src/com/hughes/android/dictionary/engine/Index.java

   1 // Copyright 2011 Google Inc. All Rights Reserved.
   2 //
   3 // Licensed under the Apache License, Version 2.0 (the "License");
   4 // you may not use this file except in compliance with the License.
   5 // You may obtain a copy of the License at
   6 //
   7 //     http://www.apache.org/licenses/LICENSE-2.0
   8 //
   9 // Unless required by applicable law or agreed to in writing, software
  10 // distributed under the License is distributed on an "AS IS" BASIS,
  11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  12 // See the License for the specific language governing permissions and
  13 // limitations under the License.
  14
  15 package com.hughes.android.dictionary.engine;
  16
  17 import java.io.DataInput;
  18 import java.io.DataOutput;
  19 import java.io.IOException;
  20 import java.io.PrintStream;
  21 import java.io.RandomAccessFile;
  22 import java.util.AbstractList;
  23 import java.util.ArrayList;
  24 import java.util.Collection;
  25 import java.util.Collections;
  26 import java.util.Comparator;
  27 import java.util.EnumMap;
  28 import java.util.HashMap;
  29 import java.util.HashSet;
  30 import java.util.List;
  31 import java.util.Map;
  32 import java.util.Set;
  33 import java.util.concurrent.atomic.AtomicBoolean;
  34 import java.util.regex.Pattern;
  35
  36 import com.hughes.android.dictionary.DictionaryInfo;
  37 import com.hughes.android.dictionary.DictionaryInfo.IndexInfo;
  38 import com.hughes.android.dictionary.engine.RowBase.RowKey;
  39 import com.hughes.util.CachingList;
  40 import com.hughes.util.DataInputBuffer;
  41 import com.hughes.util.StringUtil;
  42 import com.hughes.util.TransformingList;
  43 import com.hughes.util.raf.RAFList;
  44 import com.hughes.util.raf.RAFSerializable;
  45 import com.hughes.util.raf.RAFSerializer;
  46 import com.hughes.util.raf.SerializableSerializer;
  47 import com.hughes.util.raf.UniformRAFList;
  48 import com.ibm.icu.text.Transliterator;
  49
  50 public final class Index implements RAFSerializable<Index> {
  51
  52     private static final int CACHE_SIZE = 5000;
  53
  54     public final Dictionary dict;
  55
  56     public final String shortName; // Typically the ISO code for the language.
  57     public final String longName;
  58
  59     // persisted: tells how the entries are sorted.
  60     public final Language sortLanguage;
  61     public final String normalizerRules;
  62
  63     // Built from the two above.
  64     private Transliterator normalizer;
  65
  66     // persisted
  67     public final List<IndexEntry> sortedIndexEntries;
  68
  69     // persisted.
  70     public final Set<String> stoplist;
  71
  72     // One big list!
  73     // Various sub-types.
  74     // persisted
  75     public final List<RowBase> rows;
  76     public final boolean swapPairEntries;
  77
  78     // Version 2:
  79     @SuppressWarnings("WeakerAccess")
  80     public int mainTokenCount = -1;
  81
  82     // --------------------------------------------------------------------------
  83
  84     public Index(final Dictionary dict, final String shortName, final String longName,
  85                  final Language sortLanguage, final String normalizerRules,
  86                  final boolean swapPairEntries, final Set<String> stoplist) {
  87         this.dict = dict;
  88         this.shortName = shortName;
  89         this.longName = longName;
  90         this.sortLanguage = sortLanguage;
  91         this.normalizerRules = normalizerRules;
  92         this.swapPairEntries = swapPairEntries;
  93         sortedIndexEntries = new ArrayList<>();
  94         this.stoplist = stoplist;
  95         rows = new ArrayList<>();
  96
  97         normalizer = null;
  98     }
  99
 100     /**
 101      * Deferred initialization because it can be slow.
 102      */
 103     @SuppressWarnings("WeakerAccess")
 104     public synchronized Transliterator normalizer() {
 105         if (normalizer == null) {
 106             normalizer = TransliteratorManager.get(normalizerRules);
 107         }
 108         return normalizer;
 109     }
 110
 111     /**
 112      * Note that using this comparator probably involves doing too many text
 113      * normalizations.
