BinaryDictionary.java revision 060917ed9abff0e777e6b99e43cc4b826d1ac0c8
1/* 2 * Copyright (C) 2008 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of 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, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17package com.android.inputmethod.latin; 18 19import android.text.TextUtils; 20import android.util.SparseArray; 21 22import com.android.inputmethod.annotations.UsedForTesting; 23import com.android.inputmethod.keyboard.ProximityInfo; 24import com.android.inputmethod.latin.SuggestedWords.SuggestedWordInfo; 25import com.android.inputmethod.latin.settings.NativeSuggestOptions; 26import com.android.inputmethod.latin.utils.CollectionUtils; 27import com.android.inputmethod.latin.utils.JniUtils; 28import com.android.inputmethod.latin.utils.StringUtils; 29 30import java.util.ArrayList; 31import java.util.Arrays; 32import java.util.Locale; 33 34/** 35 * Implements a static, compacted, binary dictionary of standard words. 36 */ 37public final class BinaryDictionary extends Dictionary { 38 private static final String TAG = BinaryDictionary.class.getSimpleName(); 39 40 // Must be equal to MAX_WORD_LENGTH in native/jni/src/defines.h 41 private static final int MAX_WORD_LENGTH = Constants.DICTIONARY_MAX_WORD_LENGTH; 42 // Must be equal to MAX_RESULTS in native/jni/src/defines.h 43 private static final int MAX_RESULTS = 18; 44 // Required space count for auto commit. 45 // TODO: Remove this heuristic. 46 private static final int SPACE_COUNT_FOR_AUTO_COMMIT = 3; 47 48 private long mNativeDict; 49 private final Locale mLocale; 50 private final long mDictSize; 51 private final String mDictFilePath; 52 private final int[] mInputCodePoints = new int[MAX_WORD_LENGTH]; 53 private final int[] mOutputCodePoints = new int[MAX_WORD_LENGTH * MAX_RESULTS]; 54 private final int[] mSpaceIndices = new int[MAX_RESULTS]; 55 private final int[] mOutputScores = new int[MAX_RESULTS]; 56 private final int[] mOutputTypes = new int[MAX_RESULTS]; 57 private final int[] mOutputAutoCommitFirstWordConfidence = new int[MAX_RESULTS]; 58 59 private final NativeSuggestOptions mNativeSuggestOptions = new NativeSuggestOptions(); 60 61 private final SparseArray<DicTraverseSession> mDicTraverseSessions = 62 CollectionUtils.newSparseArray(); 63 64 // TODO: There should be a way to remove used DicTraverseSession objects from 65 // {@code mDicTraverseSessions}. 66 private DicTraverseSession getTraverseSession(final int traverseSessionId) { 67 synchronized(mDicTraverseSessions) { 68 DicTraverseSession traverseSession = mDicTraverseSessions.get(traverseSessionId); 69 if (traverseSession == null) { 70 traverseSession = mDicTraverseSessions.get(traverseSessionId); 71 if (traverseSession == null) { 72 traverseSession = new DicTraverseSession(mLocale, mNativeDict, mDictSize); 73 mDicTraverseSessions.put(traverseSessionId, traverseSession); 74 } 75 } 76 return traverseSession; 77 } 78 } 79 80 /** 81 * Constructor for the binary dictionary. This is supposed to be called from the 82 * dictionary factory. 83 * @param filename the name of the file to read through native code. 84 * @param offset the offset of the dictionary data within the file. 85 * @param length the length of the binary data. 86 * @param useFullEditDistance whether to use the full edit distance in suggestions 87 * @param dictType the dictionary type, as a human-readable string 88 * @param isUpdatable whether to open the dictionary file in writable mode. 89 */ 90 public BinaryDictionary(final String filename, final long offset, final long length, 91 final boolean useFullEditDistance, final Locale locale, final String dictType, 92 final boolean isUpdatable) { 93 super(dictType); 94 mLocale = locale; 95 mDictSize = length; 96 mDictFilePath = filename; 97 mNativeSuggestOptions.setUseFullEditDistance(useFullEditDistance); 98 loadDictionary(filename, offset, length, isUpdatable); 99 } 100 101 static { 102 JniUtils.loadNativeLibrary(); 103 } 104 105 private static native long openNative(String sourceDir, long dictOffset, long dictSize, 106 boolean isUpdatable); 107 private static native void flushNative(long dict, String filePath); 108 private static native boolean needsToRunGCNative(long dict); 109 private static native void flushWithGCNative(long dict, String filePath); 110 private static native void closeNative(long dict); 111 private static native int getProbabilityNative(long dict, int[] word); 112 private static native boolean isValidBigramNative(long