BinaryDictionary.java revision bb57090f1da9d1fc5a0eda9b627d3f8c8b25ab42
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.io.File; 31import java.util.ArrayList; 32import java.util.Arrays; 33import java.util.Locale; 34import java.util.Map; 35 36/** 37 * Implements a static, compacted, binary dictionary of standard words. 38 */ 39// TODO: All methods which should be locked need to have a suffix "Locked". 40public final class BinaryDictionary extends Dictionary { 41 private static final String TAG = BinaryDictionary.class.getSimpleName(); 42 43 // Must be equal to MAX_WORD_LENGTH in native/jni/src/defines.h 44 private static final int MAX_WORD_LENGTH = Constants.DICTIONARY_MAX_WORD_LENGTH; 45 // Must be equal to MAX_RESULTS in native/jni/src/defines.h 46 private static final int MAX_RESULTS = 18; 47 // Required space count for auto commit. 48 // TODO: Remove this heuristic. 49 private static final int SPACE_COUNT_FOR_AUTO_COMMIT = 3; 50 51 @UsedForTesting 52 public static final String UNIGRAM_COUNT_QUERY = "UNIGRAM_COUNT"; 53 @UsedForTesting 54 public static final String BIGRAM_COUNT_QUERY = "BIGRAM_COUNT"; 55 56 private long mNativeDict; 57 private final Locale mLocale; 58 private final long mDictSize; 59 private final String mDictFilePath; 60 private final int[] mInputCodePoints = new int[MAX_WORD_LENGTH]; 61 private final int[] mOutputCodePoints = new int[MAX_WORD_LENGTH * MAX_RESULTS]; 62 private final int[] mSpaceIndices = new int[MAX_RESULTS]; 63 private final int[] mOutputScores = new int[MAX_RESULTS]; 64 private final int[] mOutputTypes = new int[MAX_RESULTS]; 65 // Only one result is ever used 66 private final int[] mOutputAutoCommitFirstWordConfidence = new int[1]; 67 68 private final NativeSuggestOptions mNativeSuggestOptions = new NativeSuggestOptions(); 69 70 private final SparseArray<DicTraverseSession> mDicTraverseSessions = 71 CollectionUtils.newSparseArray(); 72 73 // TODO: There should be a way to remove used DicTraverseSession objects from 74 // {@code mDicTraverseSessions}. 75 private DicTraverseSession getTraverseSession(final int traverseSessionId) { 76 synchronized(mDicTraverseSessions) { 77 DicTraverseSession traverseSession = mDicTraverseSessions.get(traverseSessionId); 78 if (traverseSession == null) { 79 traverseSession = mDicTraverseSessions.get(traverseSessionId); 80 if (traverseSession == null) { 81 traverseSession = new DicTraverseSession(mLocale, mNativeDict, mDictSize); 82 mDicTraverseSessions.put(traverseSessionId, traverseSession); 83 } 84 } 85 return traverseSession; 86 } 87 } 88 89 /** 90 * Constructor for the binary dictionary. This is supposed to be called from the 91 * dictionary factory. 92 * @param filename the name of the file to read through native code. 93 * @param offset the offset of the dictionary data within the file. 94 * @param length the length of the binary data. 95 * @param useFullEditDistance whether to use the full edit distance in suggestions 96 * @param dictType the dictionary type, as a human-readable string 97 * @param isUpdatable whether to open the dictionary file in writable mode. 98 */ 99 public BinaryDictionary(final String filename, final long offset, final long length, 100 final boolean useFullEditDistance, final Locale locale, final String dictType, 101 final boolean isUpdatable) { 102 super(dictType); 103 mLocale = locale; 104 mDictSize = length; 105 mDictFilePath = filename; 106 mNativeSuggestOptions.setUseFullEditDistance(useFullEditDistance); 107 loadDictionary(filename, offset, length, isUpdatable); 108 } 109 110 static { 111 JniUtils.loadNativeLibrary(); 112 } 113 114 private static native boolean createEmptyDictFileNative(String filePath, long dictVersion, 115 String[] attributeKeyStringArray, String[] attributeValueStringArray); 116 private static native long openNative(String sourceDir, long dictOffset, long dictSize, 117 boolean isUpdatable); 118 private