AndroidSpellCheckerService.java revision d9df8094e625c92f6a66ebe9048afc7f88c14b33
1/* 2 * Copyright (C) 2011 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.spellcheck; 18 19import android.content.Intent; 20import android.content.SharedPreferences; 21import android.preference.PreferenceManager; 22import android.service.textservice.SpellCheckerService; 23import android.text.TextUtils; 24import android.util.Log; 25import android.util.LruCache; 26import android.view.textservice.SentenceSuggestionsInfo; 27import android.view.textservice.SuggestionsInfo; 28import android.view.textservice.TextInfo; 29 30import com.android.inputmethod.compat.SuggestionsInfoCompatUtils; 31import com.android.inputmethod.keyboard.ProximityInfo; 32import com.android.inputmethod.latin.BinaryDictionary; 33import com.android.inputmethod.latin.ContactsBinaryDictionary; 34import com.android.inputmethod.latin.Dictionary; 35import com.android.inputmethod.latin.Dictionary.WordCallback; 36import com.android.inputmethod.latin.DictionaryCollection; 37import com.android.inputmethod.latin.DictionaryFactory; 38import com.android.inputmethod.latin.LocaleUtils; 39import com.android.inputmethod.latin.R; 40import com.android.inputmethod.latin.StringUtils; 41import com.android.inputmethod.latin.SuggestedWords.SuggestedWordInfo; 42import com.android.inputmethod.latin.SynchronouslyLoadedContactsBinaryDictionary; 43import com.android.inputmethod.latin.SynchronouslyLoadedUserBinaryDictionary; 44import com.android.inputmethod.latin.UserBinaryDictionary; 45import com.android.inputmethod.latin.WhitelistDictionary; 46import com.android.inputmethod.latin.WordComposer; 47 48import java.lang.ref.WeakReference; 49import java.util.ArrayList; 50import java.util.Arrays; 51import java.util.Collections; 52import java.util.HashSet; 53import java.util.Iterator; 54import java.util.Locale; 55import java.util.Map; 56import java.util.TreeMap; 57 58/** 59 * Service for spell checking, using LatinIME's dictionaries and mechanisms. 60 */ 61public class AndroidSpellCheckerService extends SpellCheckerService 62 implements SharedPreferences.OnSharedPreferenceChangeListener { 63 private static final String TAG = AndroidSpellCheckerService.class.getSimpleName(); 64 private static final boolean DBG = false; 65 private static final int POOL_SIZE = 2; 66 67 public static final String PREF_USE_CONTACTS_KEY = "pref_spellcheck_use_contacts"; 68 69 private static final int CAPITALIZE_NONE = 0; // No caps, or mixed case 70 private static final int CAPITALIZE_FIRST = 1; // First only 71 private static final int CAPITALIZE_ALL = 2; // All caps 72 73 private final static String[] EMPTY_STRING_ARRAY = new String[0]; 74 private Map<String, DictionaryPool> mDictionaryPools = 75 Collections.synchronizedMap(new TreeMap<String, DictionaryPool>()); 76 private Map<String, UserBinaryDictionary> mUserDictionaries = 77 Collections.synchronizedMap(new TreeMap<String, UserBinaryDictionary>()); 78 private Map<String, Dictionary> mWhitelistDictionaries = 79 Collections.synchronizedMap(new TreeMap<String, Dictionary>()); 80 private ContactsBinaryDictionary mContactsDictionary; 81 82 // The threshold for a candidate to be offered as a suggestion. 83 private float mSuggestionThreshold; 84 // The threshold for a suggestion to be considered "recommended". 85 private float mRecommendedThreshold; 86 // Whether to use the contacts dictionary 87 private boolean mUseContactsDictionary; 88 private final Object mUseContactsLock = new Object(); 89 90 private final HashSet<WeakReference<DictionaryCollection>> mDictionaryCollectionsList = 91 new HashSet<WeakReference<DictionaryCollection>>(); 92 93 public static final int SCRIPT_LATIN = 0; 94 public static final int SCRIPT_CYRILLIC = 1; 95 private static final String SINGLE_QUOTE = "\u0027"; 96 private static final String APOSTROPHE = "\u2019"; 97 private static final TreeMap<String, Integer> mLanguageToScript; 98 static { 99 // List of the supported languages and their associated script. We won't check 100 // words written in another script than the selected script, because we know we 101 // don't have those in our dictionary so we will underline everything and we 102 // will never have any suggestions, so it makes no sense checking them, and this 103 // is done in {@link #shouldFilterOut}. Also, the script is used to choose which 104 // proximity to pass to the dictionary descent algorithm. 