 114      */
 115     @SuppressWarnings("WeakerAccess")
 116     public NormalizeComparator getSortComparator() {
 117         return new NormalizeComparator(normalizer(), sortLanguage.getCollator(), dict.dictFileVersion);
 118     }
 119
 120     public Index(final Dictionary dict, final DataInputBuffer raf) throws IOException {
 121         this.dict = dict;
 122         shortName = raf.readUTF();
 123         longName = raf.readUTF();
 124         final String languageCode = raf.readUTF();
 125         sortLanguage = Language.lookup(languageCode);
 126         normalizerRules = raf.readUTF();
 127         swapPairEntries = raf.readBoolean();
 128         if (dict.dictFileVersion >= 2) {
 129             mainTokenCount = raf.readInt();
 130         }
 131         sortedIndexEntries = CachingList.create(
 132                                  RAFList.create(raf, new IndexEntrySerializer(),
 133                                                 dict.dictFileVersion, dict.dictInfo + " idx " + languageCode + ": "), CACHE_SIZE, true);
 134         if (dict.dictFileVersion >= 7) {
 135             int count = StringUtil.readVarInt(raf);
 136             stoplist = new HashSet<>(count);
 137             for (int i = 0; i < count; ++i) {
 138                 stoplist.add(raf.readUTF());
 139             }
 140         } else if (dict.dictFileVersion >= 4) {
 141             stoplist = new SerializableSerializer<Set<String>>().read(raf);
 142         } else {
 143             stoplist = Collections.emptySet();
 144         }
 145         rows = CachingList.create(
 146                    UniformRAFList.create(raf, new RowBase.Serializer(this)),
 147                    CACHE_SIZE, true);
 148     }
 149
 150     @Override
 151     public void write(final DataOutput out) throws IOException {
 152         RandomAccessFile raf = (RandomAccessFile)out;
 153         raf.writeUTF(shortName);
 154         raf.writeUTF(longName);
 155         raf.writeUTF(sortLanguage.getIsoCode());
 156         raf.writeUTF(normalizerRules);
 157         raf.writeBoolean(swapPairEntries);
 158         raf.writeInt(mainTokenCount);
 159         RAFList.write(raf, sortedIndexEntries, new IndexEntrySerializer(), 32, true);
 160         StringUtil.writeVarInt(raf, stoplist.size());
 161         for (String i : stoplist) {
 162             raf.writeUTF(i);
 163         }
 164         UniformRAFList.write(raf, rows, new RowBase.Serializer(this), 3 /* bytes per entry */);
 165     }
 166
 167     public void print(final PrintStream out) {
 168         for (final RowBase row : rows) {
 169             row.print(out);
 170         }
 171     }
 172
 173     private final class IndexEntrySerializer implements RAFSerializer<IndexEntry> {
 174         @Override
 175         public IndexEntry read(DataInput raf) throws IOException {
 176             return new IndexEntry(Index.this, raf);
 177         }
 178
 179         @Override
 180         public void write(DataOutput raf, IndexEntry t) throws IOException {
 181             t.write(raf);
 182         }
 183     }
 184
 185     public static final class IndexEntry implements RAFSerializable<Index.IndexEntry> {
 186         public final String token;
 187         private final String normalizedToken;
 188         public final int startRow;
 189         final int numRows; // doesn't count the token row!