static native void flushNative(long dict, String filePath); 119 private static native boolean needsToRunGCNative(long dict, boolean mindsBlockByGC); 120 private static native void flushWithGCNative(long dict, String filePath); 121 private static native void closeNative(long dict); 122 private static native int getProbabilityNative(long dict, int[] word); 123 private static native int getBigramProbabilityNative(long dict, int[] word0, int[] word1); 124 private static native int getSuggestionsNative(long dict, long proximityInfo, 125 long traverseSession, int[] xCoordinates, int[] yCoordinates, int[] times, 126 int[] pointerIds, int[] inputCodePoints, int inputSize, int commitPoint, 127 int[] suggestOptions, int[] prevWordCodePointArray, 128 int[] outputCodePoints, int[] outputScores, int[] outputIndices, int[] outputTypes, 129 int[] outputAutoCommitFirstWordConfidence); 130 private static native float calcNormalizedScoreNative(int[] before, int[] after, int score); 131 private static native int editDistanceNative(int[] before, int[] after); 132 private static native void addUnigramWordNative(long dict, int[] word, int probability); 133 private static native void addBigramWordsNative(long dict, int[] word0, int[] word1, 134 int probability); 135 private static native void removeBigramWordsNative(long dict, int[] word0, int[] word1); 136 private static native int calculateProbabilityNative(long dict, int unigramProbability, 137 int bigramProbability); 138 private static native String getPropertyNative(long dict, String query); 139 140 @UsedForTesting 141 public static boolean createEmptyDictFile(final String filePath, final long dictVersion, 142 final Map<String, String> attributeMap) { 143 final String[] keyArray = new String[attributeMap.size()]; 144 final String[] valueArray = new String[attributeMap.size()]; 145 int index = 0; 146 for (final String key : attributeMap.keySet()) { 147 keyArray[index] = key; 148 valueArray[index] = attributeMap.get(key); 149 index++; 150 } 151 return createEmptyDictFileNative(filePath, dictVersion, keyArray, valueArray); 152 } 153 154 // TODO: Move native dict into session 155 private final void loadDictionary(final String path, final long startOffset, 156 final long length, final boolean isUpdatable) { 157 mNativeDict = openNative(path, startOffset, length, isUpdatable); 158 } 159 160 @Override 161 public ArrayList<SuggestedWordInfo> getSuggestions(final WordComposer composer, 162 final String prevWord, final ProximityInfo proximityInfo, 163 final boolean blockOffensiveWords, final int[] additionalFeaturesOptions) { 164 return getSuggestionsWithSessionId(composer, prevWord, proximityInfo, blockOffensiveWords, 165 additionalFeaturesOptions, 0 /* sessionId */); 166 } 167 168 @Override 169 public ArrayList<SuggestedWordInfo> getSuggestionsWithSessionId(final WordComposer composer, 170 final String prevWord, final ProximityInfo proximityInfo, 171 final boolean blockOffensiveWords, final int[] additionalFeaturesOptions, 172 final int sessionId) { 173 if (!isValidDictionary()) return null; 174 175 Arrays.fill(mInputCodePoints, Constants.NOT_A_CODE); 176 // TODO: toLowerCase in the native code 177 final int[] prevWordCodePointArray = (null == prevWord) 178 ? null : StringUtils.toCodePointArray(prevWord); 179 final int composerSize = composer.size(); 180 181 final boolean isGesture = composer.isBatchMode(); 182 if (composerSize <= 1 || !isGesture) { 183 if (composerSize > MAX_WORD_LENGTH - 1) return null; 184 for (int i = 0; i < composerSize; i++) { 185 mInputCodePoints[i] = composer.getCodeAt(i); 186 } 187 } 188 189 final InputPointers ips = composer.getInputPointers(); 190 final int inputSize = isGesture ? ips.getPointerSize() : composerSize; 191 mNativeSuggestOptions.setIsGesture(isGesture); 192 mNativeSuggestOptions.setAdditionalFeaturesOptions(additionalFeaturesOptions); 193 // proximityInfo and/or