105 // IMPORTANT: this only contains languages - do not write countries in there. 106 // Only the language is searched from the map. 107 mLanguageToScript = new TreeMap<String, Integer>(); 108 mLanguageToScript.put("en", SCRIPT_LATIN); 109 mLanguageToScript.put("fr", SCRIPT_LATIN); 110 mLanguageToScript.put("de", SCRIPT_LATIN); 111 mLanguageToScript.put("nl", SCRIPT_LATIN); 112 mLanguageToScript.put("cs", SCRIPT_LATIN); 113 mLanguageToScript.put("es", SCRIPT_LATIN); 114 mLanguageToScript.put("it", SCRIPT_LATIN); 115 mLanguageToScript.put("hr", SCRIPT_LATIN); 116 mLanguageToScript.put("pt", SCRIPT_LATIN); 117 mLanguageToScript.put("ru", SCRIPT_CYRILLIC); 118 // TODO: Make a persian proximity, and activate the Farsi subtype. 119 // mLanguageToScript.put("fa", SCRIPT_PERSIAN); 120 } 121 122 @Override public void onCreate() { 123 super.onCreate(); 124 mSuggestionThreshold = 125 Float.parseFloat(getString(R.string.spellchecker_suggestion_threshold_value)); 126 mRecommendedThreshold = 127 Float.parseFloat(getString(R.string.spellchecker_recommended_threshold_value)); 128 final SharedPreferences prefs = PreferenceManager.getDefaultSharedPreferences(this); 129 prefs.registerOnSharedPreferenceChangeListener(this); 130 onSharedPreferenceChanged(prefs, PREF_USE_CONTACTS_KEY); 131 } 132 133 private static int getScriptFromLocale(final Locale locale) { 134 final Integer script = mLanguageToScript.get(locale.getLanguage()); 135 if (null == script) { 136 throw new RuntimeException("We have been called with an unsupported language: \"" 137 + locale.getLanguage() + "\". Framework bug?"); 138 } 139 return script; 140 } 141 142 @Override 143 public void onSharedPreferenceChanged(final SharedPreferences prefs, final String key) { 144 if (!PREF_USE_CONTACTS_KEY.equals(key)) return; 145 synchronized(mUseContactsLock) { 146 mUseContactsDictionary = prefs.getBoolean(PREF_USE_CONTACTS_KEY, true); 147 if (mUseContactsDictionary) { 148 startUsingContactsDictionaryLocked(); 149 } else { 150 stopUsingContactsDictionaryLocked(); 151 } 152 } 153 } 154 155 private void startUsingContactsDictionaryLocked() { 156 if (null == mContactsDictionary) { 157 // TODO: use the right locale for each session 158 mContactsDictionary = 159 new SynchronouslyLoadedContactsBinaryDictionary(this, Locale.getDefault()); 160 } 161 final Iterator<WeakReference<DictionaryCollection>> iterator = 162 mDictionaryCollectionsList.iterator(); 163 while (iterator.hasNext()) { 164 final WeakReference<DictionaryCollection> dictRef = iterator.next(); 165 final DictionaryCollection dict = dictRef.get(); 166 if (null == dict) { 167 iterator.remove(); 168 } else { 169 dict.addDictionary(mContactsDictionary); 170 } 171 } 172 } 173 174 private void stopUsingContactsDictionaryLocked() { 175 if (null == mContactsDictionary) return; 176 final Dictionary contactsDict = mContactsDictionary; 177 // TODO: revert to the concrete type when USE_BINARY_CONTACTS_DICTIONARY is no longer needed 178 mContactsDictionary = null; 179 final Iterator<WeakReference<DictionaryCollection>> iterator = 180 mDictionaryCollectionsList.iterator(); 181 while (iterator.hasNext()) { 182 final WeakReference<DictionaryCollection> dictRef = iterator.next(); 183 final DictionaryCollection dict = dictRef.get(); 184 if (null == dict) { 185 iterator.remove(); 186 } else { 187 dict.removeDictionary(contactsDict); 188 } 189 } 190 contactsDict.close(); 191 } 192 193 @Override 194 public Session createSession() { 195 return new AndroidSpellCheckerSession(this); 196 } 197 198 private static SuggestionsInfo getNotInDictEmptySuggestions() { 199 return new SuggestionsInfo(0, EMPTY_STRING_ARRAY); 200 } 201 202 private static SuggestionsInfo getInDictEmptySuggestions() { 203 return new SuggestionsInfo(SuggestionsInfo.RESULT_ATTR_IN_THE_DICTIONARY, 204 EMPTY_STRING_ARRAY); 205 } 206 207 // TODO: remove this class when WordCallback is finally out of the picture and 208 // replace it by storage local to the session. 