 190         public List<HtmlEntry> htmlEntries;
 191
 192         public IndexEntry(final Index index, final String token, final String normalizedToken,
 193                           final int startRow, final int numRows) {
 194             assert token.equals(token.trim());
 195             assert token.length() > 0;
 196             this.token = token;
 197             this.normalizedToken = normalizedToken;
 198             this.startRow = startRow;
 199             this.numRows = numRows;
 200             this.htmlEntries = new ArrayList<>();
 201         }
 202
 203         IndexEntry(final Index index, final DataInput raf) throws IOException {
 204             token = raf.readUTF();
 205             if (index.dict.dictFileVersion >= 7) {
 206                 startRow = StringUtil.readVarInt(raf);
 207                 numRows = StringUtil.readVarInt(raf);
 208             } else {
 209                 startRow = raf.readInt();
 210                 numRows = raf.readInt();
 211             }
 212             final boolean hasNormalizedForm = raf.readBoolean();
 213             normalizedToken = hasNormalizedForm ? raf.readUTF() : token;
 214             if (index.dict.dictFileVersion >= 7) {
 215                 int size = StringUtil.readVarInt(raf);
 216                 if (size == 0) {
 217                     this.htmlEntries = Collections.emptyList();
 218                 } else {
 219                     final int[] htmlEntryIndices = new int[size];
 220                     for (int i = 0; i < size; ++i) {
 221                         htmlEntryIndices[i] = StringUtil.readVarInt(raf);
 222                     }
 223                     this.htmlEntries = new AbstractList<HtmlEntry>() {
 224                         @Override
 225                         public HtmlEntry get(int i) {
 226                             return index.dict.htmlEntries.get(htmlEntryIndices[i]);
 227                         }
 228
 229                         @Override
 230                         public int size() {
 231                             return htmlEntryIndices.length;
 232                         }
 233                     };
 234                 }
 235             } else if (index.dict.dictFileVersion >= 6) {
 236                 this.htmlEntries = CachingList.create(
 237                                        RAFList.create((DataInputBuffer)raf, index.dict.htmlEntryIndexSerializer,
 238                                                       index.dict.dictFileVersion,
 239                                                       index.dict.dictInfo + " htmlEntries: "), 1, false);
 240             } else {
 241                 this.htmlEntries = Collections.emptyList();
 242             }
 243         }
 244
 245         public void write(DataOutput raf) throws IOException {
 246             raf.writeUTF(token);
 247             StringUtil.writeVarInt(raf, startRow);
 248             StringUtil.writeVarInt(raf, numRows);
 249             final boolean hasNormalizedForm = !token.equals(normalizedToken);
 250             raf.writeBoolean(hasNormalizedForm);
 251             if (hasNormalizedForm) {
 252                 raf.writeUTF(normalizedToken);
 253             }
 254             StringUtil.writeVarInt(raf, htmlEntries.size());
 255             for (HtmlEntry e : htmlEntries)
 256                 StringUtil.writeVarInt(raf, e.index());
 257         }
 258
 259         public String toString() {
 260             return String.format("%s@%d(%d)", token, startRow, numRows);
 261         }
 262
 263         String normalizedToken() {
 264             return normalizedToken;
 265         }
 266     }
 267
 268     private static final TransformingList.Transformer<IndexEntry, String> INDEX_ENTRY_TO_TOKEN = new TransformingList.Transformer<IndexEntry, String>() {
 269         @Override
 270         public String transform(IndexEntry t1) {
 271             return t1.token;
 272         }
 273     };
 274
 275     public IndexEntry findExact(final String exactToken) {
 276         final int result = Collections.binarySearch(
 277                                TransformingList.create(sortedIndexEntries, INDEX_ENTRY_TO_TOKEN), exactToken,
 278                                getSortComparator());
 279         if (result >= 0) {
 280             return sortedIndexEntries.get(result);
 281         }
 282         return null;
 283     }
 284
 285     public IndexEntry findInsertionPoint(String token, final AtomicBoolean interrupted) {
 286         final int index = findInsertionPointIndex(token, interrupted);
 287         return index != -1 ? sortedIndexEntries.get(index) : null;
 288     }
 289
 290     private int compareIdx(String token, final Comparator<Object> sortCollator, int idx) {
 291         final IndexEntry entry = sortedIndexEntries.get(idx);
 292         return NormalizeComparator.compareWithoutDash(token, entry.normalizedToken(), sortCollator, dict.dictFileVersion);
 293     }
 294
 295     private int findMatchLen(final Comparator<Object> sortCollator, String a, String b) {
 296         int start = 0;
 297         int end = Math.min(a.length(), b.length());
 298         while (start < end)
 299         {
 300             int mid = (start + end + 1) / 2;
 301             if (sortCollator.compare(a.substring(0, mid), b.substring(0, mid)) == 0)
 302                 start = mid;
 303             else
 304                 end = mid - 1;
 305         }
 306         return start;
 307     }
 308
 309     private int findInsertionPointIndex(String token, final AtomicBoolean interrupted) {
 310         token = normalizeToken(token);
 311
 312         int start = 0;
 313         int end = sortedIndexEntries.size();