dict, int[] word0, int[] word1); 113 private static native int getSuggestionsNative(long dict, long proximityInfo, 114 long traverseSession, int[] xCoordinates, int[] yCoordinates, int[] times, 115 int[] pointerIds, int[] inputCodePoints, int inputSize, int commitPoint, 116 int[] suggestOptions, int[] prevWordCodePointArray, 117 int[] outputCodePoints, int[] outputScores, int[] outputIndices, int[] outputTypes, 118 int[] outputAutoCommitFirstWordConfidence); 119 private static native float calcNormalizedScoreNative(int[] before, int[] after, int score); 120 private static native int editDistanceNative(int[] before, int[] after); 121 private static native void addUnigramWordNative(long dict, int[] word, int probability); 122 private static native void addBigramWordsNative(long dict, int[] word0, int[] word1, 123 int probability); 124 private static native void removeBigramWordsNative(long dict, int[] word0, int[] word1); 125 126 // TODO: Move native dict into session 127 private final void loadDictionary(final String path, final long startOffset, 128 final long length, final boolean isUpdatable) { 129 mNativeDict = openNative(path, startOffset, length, isUpdatable); 130 } 131 132 @Override 133 public ArrayList<SuggestedWordInfo> getSuggestions(final WordComposer composer, 134 final String prevWord, final ProximityInfo proximityInfo, 135 final boolean blockOffensiveWords, final int[] additionalFeaturesOptions) { 136 return getSuggestionsWithSessionId(composer, prevWord, proximityInfo, blockOffensiveWords, 137 additionalFeaturesOptions, 0 /* sessionId */); 138 } 139 140 @Override 141 public ArrayList<SuggestedWordInfo> getSuggestionsWithSessionId(final WordComposer composer, 142 final String prevWord, final ProximityInfo proximityInfo, 143 final boolean blockOffensiveWords, final int[] additionalFeaturesOptions, 144 final int sessionId) { 145 if (!isValidDictionary()) return null; 146 147 Arrays.fill(mInputCodePoints, Constants.NOT_A_CODE); 148 // TODO: toLowerCase in the native code 149 final int[] prevWordCodePointArray = (null == prevWord) 150 ? null : StringUtils.toCodePointArray(prevWord); 151 final int composerSize = composer.size(); 152 153 final boolean isGesture = composer.isBatchMode(); 154 if (composerSize <= 1 || !isGesture) { 155 if (composerSize > MAX_WORD_LENGTH - 1) return null; 156 for (int i = 0; i < composerSize; i++) { 157 mInputCodePoints[i] = composer.getCodeAt(i); 158 } 159 } 160 161 final InputPointers ips = composer.getInputPointers(); 162 final int inputSize = isGesture ? ips.getPointerSize() : composerSize; 163 mNativeSuggestOptions.setIsGesture(isGesture); 164 mNativeSuggestOptions.setAdditionalFeaturesOptions(additionalFeaturesOptions); 165 // proximityInfo and/or prevWordForBigrams may not be null. 166 final int count = getSuggestionsNative(mNativeDict, proximityInfo.getNativeProximityInfo(), 167 getTraverseSession(sessionId).getSession(), ips.getXCoordinates(), 168 ips.getYCoordinates(), ips.getTimes(), ips.getPointerIds(), mInputCodePoints, 169 inputSize, 0 /* commitPoint */, mNativeSuggestOptions.getOptions(), 170 prevWordCodePointArray, mOutputCodePoints, mOutputScores, mSpaceIndices, 171 mOutputTypes, mOutputAutoCommitFirstWordConfidence); 172 final ArrayList<SuggestedWordInfo> suggestions = CollectionUtils.newArrayList(); 173 for (int j = 0; j < count; ++j) { 174 final int start = j * MAX_WORD_LENGTH; 175 int len = 0; 176 while (len < MAX_WORD_LENGTH && mOutputCodePoints[start + len] != 0) { 177 ++len; 178 } 179 if (len > 0) { 180 final int flags = mOutputTypes[j] & SuggestedWordInfo.KIND_MASK_FLAGS; 181 if (blockOffensiveWords 182 && 0 != (flags & SuggestedWordInfo.KIND_FLAG_POSSIBLY_OFFENSIVE) 183 && 0 == (flags & SuggestedWordInfo.KIND_FLAG_EXACT_MATCH)) { 184 // If we block potentially offensive words, and if the word is possibly 185 // offensive, then we don't output it unless it's also an exact match. 