209 private static class SuggestionsGatherer implements WordCallback { 210 public static class Result { 211 public final String[] mSuggestions; 212 public final boolean mHasRecommendedSuggestions; 213 public Result(final String[] gatheredSuggestions, 214 final boolean hasRecommendedSuggestions) { 215 mSuggestions = gatheredSuggestions; 216 mHasRecommendedSuggestions = hasRecommendedSuggestions; 217 } 218 } 219 220 private final ArrayList<CharSequence> mSuggestions; 221 private final int[] mScores; 222 private final String mOriginalText; 223 private final float mSuggestionThreshold; 224 private final float mRecommendedThreshold; 225 private final int mMaxLength; 226 private int mLength = 0; 227 228 // The two following attributes are only ever filled if the requested max length 229 // is 0 (or less, which is treated the same). 230 private String mBestSuggestion = null; 231 private int mBestScore = Integer.MIN_VALUE; // As small as possible 232 233 SuggestionsGatherer(final String originalText, final float suggestionThreshold, 234 final float recommendedThreshold, final int maxLength) { 235 mOriginalText = originalText; 236 mSuggestionThreshold = suggestionThreshold; 237 mRecommendedThreshold = recommendedThreshold; 238 mMaxLength = maxLength; 239 mSuggestions = new ArrayList<CharSequence>(maxLength + 1); 240 mScores = new int[mMaxLength]; 241 } 242 243 @Override 244 synchronized public boolean addWord(char[] word, int[] spaceIndices, int wordOffset, 245 int wordLength, int score, int dicTypeId, int dataType) { 246 return true; 247 } 248 249 synchronized public boolean oldAddWord(char[] word, int[] spaceIndices, int wordOffset, 250 int wordLength, int score, int dicTypeId /* unused */, int dataType) { 251 final int positionIndex = Arrays.binarySearch(mScores, 0, mLength, score); 252 // binarySearch returns the index if the element exists, and -<insertion index> - 1 253 // if it doesn't. See documentation for binarySearch. 254 final int insertIndex = positionIndex >= 0 ? positionIndex : -positionIndex - 1; 255 256 if (insertIndex == 0 && mLength >= mMaxLength) { 257 // In the future, we may want to keep track of the best suggestion score even if 258 // we are asked for 0 suggestions. In this case, we can use the following 259 // (tested) code to keep it: 260 // If the maxLength is 0 (should never be less, but if it is, it's treated as 0) 261 // then we need to keep track of the best suggestion in mBestScore and 262 // mBestSuggestion. This is so that we know whether the best suggestion makes 263 // the score cutoff, since we need to know that to return a meaningful 264 // looksLikeTypo. 265 // if (0 >= mMaxLength) { 266 // if (score > mBestScore) { 267 // mBestScore = score; 268 // mBestSuggestion = new String(word, wordOffset, wordLength); 269 // } 270 // } 271 return true; 272 } 273 if (insertIndex >= mMaxLength) { 274 // We found a suggestion, but its score is too weak to be kept considering 275 // the suggestion limit. 276 return true; 277 } 278 279 // Compute the normalized score and skip this word if it's normalized score does not 280 // make the threshold. 281 final String wordString = new String(word, wordOffset, wordLength); 282 final float normalizedScore = 283 BinaryDictionary.calcNormalizedScore(mOriginalText, wordString, score); 284 if (normalizedScore < mSuggestionThreshold) { 285 if (DBG) Log.i(TAG, wordString + " does not make the score threshold"); 286 return true; 287 } 288 289 if (mLength < mMaxLength) { 290 final int copyLen = mLength - insertIndex; 291 ++mLength; 292 System.arraycopy(mScores, insertIndex, mScores, insertIndex + 1, copyLen); 293 mSuggestions.add(insertIndex, wordString); 294 } else { 295 System.arraycopy(mScores, 1, mScores, 0, insertIndex); 296 mSuggestions.add(insertIndex, wordString); 297 mSuggestions.remove(0); 298 } 299 mScores[insertIndex] = score; 300 301 return true; 302 } 303 304 public Result getResults(final int capitalizeType, final Locale locale) { 305 final String[] gatheredSuggestions; 306 final boolean hasRecommendedSuggestions; 307 if (0 == mLength) { 308 // Either we found no suggestions, or we found some BUT the max length was 0. 309 // If we found some mBestSuggestion will not be null. If it is null, then 310 // we found none, regardless of the max length. 