/*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation. Oracle designates this
* particular file as subject to the "Classpath" exception as provided
* by Oracle in the LICENSE file that accompanied this code.
*
* This code is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
* version 2 for more details (a copy is included in the LICENSE file that
* accompanied this code).
*
* You should have received a copy of the GNU General Public License version
* 2 along with this work; if not, write to the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
*
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
* or visit www.oracle.com if you need additional information or have any
* questions.
*/
/*
*
* (C) Copyright Taligent, Inc. 1996, 1997 - All Rights Reserved
* (C) Copyright IBM Corp. 1996 - 2002 - All Rights Reserved
*
* The original version of this source code and documentation
* is copyrighted and owned by Taligent, Inc., a wholly-owned
* subsidiary of IBM. These materials are provided under terms
* of a License Agreement between Taligent and Sun. This technology
* is protected by multiple US and International patents.
*
* This notice and attribution to Taligent may not be removed.
* Taligent is a registered trademark of Taligent, Inc.
*/
/**
* A subclass of RuleBasedBreakIterator that adds the ability to use a dictionary
* to further subdivide ranges of text beyond what is possible using just the
* state-table-based algorithm. This is necessary, for example, to handle
* word and line breaking in Thai, which doesn't use spaces between words. The
* state-table-based algorithm used by RuleBasedBreakIterator is used to divide
* up text as far as possible, and then contiguous ranges of letters are
* repeatedly compared against a list of known words (i.e., the dictionary)
* to divide them up into words.
*
* DictionaryBasedBreakIterator uses the same rule language as RuleBasedBreakIterator,
* but adds one more special substitution name: <dictionary>. This substitution
* name is used to identify characters in words in the dictionary. The idea is that
* if the iterator passes over a chunk of text that includes two or more characters
* in a row that are included in <dictionary>, it goes back through that range and
* derives additional break positions (if possible) using the dictionary.
*
* DictionaryBasedBreakIterator is also constructed with the filename of a dictionary
* file. It follows a prescribed search path to locate the dictionary (right now,
* and won't find it in JAR files, but this location is likely to change). The
* dictionary file is in a serialized binary format. We have a very primitive (and
* slow) BuildDictionaryFile utility for creating dictionary files, but aren't
* currently making it public. Contact us for help.
*/
/**
* a list of known words that is used to divide up contiguous ranges of letters,
* stored in a compressed, indexed, format that offers fast access
*/
/**
* a list of flags indicating which character categories are contained in
* the dictionary file (this is used to determine which ranges of characters
* to apply the dictionary to)
*/
private boolean[] categoryFlags;
/**
* a temporary hiding place for the number of dictionary characters in the
* last range passed over by next()
*/
private int dictionaryCharCount;
/**
* when a range of characters is divided up using the dictionary, the break
* positions that are discovered are stored here, preventing us from having
* to use either the dictionary or the state table again until the iterator
* leaves this range of text
*/
private int[] cachedBreakPositions;
/**
* if cachedBreakPositions is not null, this indicates which item in the
* cache the current iteration position refers to
*/
private int positionInCache;
/**
* Constructs a DictionaryBasedBreakIterator.
* @param description Same as the description parameter on RuleBasedBreakIterator,
* except for the special meaning of "<dictionary>". This parameter is just
* passed through to RuleBasedBreakIterator's constructor.
* @param dictionaryFilename The filename of the dictionary file to use
*/
throws IOException {
super(dataFile);
byte[] tmp = super.getAdditionalData();
super.setAdditionalData(null);
}
}
}
}
dictionaryCharCount = 0;
positionInCache = 0;
}
/**
* 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.
*/
public int first() {
dictionaryCharCount = 0;
positionInCache = 0;
return super.first();
}
/**
* 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.
*/
public int last() {
dictionaryCharCount = 0;
positionInCache = 0;
return super.last();
}
/**
* Advances the iterator one step backwards.
* @return The position of the last boundary position before the
* current iteration position
*/
public int previous() {
// if we have cached break positions and we're still in the range
// covered by them, just move one step backward in the cache
return cachedBreakPositions[positionInCache];
}
// otherwise, dump the cache and use the inherited previous() method to move
// backward. This may fill up the cache with new break positions, in which
// case we have to mark our position in the cache
else {
if (cachedBreakPositions != null) {
}
return result;
}
}
/**
* Sets the current iteration position to the last boundary position
* before the specified position.
* @param offset The position to begin searching from
* @return The position of the last boundary before "offset"
*/
// if we have no cached break positions, or "offset" is outside the
// range covered by the cache, we can just call the inherited routine
// (which will eventually call other routines in this class that may
// refresh the cache)
}
// on the other hand, if "offset" is within the range covered by the cache,
// then all we have to do is search the cache for the last break position
// before "offset"
else {
positionInCache = 0;
}
}
}
/**
* Sets the current iteration position to the first boundary position after
* the specified position.
* @param offset The position to begin searching forward from
* @return The position of the first boundary after "offset"
*/
// if we have no cached break positions, or if "offset" is outside the
// range covered by the cache, then dump the cache and call our
// inherited following() method. This will call other methods in this
// class that may refresh the cache.
}
// on the other hand, if "offset" is within the range covered by the
// cache, then just search the cache for the first break position
// after "offset"
else {
positionInCache = 0;
}
}
}
/**
* This is the implementation function for next().
