BinaryDictionary.java revision 1250cdeab2da2c7f7d9c732a816dbc36e49fd61a
1/* 2 * Copyright (C) 2008 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); you may not 5 * use this file except in compliance with the License. You may obtain a copy of 6 * the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT 12 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the 13 * License for the specific language governing permissions and limitations under 14 * the License. 15 */ 16 17package com.android.inputmethod.latin; 18 19import android.content.Context; 20import android.text.TextUtils; 21 22import com.android.inputmethod.keyboard.ProximityInfo; 23import com.android.inputmethod.latin.SuggestedWords.SuggestedWordInfo; 24 25import java.util.ArrayList; 26import java.util.Arrays; 27import java.util.Locale; 28 29/** 30 * Implements a static, compacted, binary dictionary of standard words. 31 */ 32public class BinaryDictionary extends Dictionary { 33 34 public static final String DICTIONARY_PACK_AUTHORITY = 35 "com.android.inputmethod.latin.dictionarypack"; 36 37 /** 38 * There is a difference between what java and native code can handle. 39 * This value should only be used in BinaryDictionary.java 40 * It is necessary to keep it at this value because some languages e.g. German have 41 * really long words. 42 */ 43 public static final int MAX_WORD_LENGTH = 48; 44 public static final int MAX_WORDS = 18; 45 public static final int MAX_SPACES = 16; 46 47 private static final String TAG = "BinaryDictionary"; 48 private static final int MAX_BIGRAMS = 60; 49 private static final int MAX_RESULTS = MAX_BIGRAMS > MAX_WORDS ? MAX_BIGRAMS : MAX_WORDS; 50 51 private static final int TYPED_LETTER_MULTIPLIER = 2; 52 53 private long mNativeDict; 54 private final int[] mInputCodes = new int[MAX_WORD_LENGTH]; 55 private final char[] mOutputChars = new char[MAX_WORD_LENGTH * MAX_RESULTS]; 56 private final int[] mSpaceIndices = new int[MAX_SPACES]; 57 private final int[] mOutputScores = new int[MAX_RESULTS]; 58 59 private final boolean mUseFullEditDistance; 60 61 /** 62 * Constructor for the binary dictionary. This is supposed to be called from the 63 * dictionary factory. 64 * All implementations should pass null into flagArray, except for testing purposes. 65 * @param context the context to access the environment from. 66 * @param filename the name of the file to read through native code. 67 * @param offset the offset of the dictionary data within the file. 68 * @param length the length of the binary data. 69 * @param useFullEditDistance whether to use the full edit distance in suggestions 70 * @param dictType the dictionary type, as a human-readable string 71 */ 72 public BinaryDictionary(final Context context, 73 final String filename, final long offset, final long length, 74 final boolean useFullEditDistance, final Locale locale, final String dictType) { 75 super(dictType); 76 mUseFullEditDistance = useFullEditDistance; 77 loadDictionary(filename, offset, length); 78 } 79 80 static { 81 JniUtils.loadNativeLibrary(); 82 } 83 84 private native long openNative(String sourceDir, long dictOffset, long dictSize, 85 int typedLetterMultiplier, int fullWordMultiplier, int maxWordLength, int maxWords); 86 private native void closeNative(long dict); 87 private native int getFrequencyNative(long dict, int[] word, int wordLength); 88 private native boolean isValidBigramNative(long dict, int[] word1, int[] word2); 89 private native int getSuggestionsNative(long dict, long proximityInfo, int[] xCoordinates, 90 int[] yCoordinates, int[] times, int[] pointerIds, int[] inputCodes, int codesSize, 91 int commitPoint, boolean isGesture, 92 int[] prevWordCodePointArray, boolean useFullEditDistance, char[] outputChars, 93 int[] scores, int[] outputIndices); 94 private native int getBigramsNative(long dict, int[] prevWord, int prevWordLength, 95 int[] inputCodes, int inputCodesLength, char[] outputChars, int[] scores, 96 int maxWordLength, int maxBigrams); 97 private static native float calcNormalizedScoreNative( 98 char[] before, int beforeLength, char[] after, int afterLength, int score); 99 private static native int editDistanceNative( 100 char[] before, int beforeLength, char[] after, int afterLength); 101 102 private final void loadDictionary(String path, long startOffset, long length) { 103 mNativeDict = openNative(path, startOffset, length, 