311 if (null == mBestSuggestion) { 312 gatheredSuggestions = null; 313 hasRecommendedSuggestions = false; 314 } else { 315 gatheredSuggestions = EMPTY_STRING_ARRAY; 316 final float normalizedScore = BinaryDictionary.calcNormalizedScore( 317 mOriginalText, mBestSuggestion, mBestScore); 318 hasRecommendedSuggestions = (normalizedScore > mRecommendedThreshold); 319 } 320 } else { 321 if (DBG) { 322 if (mLength != mSuggestions.size()) { 323 Log.e(TAG, "Suggestion size is not the same as stored mLength"); 324 } 325 for (int i = mLength - 1; i >= 0; --i) { 326 Log.i(TAG, "" + mScores[i] + " " + mSuggestions.get(i)); 327 } 328 } 329 Collections.reverse(mSuggestions); 330 StringUtils.removeDupes(mSuggestions); 331 if (CAPITALIZE_ALL == capitalizeType) { 332 for (int i = 0; i < mSuggestions.size(); ++i) { 333 // get(i) returns a CharSequence which is actually a String so .toString() 334 // should return the same object. 335 mSuggestions.set(i, mSuggestions.get(i).toString().toUpperCase(locale)); 336 } 337 } else if (CAPITALIZE_FIRST == capitalizeType) { 338 for (int i = 0; i < mSuggestions.size(); ++i) { 339 // Likewise 340 mSuggestions.set(i, StringUtils.toTitleCase( 341 mSuggestions.get(i).toString(), locale)); 342 } 343 } 344 // This returns a String[], while toArray() returns an Object[] which cannot be cast 345 // into a String[]. 346 gatheredSuggestions = mSuggestions.toArray(EMPTY_STRING_ARRAY); 347 348 final int bestScore = mScores[mLength - 1]; 349 final CharSequence bestSuggestion = mSuggestions.get(0); 350 final float normalizedScore = 351 BinaryDictionary.calcNormalizedScore( 352 mOriginalText, bestSuggestion.toString(), bestScore); 353 hasRecommendedSuggestions = (normalizedScore > mRecommendedThreshold); 354 if (DBG) { 355 Log.i(TAG, "Best suggestion : " + bestSuggestion + ", score " + bestScore); 356 Log.i(TAG, "Normalized score = " + normalizedScore 357 + " (threshold " + mRecommendedThreshold 358 + ") => hasRecommendedSuggestions = " + hasRecommendedSuggestions); 359 } 360 } 361 return new Result(gatheredSuggestions, hasRecommendedSuggestions); 362 } 363 } 364 365 @Override 366 public boolean onUnbind(final Intent intent) { 367 closeAllDictionaries(); 368 return false; 369 } 370 371 private void closeAllDictionaries() { 372 final Map<String, DictionaryPool> oldPools = mDictionaryPools; 373 mDictionaryPools = Collections.synchronizedMap(new TreeMap<String, DictionaryPool>()); 374 final Map<String, UserBinaryDictionary> oldUserDictionaries = mUserDictionaries; 375 mUserDictionaries = 376 Collections.synchronizedMap(new TreeMap<String, UserBinaryDictionary>()); 377 final Map<String, Dictionary> oldWhitelistDictionaries = mWhitelistDictionaries; 378 mWhitelistDictionaries = Collections.synchronizedMap(new TreeMap<String, Dictionary>()); 379 new Thread("spellchecker_close_dicts") { 380 @Override 381 public void run() { 382 for (DictionaryPool pool : oldPools.values()) { 383 pool.close(); 384 } 385 for (Dictionary dict : oldUserDictionaries.values()) { 386 dict.close(); 387 } 388 for (Dictionary dict : oldWhitelistDictionaries.values()) { 389 dict.close(); 390 } 391 synchronized (mUseContactsLock) { 392 if (null != mContactsDictionary) { 393 // The synchronously loaded contacts dictionary should have been in one 394 // or several pools, but it is shielded against multiple closing and it's 395 // safe to call it several times. 396 final ContactsBinaryDictionary dictToClose = mContactsDictionary; 397 // TODO: revert to the concrete type when USE_BINARY_CONTACTS_DICTIONARY 398 // is no longer needed 399 mContactsDictionary = null; 400 dictToClose.close(); 401 } 402 } 403 } 404 }.start(); 405 } 406 407 private DictionaryPool getDictionaryPool(final String locale) { 408 DictionaryPool pool = mDictionaryPools.get(locale); 409 if (null == pool) { 410 final Locale localeObject = LocaleUtils.constructLocaleFromString(locale); 411 pool = new DictionaryPool(POOL_SIZE, this, localeObject); 412 mDictionaryPools.put(locale, pool); 413 } 414 return pool; 415 } 416 417 public DictAndProximity createDictAndProximity(final Locale locale) { 418 final int script = getScriptFromLocale(locale); 419 final ProximityInfo proximityInfo = ProximityInfo.createSpellCheckerProximityInfo( 420 SpellCheckerProximityInfo.getProximityForScript(script), 421 SpellCheckerProximityInfo.ROW_SIZE, 422 SpellCheckerProximityInfo.PROXIMITY_GRID_WIDTH, 423 SpellCheckerProximityInfo.PROXIMITY_GRID_HEIGHT); 424 final DictionaryCollection dictionaryCollection = 425 DictionaryFactory.createMainDictionaryFromManager(this, locale, 426 true /* useFullEditDistance */); 427 final String