prevWordForBigrams may not be null. 194 final int count = getSuggestionsNative(mNativeDict, proximityInfo.getNativeProximityInfo(), 195 getTraverseSession(sessionId).getSession(), ips.getXCoordinates(), 196 ips.getYCoordinates(), ips.getTimes(), ips.getPointerIds(), mInputCodePoints, 197 inputSize, 0 /* commitPoint */, mNativeSuggestOptions.getOptions(), 198 prevWordCodePointArray, mOutputCodePoints, mOutputScores, mSpaceIndices, 199 mOutputTypes, mOutputAutoCommitFirstWordConfidence); 200 final ArrayList<SuggestedWordInfo> suggestions = CollectionUtils.newArrayList(); 201 for (int j = 0; j < count; ++j) { 202 final int start = j * MAX_WORD_LENGTH; 203 int len = 0; 204 while (len < MAX_WORD_LENGTH && mOutputCodePoints[start + len] != 0) { 205 ++len; 206 } 207 if (len > 0) { 208 final int flags = mOutputTypes[j] & SuggestedWordInfo.KIND_MASK_FLAGS; 209 if (blockOffensiveWords 210 && 0 != (flags & SuggestedWordInfo.KIND_FLAG_POSSIBLY_OFFENSIVE) 211 && 0 == (flags & SuggestedWordInfo.KIND_FLAG_EXACT_MATCH)) { 212 // If we block potentially offensive words, and if the word is possibly 213 // offensive, then we don't output it unless it's also an exact match. 214 continue; 215 } 216 final int kind = mOutputTypes[j] & SuggestedWordInfo.KIND_MASK_KIND; 217 final int score = SuggestedWordInfo.KIND_WHITELIST == kind 218 ? SuggestedWordInfo.MAX_SCORE : mOutputScores[j]; 219 // TODO: check that all users of the `kind' parameter are ready to accept 220 // flags too and pass mOutputTypes[j] instead of kind 221 suggestions.add(new SuggestedWordInfo(new String(mOutputCodePoints, start, len), 222 score, kind, this /* sourceDict */, 223 mSpaceIndices[j] /* indexOfTouchPointOfSecondWord */, 224 mOutputAutoCommitFirstWordConfidence[0])); 225 } 226 } 227 return suggestions; 228 } 229 230 public boolean isValidDictionary() { 231 return mNativeDict != 0; 232 } 233 234 public static float calcNormalizedScore(final String before, final String after, 235 final int score) { 236 return calcNormalizedScoreNative(StringUtils.toCodePointArray(before), 237 StringUtils.toCodePointArray(after), score); 238 } 239 240 public static int editDistance(final String before, final String after) { 241 if (before == null || after == null) { 242 throw new IllegalArgumentException(); 243 } 244 return editDistanceNative(StringUtils.toCodePointArray(before), 245 StringUtils.toCodePointArray(after)); 246 } 247 248 @Override 249 public boolean isValidWord(final String word) { 250 return getFrequency(word) != NOT_A_PROBABILITY; 251 } 252 253 @Override 254 public int getFrequency(final String word) { 255 if (word == null) return NOT_A_PROBABILITY; 256 int[] codePoints = StringUtils.toCodePointArray(word); 257 return getProbabilityNative(mNativeDict, codePoints); 258 } 259 260 // TODO: Add a batch process version (isValidBigramMultiple?) to avoid excessive numbers of jni 261 // calls when checking for changes in an entire dictionary. 262 public boolean isValidBigram(final String word0, final String word1) { 263 return getBigramProbability(word0, word1) != NOT_A_PROBABILITY; 264 } 265 266 public int getBigramProbability(final String word0, final String word1) { 267 if (TextUtils.isEmpty(word0) || TextUtils.isEmpty(word1)) return NOT_A_PROBABILITY; 268 final int[] codePoints0 = StringUtils.toCodePointArray(word0); 269 final int[] codePoints1 = StringUtils.toCodePointArray(word1); 270 return getBigramProbabilityNative(mNativeDict, codePoints0, codePoints1); 271 } 272 273 private void runGCIfRequired() { 274 if (needsToRunGC(true /* mindsBlockByGC */)) { 275 flushWithGC(); 276 } 277 } 278 279 // Add a unigram entry to binary dictionary in native code. 280 public void addUnigramWord(final String word, final int probability) { 281 if (TextUtils.isEmpty(word)) { 282 return; 283 } 284 runGCIfRequired(); 285 final int[] codePoints = StringUtils.toCodePointArray(word); 286 