 314
 315         final Comparator<Object> sortCollator = sortLanguage.getCollator();
 316         while (start < end) {
 317             final int mid = (start + end) / 2;
 318             if (interrupted.get()) {
 319                 return -1;
 320             }
 321             final IndexEntry midEntry = sortedIndexEntries.get(mid);
 322
 323             int comp = NormalizeComparator.compareWithoutDash(token, midEntry.normalizedToken(), sortCollator, dict.dictFileVersion);
 324             if (comp == 0)
 325                 comp = sortCollator.compare(token, midEntry.normalizedToken());
 326             if (comp == 0) {
 327                 return windBackCase(token, mid, interrupted);
 328             } else if (comp < 0) {
 329                 // System.out.println("Upper bound: " + midEntry + ", norm=" +
 330                 // midEntry.normalizedToken() + ", mid=" + mid);
 331
 332                 // Hack for robustness if sort order is broken
 333                 if (mid + 2 < end &&
 334                     compareIdx(token, sortCollator, mid + 1) > 0 &&
 335                     compareIdx(token, sortCollator, mid + 2) > 0) {
 336                     start = mid;
 337                 } else {
 338                     end = mid;
 339                 }
 340             } else {
 341                 // System.out.println("Lower bound: " + midEntry + ", norm=" +
 342                 // midEntry.normalizedToken() + ", mid=" + mid);
 343
 344                 // Hack for robustness if sort order is broken
 345                 if (mid - 2 >= start &&
 346                     compareIdx(token, sortCollator, mid - 1) < 0 &&
 347                     compareIdx(token, sortCollator, mid - 2) < 0) {
 348                     end = mid + 1;
 349                 } else {
 350                     start = mid + 1;
 351                 }
 352             }
 353         }
 354
 355         // if the word before is the better match, move
 356         // our result to it
 357         if (start > 0 && start < sortedIndexEntries.size()) {
 358             String prev = sortedIndexEntries.get(start - 1).normalizedToken();
 359             String next = sortedIndexEntries.get(start).normalizedToken();
 360             if (findMatchLen(sortCollator, token, prev) >= findMatchLen(sortCollator, token, next))
 361                 start--;
 362         }
 363
 364         // If we search for a substring of a string that's in there, return
 365         // that.
 366         int result = Math.min(start, sortedIndexEntries.size() - 1);
 367         result = windBackCase(sortedIndexEntries.get(result).normalizedToken(), result, interrupted);
 368         return result;
 369     }
 370
 371     private int windBackCase(final String token, int result, final AtomicBoolean interrupted) {
 372         while (result > 0 && sortedIndexEntries.get(result - 1).normalizedToken().equals(token)) {
 373             --result;
 374             if (interrupted.get()) {
 375                 return result;
 376             }
 377         }
 378         return result;
 379     }
 380
 381     public IndexInfo getIndexInfo() {
 382         return new DictionaryInfo.IndexInfo(shortName, sortedIndexEntries.size(), mainTokenCount);
 383     }
 384
 385     private static final int MAX_SEARCH_ROWS = 1000;
 386
 387     private final Map<String, Integer> prefixToNumRows = new HashMap<>();
 388
 389     private synchronized int getUpperBoundOnRowsStartingWith(final String normalizedPrefix,
 390                                                              final int maxRows, final AtomicBoolean interrupted) {
 391         final Integer numRows = prefixToNumRows.get(normalizedPrefix);
 392         if (numRows != null) {
 393             return numRows;
 394         }
 395         final int insertionPointIndex = findInsertionPointIndex(normalizedPrefix, interrupted);
 396
 397         int rowCount = 0;
 398         for (int index = insertionPointIndex; index < sortedIndexEntries.size(); ++index) {
 399             if (interrupted.get()) {
 400                 return -1;
 401             }
 402             final IndexEntry indexEntry = sortedIndexEntries.get(index);
 403             if (!indexEntry.normalizedToken.startsWith(normalizedPrefix) &&
 404                 !NormalizeComparator.withoutDash(indexEntry.normalizedToken).startsWith(normalizedPrefix)) {
 405                 break;
 406             }
 407             rowCount += indexEntry.numRows + indexEntry.htmlEntries.size();
 408             if (rowCount > maxRows) {
 409                 System.out.println("Giving up, too many words with prefix: " + normalizedPrefix);
 410                 break;
 411             }
 412         }
 413         prefixToNumRows.put(normalizedPrefix, rowCount);
 414         return rowCount;
 415     }
 416
 417     public final List<RowBase> multiWordSearch(
 418         final String searchText, final List<String> searchTokens,
 419         final AtomicBoolean interrupted) {
 420         final long startMills = System.currentTimeMillis();
 421         final List<RowBase> result = new ArrayList<>();
 422
 423         final Set<String> normalizedNonStoplist = new HashSet<>();
 424
 425         String bestPrefix = null;
 426         int leastRows = Integer.MAX_VALUE;
 427         final StringBuilder searchTokensRegex = new StringBuilder();
 428         for (int i = 0; i < searchTokens.size(); ++i) {
 429             if (interrupted.get()) {
 430                 return null;
 431             }
 432             final String searchToken = searchTokens.get(i);
 433             final String normalized = normalizeToken(searchTokens.get(i));
 434             // Normalize them all.