localeStr = locale.toString(); 428 UserBinaryDictionary userDictionary = mUserDictionaries.get(localeStr); 429 if (null == userDictionary) { 430 userDictionary = new SynchronouslyLoadedUserBinaryDictionary(this, localeStr, true); 431 mUserDictionaries.put(localeStr, userDictionary); 432 } 433 dictionaryCollection.addDictionary(userDictionary); 434 Dictionary whitelistDictionary = mWhitelistDictionaries.get(localeStr); 435 if (null == whitelistDictionary) { 436 whitelistDictionary = new WhitelistDictionary(this, locale); 437 mWhitelistDictionaries.put(localeStr, whitelistDictionary); 438 } 439 dictionaryCollection.addDictionary(whitelistDictionary); 440 synchronized (mUseContactsLock) { 441 if (mUseContactsDictionary) { 442 if (null == mContactsDictionary) { 443 // TODO: use the right locale. We can't do it right now because the 444 // spell checker is reusing the contacts dictionary across sessions 445 // without regard for their locale, so we need to fix that first. 446 mContactsDictionary = new SynchronouslyLoadedContactsBinaryDictionary(this, 447 Locale.getDefault()); 448 } 449 } 450 dictionaryCollection.addDictionary(mContactsDictionary); 451 mDictionaryCollectionsList.add( 452 new WeakReference<DictionaryCollection>(dictionaryCollection)); 453 } 454 return new DictAndProximity(dictionaryCollection, proximityInfo); 455 } 456 457 // This method assumes the text is not empty or null. 458 private static int getCapitalizationType(String text) { 459 // If the first char is not uppercase, then the word is either all lower case, 460 // and in either case we return CAPITALIZE_NONE. 461 if (!Character.isUpperCase(text.codePointAt(0))) return CAPITALIZE_NONE; 462 final int len = text.length(); 463 int capsCount = 1; 464 for (int i = 1; i < len; i = text.offsetByCodePoints(i, 1)) { 465 if (1 != capsCount && i != capsCount) break; 466 if (Character.isUpperCase(text.codePointAt(i))) ++capsCount; 467 } 468 // We know the first char is upper case. So we want to test if either everything 469 // else is lower case, or if everything else is upper case. If the string is 470 // exactly one char long, then we will arrive here with capsCount 1, and this is 471 // correct, too. 472 if (1 == capsCount) return CAPITALIZE_FIRST; 473 return (len == capsCount ? CAPITALIZE_ALL : CAPITALIZE_NONE); 474 } 475 476 private static class AndroidSpellCheckerSession extends Session { 477 // Immutable, but need the locale which is not available in the constructor yet 478 private DictionaryPool mDictionaryPool; 479 // Likewise 480 private Locale mLocale; 481 // Cache this for performance 482 private int mScript; // One of SCRIPT_LATIN or SCRIPT_CYRILLIC for now. 483 484 private final AndroidSpellCheckerService mService; 485 486 private final SuggestionsCache mSuggestionsCache = new SuggestionsCache(); 487 488 private static class SuggestionsParams { 489 public final String[] mSuggestions; 490 public final int mFlags; 491 public SuggestionsParams(String[] suggestions, int flags) { 492 mSuggestions = suggestions; 493 mFlags = flags; 494 } 495 } 496 497 private static class SuggestionsCache { 498 private static final char CHAR_DELIMITER = '\uFFFC'; 499 private static final int MAX_CACHE_SIZE = 50; 500 private final LruCache<String, SuggestionsParams> mUnigramSuggestionsInfoCache = 501 new LruCache<String, SuggestionsParams>(MAX_CACHE_SIZE); 502 503 // TODO: Support n-gram input 504 private static String generateKey(String query, String prevWord) { 505 if (TextUtils.isEmpty(query) || TextUtils.isEmpty(prevWord)) { 506 return query; 507 } 508 return query + CHAR_DELIMITER + prevWord; 509 } 510 511 // TODO: Support n-gram input 512 public SuggestionsParams getSuggestionsFromCache(String query, String prevWord) { 513 return mUnigramSuggestionsInfoCache.get(generateKey(query, prevWord)); 514 } 515 516 // TODO: Support n-gram input 517 public void putSuggestionsToCache( 518 String query, String prevWord, String[] suggestions, int flags) { 519 if (suggestions == null || TextUtils.isEmpty(query)) { 520 return; 521 } 522 mUnigramSuggestionsInfoCache.put( 523 generateKey(query, prevWord), new SuggestionsParams(suggestions, flags)); 524 } 525 } 526 527 AndroidSpellCheckerSession(final AndroidSpellCheckerService service) { 528 mService = service; 529 } 530 531 @Override 532 public void onCreate() { 533 final String localeString = getLocale(); 534 mDictionaryPool = mService.getDictionaryPool(localeString); 535 mLocale = LocaleUtils.constructLocaleFromString(localeString); 536 mScript = getScriptFromLocale(mLocale); 537 } 538 539 /* 540 * Returns whether the code point is a letter that makes sense for the specified 541 * locale for this spell checker. 