186 continue; 187 } 188 final int kind = mOutputTypes[j] & SuggestedWordInfo.KIND_MASK_KIND; 189 final int score = SuggestedWordInfo.KIND_WHITELIST == kind 190 ? SuggestedWordInfo.MAX_SCORE : mOutputScores[j]; 191 // TODO: check that all users of the `kind' parameter are ready to accept 192 // flags too and pass mOutputTypes[j] instead of kind 193 suggestions.add(new SuggestedWordInfo(new String(mOutputCodePoints, start, len), 194 score, kind, this /* sourceDict */, 195 mSpaceIndices[j] /* indexOfTouchPointOfSecondWord */, 196 mOutputAutoCommitFirstWordConfidence[0])); 197 } 198 } 199 return suggestions; 200 } 201 202 public boolean isValidDictionary() { 203 return mNativeDict != 0; 204 } 205 206 public static float calcNormalizedScore(final String before, final String after, 207 final int score) { 208 return calcNormalizedScoreNative(StringUtils.toCodePointArray(before), 209 StringUtils.toCodePointArray(after), score); 210 } 211 212 public static int editDistance(final String before, final String after) { 213 if (before == null || after == null) { 214 throw new IllegalArgumentException(); 215 } 216 return editDistanceNative(StringUtils.toCodePointArray(before), 217 StringUtils.toCodePointArray(after)); 218 } 219 220 @Override 221 public boolean isValidWord(final String word) { 222 return getFrequency(word) >= 0; 223 } 224 225 @Override 226 public int getFrequency(final String word) { 227 if (word == null) return -1; 228 int[] codePoints = StringUtils.toCodePointArray(word); 229 return getProbabilityNative(mNativeDict, codePoints); 230 } 231 232 // TODO: Add a batch process version (isValidBigramMultiple?) to avoid excessive numbers of jni 233 // calls when checking for changes in an entire dictionary. 234 public boolean isValidBigram(final String word0, final String word1) { 235 if (TextUtils.isEmpty(word0) || TextUtils.isEmpty(word1)) return false; 236 final int[] codePoints0 = StringUtils.toCodePointArray(word0); 237 final int[] codePoints1 = StringUtils.toCodePointArray(word1); 238 return isValidBigramNative(mNativeDict, codePoints0, codePoints1); 239 } 240 241 // Add a unigram entry to binary dictionary in native code. 242 public void addUnigramWord(final String word, final int probability) { 243 if (TextUtils.isEmpty(word)) { 244 return; 245 } 246 final int[] codePoints = StringUtils.toCodePointArray(word); 247 addUnigramWordNative(mNativeDict, codePoints, probability); 248 } 249 250 // Add a bigram entry to binary dictionary in native code. 251 public void addBigramWords(final String word0, final String word1, final int probability) { 252 if (TextUtils.isEmpty(word0) || TextUtils.isEmpty(word1)) { 253 return; 254 } 255 final int[] codePoints0 = StringUtils.toCodePointArray(word0); 256 final int[] codePoints1 = StringUtils.toCodePointArray(word1); 257 addBigramWordsNative(mNativeDict, codePoints0, codePoints1, probability); 258 } 259 260 // Remove a bigram entry form binary dictionary in native code. 261 public void removeBigramWords(final String word0, final String word1) { 262 if (TextUtils.isEmpty(word0) || TextUtils.isEmpty(word1)) { 263 return; 264 } 265 final int[] codePoints0 = StringUtils.toCodePointArray(word0); 266 final int[] codePoints1 = StringUtils.toCodePointArray(word1); 267 removeBigramWordsNative(mNativeDict, codePoints0, codePoints1); 268 } 269 270 @UsedForTesting 271 public void flush() { 272 if (!isValidDictionary()) return; 273 flushNative(mNativeDict, mDictFilePath); 274 } 275 276 @UsedForTesting 277 public void flushWithGC() { 278 if (!isValidDictionary()) return; 279 flushWithGCNative(mNativeDict, mDictFilePath); 280 } 281 282 @UsedForTesting 283 public boolean needsToRunGC() { 284 if (!isValidDictionary()) return false; 285 return needsToRunGCNative(mNativeDict); 286 } 287 288 @Override 289 public boolean shouldAutoCommit(final SuggestedWordInfo candidate) { 290 // TODO: actually use the confidence rather than use this completely broken heuristic 291 final String word = candidate.mWord; 292 final int length = word.length(); 293 int remainingSpaces = SPACE_COUNT_FOR_AUTO_COMMIT; 294 for (int i = 0; i < length; ++i) { 295 // This is okay because no low-surrogate and no high-surrogate can ever match the 296 // space character, so we don't need to take care of iterating on code points. 297 if (Constants.CODE_SPACE == word.charAt(i)) { 298 if (0 >= --remainingSpaces) return true; 299 } 300 } 301 return false; 302 } 303 304 @Override 305 public void close() { 306 synchronized (mDicTraverseSessions) { 307 final int sessionsSize = mDicTraverseSessions.size(); 308 for (int index = 0; index < sessionsSize; ++index) { 309 final DicTraverseSession traverseSession = mDicTraverseSessions.valueAt(index); 310 if (traverseSession != null) { 311 traverseSession.close(); 312 } 313 } 314 } 315 closeInternal(); 316 } 317 318 private synchronized void closeInternal() { 319 if (mNativeDict != 0) { 320 closeNative(mNativeDict); 321 mNativeDict = 0; 322 } 323 } 324 325 @Override 326 protected void finalize() throws Throwable { 327 try { 328 closeInternal(); 329 } finally { 330 super.finalize(); 331 } 332 } 333} 334