, "");

        /* Read data: totalDataSize + 8(for checksum) */
        len = BreakIterator.getInt(buf, ++offset);
        buf = new byte[len];
        if ( != len) {
            throw new MissingResourceException("Wrong data length",
                                               datafile, "");


        return buf;
protected voidreadTables(java.lang.String datafile)
Read datafile. The datafile's format is as follows:
BreakIteratorData {
u1 magic[7];
u1 version;
u4 totalDataSize;
header_info header;
body value;
totalDataSize is the summation of the size of header_info and body in byte count.

In header, each field except for checksum implies the length of each field. Since BMPdataLength is a fixed-length data(512 entries), its length isn't included in header. checksum is a CRC32 value of all in body.

header_info {
u4 stateTableLength;
u4 backwardsStateTableLength;
u4 endStatesLength;
u4 lookaheadStatesLength;
u4 BMPdataLength;
u4 nonBMPdataLength;
u4 additionalDataLength;
u8 checksum;

Finally, BMPindices and BMPdata are set to charCategoryTable. nonBMPdata is set to supplementaryCharCategoryTable.

body {
u2 stateTable[stateTableLength];
u2 backwardsStateTable[backwardsStateTableLength];
u1 endStates[endStatesLength];
u1 lookaheadStates[lookaheadStatesLength];
u2 BMPindices[512];
u1 BMPdata[BMPdataLength];
u4 nonBMPdata[numNonBMPdataLength];
u1 additionalData[additionalDataLength];

        byte[] buffer = readFile(datafile);

        /* Read header_info. */
        int stateTableLength = BreakIterator.getInt(buffer, 0);
        int backwardsStateTableLength = BreakIterator.getInt(buffer, 4);
        int endStatesLength = BreakIterator.getInt(buffer, 8);
        int lookaheadStatesLength = BreakIterator.getInt(buffer, 12);
        int BMPdataLength = BreakIterator.getInt(buffer, 16);
        int nonBMPdataLength = BreakIterator.getInt(buffer, 20);
        int additionalDataLength = BreakIterator.getInt(buffer, 24);
        checksum = BreakIterator.getLong(buffer, 28);

        /* Read stateTable[numCategories * numRows] */
        stateTable = new short[stateTableLength];
        int offset = HEADER_LENGTH;
        for (int i = 0; i < stateTableLength; i++, offset+=2) {
           stateTable[i] = BreakIterator.getShort(buffer, offset);

        /* Read backwardsStateTable[numCategories * numRows] */
        backwardsStateTable = new short[backwardsStateTableLength];
        for (int i = 0; i < backwardsStateTableLength; i++, offset+=2) {
           backwardsStateTable[i] = BreakIterator.getShort(buffer, offset);

        /* Read endStates[numRows] */
        endStates = new boolean[endStatesLength];
        for (int i = 0; i < endStatesLength; i++, offset++) {
           endStates[i] = buffer[offset] == 1; 

        /* Read lookaheadStates[numRows] */
        lookaheadStates = new boolean[lookaheadStatesLength];
        for (int i = 0; i < lookaheadStatesLength; i++, offset++) {
           lookaheadStates[i] = buffer[offset] == 1; 

        /* Read a category table and indices for BMP characters. */
        short[] temp1 = new short[BMP_INDICES_LENGTH];  // BMPindices
        for (int i = 0; i < BMP_INDICES_LENGTH; i++, offset+=2) {
            temp1[i] = BreakIterator.getShort(buffer, offset);
        byte[] temp2 = new byte[BMPdataLength];  // BMPdata
        System.arraycopy(buffer, offset, temp2, 0, BMPdataLength);
        offset += BMPdataLength;
        charCategoryTable = new CompactByteArray(temp1, temp2);

        /* Read a category table for non-BMP characters. */
        int[] temp3 = new int[nonBMPdataLength];
        for (int i = 0; i < nonBMPdataLength; i++, offset+=4) {
            temp3[i] = BreakIterator.getInt(buffer, offset);
        supplementaryCharCategoryTable = new SupplementaryCharacterData(temp3);

        /* Read additional data */ 
        if (additionalDataLength > 0) {
            additionalData = new byte[additionalDataLength];
            System.arraycopy(buffer, offset, additionalData, 0, additionalDataLength);

        /* Set numCategories */
        numCategories = stateTable.length / endStates.length;
voidsetAdditionalData(byte[] b)

        additionalData = b;
public voidsetText(java.text.CharacterIterator newText)
Set the iterator to analyze a new piece of text. This function resets the current iteration position to the beginning of the text.

newText An iterator over the text to analyze.

        // Test iterator to see if we need to wrap it in a SafeCharIterator.
        // The correct behavior for CharacterIterators is to allow the
        // position to be set to the endpoint of the iterator.  Many
        // CharacterIterators do not uphold this, so this is a workaround
        // to permit them to use this class.
        int end = newText.getEndIndex();
        boolean goodIterator;
        try {
            newText.setIndex(end);  // some buggy iterators throw an exception here
            goodIterator = newText.getIndex() == end;
        catch(IllegalArgumentException e) {
            goodIterator = false;

        if (goodIterator) {
            text = newText;
        else {
            text = new SafeCharIterator(newText);
public java.lang.StringtoString()
Returns text

        StringBuffer sb = new StringBuffer();
        sb.append("checksum=0x" + Long.toHexString(checksum));
        return sb.toString();