542 * The dictionaries supported by Latin IME are described in res/xml/spellchecker.xml 543 * and is limited to EFIGS languages and Russian. 544 * Hence at the moment this explicitly tests for Cyrillic characters or Latin characters 545 * as appropriate, and explicitly excludes CJK, Arabic and Hebrew characters. 546 */ 547 private static boolean isLetterCheckableByLanguage(final int codePoint, 548 final int script) { 549 switch (script) { 550 case SCRIPT_LATIN: 551 // Our supported latin script dictionaries (EFIGS) at the moment only include 552 // characters in the C0, C1, Latin Extended A and B, IPA extensions unicode 553 // blocks. As it happens, those are back-to-back in the code range 0x40 to 0x2AF, 554 // so the below is a very efficient way to test for it. As for the 0-0x3F, it's 555 // excluded from isLetter anyway. 556 return codePoint <= 0x2AF && Character.isLetter(codePoint); 557 case SCRIPT_CYRILLIC: 558 // All Cyrillic characters are in the 400~52F block. There are some in the upper 559 // Unicode range, but they are archaic characters that are not used in modern 560 // russian and are not used by our dictionary. 561 return codePoint >= 0x400 && codePoint <= 0x52F && Character.isLetter(codePoint); 562 default: 563 // Should never come here 564 throw new RuntimeException("Impossible value of script: " + script); 565 } 566 } 567 568 /** 569 * Finds out whether a particular string should be filtered out of spell checking. 570 * 571 * This will loosely match URLs, numbers, symbols. To avoid always underlining words that 572 * we know we will never recognize, this accepts a script identifier that should be one 573 * of the SCRIPT_* constants defined above, to rule out quickly characters from very 574 * different languages. 575 * 576 * @param text the string to evaluate. 577 * @param script the identifier for the script this spell checker recognizes 578 * @return true if we should filter this text out, false otherwise 579 */ 580 private static boolean shouldFilterOut(final String text, final int script) { 581 if (TextUtils.isEmpty(text) || text.length() <= 1) return true; 582 583 // TODO: check if an equivalent processing can't be done more quickly with a 584 // compiled regexp. 585 // Filter by first letter 586 final int firstCodePoint = text.codePointAt(0); 587 // Filter out words that don't start with a letter or an apostrophe 588 if (!isLetterCheckableByLanguage(firstCodePoint, script) 589 && '\'' != firstCodePoint) return true; 590 591 // Filter contents 592 final int length = text.length(); 593 int letterCount = 0; 594 for (int i = 0; i < length; i = text.offsetByCodePoints(i, 1)) { 595 final int codePoint = text.codePointAt(i); 596 // Any word containing a '@' is probably an e-mail address 597 // Any word containing a '/' is probably either an ad-hoc combination of two 598 // words or a URI - in either case we don't want to spell check that 599 if ('@' == codePoint || '/' == codePoint) return true; 600 if (isLetterCheckableByLanguage(codePoint, script)) ++letterCount; 601 } 602 // Guestimate heuristic: perform spell checking if at least 3/4 of the characters 603 // in this word are letters 604 return (letterCount * 4 < length * 3); 605 } 606 607 private SentenceSuggestionsInfo fixWronglyInvalidatedWordWithSingleQuote( 608 TextInfo ti, SentenceSuggestionsInfo ssi) { 609 final String typedText = ti.getText(); 610 if (!typedText.contains(SINGLE_QUOTE)) { 611 return null; 612 } 613 final int N = ssi.getSuggestionsCount(); 614 final ArrayList<Integer> additionalOffsets = new ArrayList<Integer>(); 615 final ArrayList<Integer> additionalLengths = new ArrayList<Integer>(); 616 final ArrayList<SuggestionsInfo> additionalSuggestionsInfos = 617 new ArrayList<SuggestionsInfo>(); 618 String currentWord = null; 619 for (int i = 0; i < N; ++i) { 620 final SuggestionsInfo si = ssi.getSuggestionsInfoAt(i); 621 final int flags = si.getSuggestionsAttributes(); 622 if ((flags & SuggestionsInfo.RESULT_ATTR_IN_THE_DICTIONARY) == 0) { 623 continue; 624 } 625 final int offset = ssi.getOffsetAt(i); 626 final int length = ssi.getLengthAt(i); 627 final String subText = typedText.substring(offset, offset + length); 628 final String prevWord = currentWord; 629 currentWord = subText; 630 if (!subText.contains(SINGLE_QUOTE)) { 631 continue; 632 } 633 final String[] splitTexts = subText.split(SINGLE_QUOTE, -1); 634 if (splitTexts == null || splitTexts.length <= 1) { 635 continue; 636 } 637 final int splitNum = splitTexts.length; 638 for (int j = 0; j < splitNum; ++j) { 639 final String splitText = splitTexts[j]; 640 if (TextUtils.isEmpty(splitText)) { 641 continue; 642 } 643 if (mSuggestionsCache.getSuggestionsFromCache( 644 splitText, prevWord) == null) { 645 continue; 646 } 647 final int newLength = splitText.length(); 648 // Neither RESULT_ATTR_IN_THE_DICTIONARY nor RESULT_ATTR_LOOKS_LIKE_TYPO 649 final int newFlags = 0; 650 final SuggestionsInfo newSi = new SuggestionsInfo(newFlags, EMPTY_STRING_ARRAY); 651 newSi.setCookieAndSequence(si.getCookie(), si.getSequence()); 652 if (DBG) { 653 Log.d(TAG, "Override and remove old span over: " 654 + splitText + ", " + offset + "," + newLength); 655 } 656 additionalOffsets.add(offset); 657 additionalLengths.add(newLength); 658 additionalSuggestionsInfos.add(newSi); 659 } 660 } 661 final int additionalSize = additionalOffsets.size(); 662 if (additionalSize <= 0) { 663 return null; 664 } 665 final int suggestionsSize = N + additionalSize; 666 final int[] newOffsets = new int[suggestionsSize]; 667 final int[] newLengths = new int[suggestionsSize]; 668 final SuggestionsInfo[] newSuggestionsInfos = new SuggestionsInfo[suggestionsSize]; 669 int i; 670 for (i = 0; i < N; ++i) { 671 newOffsets[i] = ssi.getOffsetAt(i); 672 newLengths[i] = ssi.getLengthAt(i); 673 newSuggestionsInfos[i] = ssi.getSuggestionsInfoAt(i); 674 } 675 for (; i < suggestionsSize; ++i) { 676 newOffsets[i] = additionalOffsets.get(i - N); 677 newLengths[i] = additionalLengths.get(i - N); 678 newSuggestionsInfos[i] = additionalSuggestionsInfos.get(i - N); 679 } 680 return new SentenceSuggestionsInfo(newSuggestionsInfos, newOffsets, newLengths); 681 } 682 683 @Override 684 public SentenceSuggestionsInfo[] onGetSentenceSuggestionsMultiple( 685 TextInfo[] textInfos, int suggestionsLimit) { 686 final SentenceSuggestionsInfo[] retval = super.onGetSentenceSuggestionsMultiple( 687 textInfos, suggestionsLimit); 688 if (retval == null || retval.length != textInfos.length) { 689 return retval; 690 } 691 for (int i = 0; i < retval.length; ++i) { 692 final SentenceSuggestionsInfo tempSsi = 693 fixWronglyInvalidatedWordWithSingleQuote(textInfos[i], retval[i]); 694 if (tempSsi != null) { 695 retval[i] = tempSsi; 696 } 697 } 698 return retval; 699 } 700 701 @Override 702 public SuggestionsInfo[] onGetSuggestionsMultiple(TextInfo[] textInfos, 703 int suggestionsLimit, boolean sequentialWords) { 704 final int length = textInfos.length; 705 final SuggestionsInfo[] retval = new SuggestionsInfo[length]; 706 for (int i = 0; i < length; ++i) { 707 final String prevWord; 708 if (sequentialWords && i > 0) { 709 final String prevWordCandidate = textInfos[i - 1].getText(); 710 // Note that an empty string would be used to indicate the initial word 711 // in the future. 712 prevWord = TextUtils.isEmpty(prevWordCandidate) ? null : prevWordCandidate; 713 } else { 714 prevWord = null; 715 } 716 retval[i] = onGetSuggestions(textInfos[i], prevWord, suggestionsLimit); 717 retval[i].setCookieAndSequence( 718 textInfos[i].getCookie(), textInfos[i].getSequence()); 719 } 720 return retval; 721 } 722 723 // Note : this must be reentrant 724 /** 725 * Gets a list of suggestions for a specific string. This returns a list of possible 726 * corrections for the text passed as an argument. It may split or group words, and 727 * even perform grammatical analysis. 