addUnigramWordNative(mNativeDict, codePoints, probability); 287 } 288 289 // Add a bigram entry to binary dictionary in native code. 290 public void addBigramWords(final String word0, final String word1, final int probability) { 291 if (TextUtils.isEmpty(word0) || TextUtils.isEmpty(word1)) { 292 return; 293 } 294 runGCIfRequired(); 295 final int[] codePoints0 = StringUtils.toCodePointArray(word0); 296 final int[] codePoints1 = StringUtils.toCodePointArray(word1); 297 addBigramWordsNative(mNativeDict, codePoints0, codePoints1, probability); 298 } 299 300 // Remove a bigram entry form binary dictionary in native code. 301 public void removeBigramWords(final String word0, final String word1) { 302 if (TextUtils.isEmpty(word0) || TextUtils.isEmpty(word1)) { 303 return; 304 } 305 runGCIfRequired(); 306 final int[] codePoints0 = StringUtils.toCodePointArray(word0); 307 final int[] codePoints1 = StringUtils.toCodePointArray(word1); 308 removeBigramWordsNative(mNativeDict, codePoints0, codePoints1); 309 } 310 311 private void reopen() { 312 close(); 313 final File dictFile = new File(mDictFilePath); 314 mNativeDict = openNative(dictFile.getAbsolutePath(), 0 /* startOffset */, 315 dictFile.length(), true /* isUpdatable */); 316 } 317 318 public void flush() { 319 if (!isValidDictionary()) return; 320 flushNative(mNativeDict, mDictFilePath); 321 reopen(); 322 } 323 324 public void flushWithGC() { 325 if (!isValidDictionary()) return; 326 flushWithGCNative(mNativeDict, mDictFilePath); 327 reopen(); 328 } 329 330 /** 331 * Checks whether GC is needed to run or not. 332 * @param mindsBlockByGC Whether to mind operations blocked by GC. We don't need to care about 333 * the blocking in some situations such as in idle time or just before closing. 334 * @return whether GC is needed to run or not. 335 */ 336 public boolean needsToRunGC(final boolean mindsBlockByGC) { 337 if (!isValidDictionary()) return false; 338 return needsToRunGCNative(mNativeDict, mindsBlockByGC); 339 } 340 341 @UsedForTesting 342 public int calculateProbability(final int unigramProbability, final int bigramProbability) { 343 if (!isValidDictionary()) return NOT_A_PROBABILITY; 344 return calculateProbabilityNative(mNativeDict, unigramProbability, bigramProbability); 345 } 346 347 @UsedForTesting 348 public String getPropertyForTests(String query) { 349 if (!isValidDictionary()) return ""; 350 return getPropertyNative(mNativeDict, query); 351 } 352 353 @Override 354 public boolean shouldAutoCommit(final SuggestedWordInfo candidate) { 355 // TODO: actually use the confidence rather than use this completely broken heuristic 356 final String word = candidate.mWord; 357 final int length = word.length(); 358 int remainingSpaces = SPACE_COUNT_FOR_AUTO_COMMIT; 359 for (int i = 0; i < length; ++i) { 360 // This is okay because no low-surrogate and no high-surrogate can ever match the 361 // space character, so we don't need to take care of iterating on code points. 362 if (Constants.CODE_SPACE == word.charAt(i)) { 363 if (0 >= --remainingSpaces) return true; 364 } 365 } 366 return false; 367 } 368 369 @Override 370 public void close() { 371 synchronized (mDicTraverseSessions) { 372 final int sessionsSize = mDicTraverseSessions.size(); 373 for (int index = 0; index < sessionsSize; ++index) { 374 final DicTraverseSession traverseSession = mDicTraverseSessions.valueAt(index); 375 if (traverseSession != null) { 376 traverseSession.close(); 377 } 378 } 379 mDicTraverseSessions.clear(); 380 } 381 closeInternalLocked(); 382 } 383 384 private synchronized void closeInternalLocked() { 385 if (mNativeDict != 0) { 386 closeNative(mNativeDict); 387 mNativeDict = 0; 388 } 389 } 390 391 // TODO: Manage BinaryDictionary instances without using WeakReference or something. 392 @Override 393 protected void finalize() throws Throwable { 394 try { 395 closeInternalLocked(); 396 } finally { 397 super.finalize(); 398 } 399 } 400} 401