104 TYPED_LETTER_MULTIPLIER, FULL_WORD_SCORE_MULTIPLIER, MAX_WORD_LENGTH, MAX_WORDS); 105 } 106 107 @Override 108 public ArrayList<SuggestedWordInfo> getSuggestions(final WordComposer composer, 109 final CharSequence prevWord, final ProximityInfo proximityInfo) { 110 if (!isValidDictionary()) return null; 111 Arrays.fill(mInputCodes, WordComposer.NOT_A_CODE); 112 Arrays.fill(mOutputChars, (char) 0); 113 Arrays.fill(mOutputScores, 0); 114 final boolean isGesture = composer.isBatchMode(); 115 final boolean isValidPrevWord = (prevWord == null) 116 // TODO: Remove this check. Unfortunately the current gesture recognition code has 117 // a double-free bug. 118 && !isGesture; 119 // TODO: toLowerCase in the native code 120 final int[] prevWordCodePointArray = isValidPrevWord 121 ? StringUtils.toCodePointArray(prevWord.toString()) : null; 122 final int composerSize = composer.size(); 123 124 if (composerSize <= 1 || !composer.isBatchMode()) { 125 if (composerSize > MAX_WORD_LENGTH - 1) return null; 126 for (int i = 0; i < composerSize; i++) { 127 mInputCodes[i] = composer.getCodeAt(i); 128 } 129 } 130 131 final int count; 132 if (!isGesture && composer.size() <= 1) { 133 if (TextUtils.isEmpty(prevWord)) return null; 134 int tmpCount = getBigramsNative(mNativeDict, prevWordCodePointArray, 135 prevWordCodePointArray.length, mInputCodes, composerSize, 136 mOutputChars, mOutputScores, MAX_WORD_LENGTH, MAX_BIGRAMS); 137 count = Math.min(tmpCount, MAX_BIGRAMS); 138 } else { 139 final InputPointers ips = composer.getInputPointers(); 140 final int codesSize; 141 if (isGesture) { 142 codesSize = ips.getPointerSize(); 143 } else { 144 codesSize = composer.size(); 145 } 146 147 // proximityInfo and/or prevWordForBigrams may not be null. 148 count = getSuggestionsNative(mNativeDict, proximityInfo.getNativeProximityInfo(), 149 ips.getXCoordinates(), ips.getYCoordinates(), ips.getTimes(), ips.getPointerIds(), 150 mInputCodes, codesSize, 0 /* unused */, isGesture, prevWordCodePointArray, 151 mUseFullEditDistance, mOutputChars, mOutputScores, mSpaceIndices); 152 } 153 final ArrayList<SuggestedWordInfo> suggestions = new ArrayList<SuggestedWordInfo>(); 154 for (int j = 0; j < count; ++j) { 155 if (composer.size() > 0 && mOutputScores[j] < 1) break; 156 final int start = j * MAX_WORD_LENGTH; 157 int len = 0; 158 while (len < MAX_WORD_LENGTH && mOutputChars[start + len] != 0) { 159 ++len; 160 } 161 if (len > 0) { 162 suggestions.add(new SuggestedWordInfo( 163 new String(mOutputChars, start, len), 164 mOutputScores[j], SuggestedWordInfo.KIND_CORRECTION, mDictType)); 165 } 166 } 167 return suggestions; 168 } 169 170 /* package for test */ boolean isValidDictionary() { 171 return mNativeDict != 0; 172 } 173 174 public static float calcNormalizedScore(String before, String after, int score) { 175 return calcNormalizedScoreNative(before.toCharArray(), before.length(), 176 after.toCharArray(), after.length(), score); 177 } 178 179 public static int editDistance(String before, String after) { 180 return editDistanceNative( 181 before.toCharArray(), before.length(), after.toCharArray(), after.length()); 182 } 183 184 @Override 185 public boolean isValidWord(CharSequence word) { 186 return getFrequency(word) >= 0; 187 } 188 189 @Override 190 public int getFrequency(CharSequence word) { 191 if (word == null) return -1; 192 int[] chars = StringUtils.toCodePointArray(word.toString()); 193 return getFrequencyNative(mNativeDict, chars, chars.length); 194 } 195 196 // TODO: Add a batch process version (isValidBigramMultiple?) to avoid excessive numbers of jni 197 // calls when checking for changes in an entire dictionary. 198 public boolean isValidBigram(CharSequence word1, CharSequence word2) { 199 if (TextUtils.isEmpty(word1) || TextUtils.isEmpty(word2)) return false; 200 int[] chars1 = StringUtils.toCodePointArray(word1.toString()); 201 int[] chars2 = StringUtils.toCodePointArray(word2.toString()); 202 return isValidBigramNative(mNativeDict, chars1, chars2); 203 } 204 205 @Override 206 public synchronized void close() { 207 closeInternal(); 208 } 209 210 private void closeInternal() { 211 if (mNativeDict != 0) { 212 closeNative(mNativeDict); 213 mNativeDict = 0; 214 } 215 } 216 217 @Override 218 protected void finalize() throws Throwable { 219 try { 220 closeInternal(); 221 } finally { 222 super.finalize(); 223 } 224 } 225} 226