728 */ 729 @Override 730 public SuggestionsInfo onGetSuggestions(final TextInfo textInfo, 731 final int suggestionsLimit) { 732 return onGetSuggestions(textInfo, null, suggestionsLimit); 733 } 734 735 private SuggestionsInfo onGetSuggestions( 736 final TextInfo textInfo, final String prevWord, final int suggestionsLimit) { 737 try { 738 final String inText = textInfo.getText(); 739 final SuggestionsParams cachedSuggestionsParams = 740 mSuggestionsCache.getSuggestionsFromCache(inText, prevWord); 741 if (cachedSuggestionsParams != null) { 742 if (DBG) { 743 Log.d(TAG, "Cache hit: " + inText + ", " + cachedSuggestionsParams.mFlags); 744 } 745 return new SuggestionsInfo( 746 cachedSuggestionsParams.mFlags, cachedSuggestionsParams.mSuggestions); 747 } 748 749 if (shouldFilterOut(inText, mScript)) { 750 DictAndProximity dictInfo = null; 751 try { 752 dictInfo = mDictionaryPool.takeOrGetNull(); 753 if (null == dictInfo) return getNotInDictEmptySuggestions(); 754 return dictInfo.mDictionary.isValidWord(inText) ? 755 getInDictEmptySuggestions() : getNotInDictEmptySuggestions(); 756 } finally { 757 if (null != dictInfo) { 758 if (!mDictionaryPool.offer(dictInfo)) { 759 Log.e(TAG, "Can't re-insert a dictionary into its pool"); 760 } 761 } 762 } 763 } 764 final String text = inText.replaceAll(APOSTROPHE, SINGLE_QUOTE); 765 766 // TODO: Don't gather suggestions if the limit is <= 0 unless necessary 767 final SuggestionsGatherer suggestionsGatherer = new SuggestionsGatherer(text, 768 mService.mSuggestionThreshold, mService.mRecommendedThreshold, 769 suggestionsLimit); 770 final WordComposer composer = new WordComposer(); 771 final int length = text.length(); 772 for (int i = 0; i < length; i = text.offsetByCodePoints(i, 1)) { 773 final int codePoint = text.codePointAt(i); 774 // The getXYForCodePointAndScript method returns (Y << 16) + X 775 final int xy = SpellCheckerProximityInfo.getXYForCodePointAndScript( 776 codePoint, mScript); 777 if (SpellCheckerProximityInfo.NOT_A_COORDINATE_PAIR == xy) { 778 composer.add(codePoint, WordComposer.NOT_A_COORDINATE, 779 WordComposer.NOT_A_COORDINATE, null); 780 } else { 781 composer.add(codePoint, xy & 0xFFFF, xy >> 16, null); 782 } 783 } 784 785 final int capitalizeType = getCapitalizationType(text); 786 boolean isInDict = true; 787 DictAndProximity dictInfo = null; 788 try { 789 dictInfo = mDictionaryPool.takeOrGetNull(); 790 if (null == dictInfo) return getNotInDictEmptySuggestions(); 791 final ArrayList<SuggestedWordInfo> suggestions = dictInfo.mDictionary.getWords( 792 composer, prevWord, suggestionsGatherer, dictInfo.mProximityInfo); 793 for (final SuggestedWordInfo suggestion : suggestions) { 794 final String suggestionStr = suggestion.mWord.toString(); 795 suggestionsGatherer.oldAddWord(suggestionStr.toCharArray(), 0, 796 suggestionStr.length(), suggestion.mScore, 0 /* ignored */, 797 Dictionary.UNIGRAM); 798 } 799 isInDict = dictInfo.mDictionary.isValidWord(text); 800 if (!isInDict && CAPITALIZE_NONE != capitalizeType) { 801 // We want to test the word again if it's all caps or first caps only. 802 // If it's fully down, we already tested it, if it's mixed case, we don't 803 // want to test a lowercase version of it. 804 isInDict = dictInfo.mDictionary.isValidWord(text.toLowerCase(mLocale)); 805 } 806 } finally { 807 if (null != dictInfo) { 808 if (!mDictionaryPool.offer(dictInfo)) { 809 Log.e(TAG, "Can't re-insert a dictionary into its pool"); 810 } 811 } 812 } 813 814 final SuggestionsGatherer.Result result = suggestionsGatherer.getResults( 815 capitalizeType, mLocale); 816 817 if (DBG) { 818 Log.i(TAG, "Spell checking results for " + text + " with suggestion limit " 819 + suggestionsLimit); 820 Log.i(TAG, "IsInDict = " + isInDict); 821 Log.i(TAG, "LooksLikeTypo = " + (!isInDict)); 822 Log.i(TAG, "HasRecommendedSuggestions = " + result.mHasRecommendedSuggestions); 823 if (null != result.mSuggestions) { 824 for (String suggestion : result.mSuggestions) { 825 Log.i(TAG, suggestion); 826 } 827 } 828 } 829 830 final int flags = 831 (isInDict ? SuggestionsInfo.RESULT_ATTR_IN_THE_DICTIONARY 832 : SuggestionsInfo.RESULT_ATTR_LOOKS_LIKE_TYPO) 833 | (result.mHasRecommendedSuggestions 834 ? SuggestionsInfoCompatUtils 835 .getValueOf_RESULT_ATTR_HAS_RECOMMENDED_SUGGESTIONS() 836 : 0); 837 final SuggestionsInfo retval = new SuggestionsInfo(flags, result.mSuggestions); 838 mSuggestionsCache.putSuggestionsToCache(text, prevWord, result.mSuggestions, flags); 839 return retval; 840 } catch (RuntimeException e) { 841 // Don't kill the keyboard if there is a bug in the spell checker 842 if (DBG) { 843 throw e; 844 } else { 845 Log.e(TAG, "Exception while spellcheking: " + e); 846 return getNotInDictEmptySuggestions(); 847 } 848 } 849 } 850 } 851} 852