*/
protected int handleNext() {
// if there are no cached break positions, or if we've just moved
// off the end of the range covered by the cache, we have to dump
// and possibly regenerate the cache
if (cachedBreakPositions == null ||
// start by using the inherited handleNext() to find a tentative return
// value. dictionaryCharCount tells us how many dictionary characters
// we passed over on our way to the tentative return value
dictionaryCharCount = 0;
int result = super.handleNext();
// if we passed over more than one dictionary character, then we use
// divideUpDictionaryRange() to regenerate the cached break positions
// for the new range
}
// otherwise, the value we got back from the inherited fuction
// is our return value, and we can dump the cache
else {
return result;
}
}
// if the cache of break positions has been regenerated (or existed all
// along), then just advance to the next break position in the cache
// and return it
if (cachedBreakPositions != null) {
return cachedBreakPositions[positionInCache];
}
return -9999; // SHOULD NEVER GET HERE!
}
/**
* Looks up a character category for a character.
*/
protected int lookupCategory(int c) {
// this override of lookupCategory() exists only to keep track of whether we've
// passed over any dictionary characters. It calls the inherited lookupCategory()
// to do the real work, and then checks whether its return value is one of the
// categories represented in the dictionary. If it is, bump the dictionary-
// character count.
int result = super.lookupCategory(c);
}
return result;
}
/**
* This is the function that actually implements the dictionary-based
* algorithm. Given the endpoints of a range of text, it uses the
* dictionary to determine the positions of any boundaries in this
* range. It stores all the boundary positions it discovers in
* cachedBreakPositions so that we only have to do this work once
* for each time we enter the range.
*/
// the range we're dividing may begin or end with non-dictionary characters
// (i.e., for line breaking, we may have leading or trailing punctuation
// that needs to be kept with the word). Seek from the beginning of the
// range to the first dictionary character
int c = getCurrent();
int category = lookupCategory(c);
c = getNext();
category = lookupCategory(c);
}
// initialize. We maintain two stacks: currentBreakPositions contains
// the list of break positions that will be returned if we successfully
// finish traversing the whole range now. possibleBreakPositions lists
// all other possible word ends we've passed along the way. (Whenever
// we reach an error [a sequence of characters that can't begin any word
// in the dictionary], we back up, possibly delete some breaks from
// currentBreakPositions, move a break from possibleBreakPositions
// to currentBreakPositions, and start over from there. This process
// continues in this way until we either successfully make it all the way
// across the range, or exhaust all of our combinations of break
// positions.)
// the dictionary is implemented as a trie, which is treated as a state
// machine. -1 represents the end of a legal word. Every word in the
// dictionary is represented by a path from the root node to -1. A path
// that ends in state 0 is an illegal combination of characters.
int state = 0;
// these two variables are used for error handling. We keep track of the
// farthest we've gotten through the range being divided, and the combination
// of breaks that got us that far. If we use up all possible break
// combinations, the text contains an error or a word that's not in the
// dictionary. In this case, we "bless" the break positions that got us the
// farthest as real break positions, and then start over from scratch with
// the character where the error occurred.
// initialize (we always exit the loop with a break statement)
c = getCurrent();
while (true) {
// if we can transition to state "-1" from our current state, we're
// on the last character of a legal word. Push that position onto
// the possible-break-positions stack
}
// look up the new state to transition to in the dictionary
// if the character we're sitting on causes us to transition to
// the "end of word" state, then it was a non-dictionary character
// and we've successfully traversed the whole range. Drop out
// of the loop.
if (state == -1) {
break;
}
// if the character we're sitting on causes us to transition to
// the error state, or if we've gone off the end of the range
// without transitioning to the "end of word" state, we've hit
// an error...
// if this is the farthest we've gotten, take note of it in
// case there's an error in the text
}
// wrongBreakPositions is a list of all break positions
// we've tried starting that didn't allow us to traverse
// all the way through the text. Every time we pop a
//break position off of currentBreakPositions, we put it
// into wrongBreakPositions to avoid trying it again later.
// If we make it to this spot, we're either going to back
// up to a break in possibleBreakPositions and try starting
// over from there, or we've exhausted all possible break
// positions and are going to do the fallback procedure.
// This loop prevents us from messing with anything in
// possibleBreakPositions that didn't work as a starting
// point the last time we tried it (this is to prevent a bunch of
// repetitive checks from slowing down some extreme cases)
possibleBreakPositions.peek())) {
}
// if we've used up all possible break-position combinations, there's
// an error or an unknown word in the text. In this case, we start
// over, treating the farthest character we've reached as the beginning
// of the range, and "blessing" the break positions that got us that
// far as real break positions
if (possibleBreakPositions.isEmpty()) {
if (bestBreakPositions != null) {
if (farthestEndPoint < endPos) {
}
else {
break;
}
}
else {
}
getNext();
}
}
// if we still have more break positions we can try, then promote the
// last break in possibleBreakPositions into currentBreakPositions,
// and get rid of all entries in currentBreakPositions that come after
// it. Then back up to that position and start over from there (i.e.,
// treat that position as the beginning of a new word)
else {
}
}
// re-sync "c" for the next go-round, and drop out of the loop if
// we've made it off the end of the range
c = getCurrent();
break;
}
}
// if we didn't hit any exceptional conditions on this last iteration,
// just advance to the next character and loop
else {
c = getNext();
}
}
// dump the last break position in the list, and replace it with the actual
// end of the range (which may be the same character, or may be further on
// because the range actually ended with non-dictionary characters we want to
// keep with the word)
if (!currentBreakPositions.isEmpty()) {
}
// create a regular array to hold the break positions and copy
// the break positions from the stack to the array (in addition,
// our starting position goes into this array as a break position).
// This array becomes the cache of break positions used by next()
// and previous(), so this is where we actually refresh the cache.
}
positionInCache = 0;
}
}