 435             searchTokens.set(i, normalized);
 436
 437             if (!stoplist.contains(searchToken)) {
 438                 if (normalizedNonStoplist.add(normalized)) {
 439                     final int numRows = getUpperBoundOnRowsStartingWith(normalized,
 440                                         MAX_SEARCH_ROWS, interrupted);
 441                     if (numRows != -1 && numRows < leastRows) {
 442                         if (numRows == 0) {
 443                             // We really are done here.
 444                             return Collections.emptyList();
 445                         }
 446                         leastRows = numRows;
 447                         bestPrefix = normalized;
 448                     }
 449                 }
 450             }
 451
 452             if (searchTokensRegex.length() > 0) {
 453                 searchTokensRegex.append("[\\s]*");
 454             }
 455             searchTokensRegex.append(Pattern.quote(normalized));
 456         }
 457         final Pattern pattern = Pattern.compile(searchTokensRegex.toString());
 458
 459         if (bestPrefix == null) {
 460             bestPrefix = searchTokens.get(0);
 461             System.out.println("Everything was in the stoplist!");
 462         }
 463         System.out.println("Searching using prefix: " + bestPrefix + ", leastRows=" + leastRows
 464                            + ", searchTokens=" + searchTokens);
 465
 466         // Place to store the things that match.
 467         final Map<RowMatchType, List<RowBase>> matches = new EnumMap<>(
 468                 RowMatchType.class);
 469         for (final RowMatchType rowMatchType : RowMatchType.values()) {
 470             if (rowMatchType != RowMatchType.NO_MATCH) {
 471                 matches.put(rowMatchType, new ArrayList<RowBase>());
 472             }
 473         }
 474
 475         int matchCount = 0;
 476
 477         final int exactMatchIndex = findInsertionPointIndex(searchText, interrupted);
 478         if (exactMatchIndex != -1) {
 479             final IndexEntry exactMatch = sortedIndexEntries.get(exactMatchIndex);
 480             if (pattern.matcher(exactMatch.token).find()) {
 481                 matches.get(RowMatchType.TITLE_MATCH).add(rows.get(exactMatch.startRow));
 482             }
 483         }
 484
 485         final String searchToken = bestPrefix;
 486         final int insertionPointIndex = findInsertionPointIndex(searchToken, interrupted);
 487         final Set<RowKey> rowsAlreadySeen = new HashSet<>();
 488         for (int index = insertionPointIndex; index < sortedIndexEntries.size()
 489                 && matchCount < MAX_SEARCH_ROWS; ++index) {
 490             if (interrupted.get()) {
 491                 return null;
 492             }
 493             final IndexEntry indexEntry = sortedIndexEntries.get(index);
 494             if (!indexEntry.normalizedToken.startsWith(searchToken) &&
 495                 !NormalizeComparator.withoutDash(indexEntry.normalizedToken).startsWith(searchToken)) {
 496                 break;
 497             }
 498
 499             // System.out.println("Searching indexEntry: " + indexEntry.token);
 500
 501             // Extra +1 to skip token row.
 502             for (int rowIndex = indexEntry.startRow + 1; rowIndex < indexEntry.startRow + 1
 503                     + indexEntry.numRows
 504                     && rowIndex < rows.size(); ++rowIndex) {
 505                 if (interrupted.get()) {
 506                     return null;
 507                 }
 508                 final RowBase row = rows.get(rowIndex);
 509                 final RowBase.RowKey rowKey = row.getRowKey();
 510                 if (rowsAlreadySeen.contains(rowKey)) {
 511                     continue;
 512                 }
 513                 rowsAlreadySeen.add(rowKey);
 514                 final RowMatchType matchType = row.matches(searchTokens, pattern, normalizer(),
 515                                                swapPairEntries);
 516                 if (matchType != RowMatchType.NO_MATCH) {
 517                     matches.get(matchType).add(row);
 518                     ++matchCount;
 519                 }
 520             }
 521         }
 522         // } // searchTokens
 523
 524         // Sort them into a reasonable order.
 525         final RowBase.LengthComparator lengthComparator = new RowBase.LengthComparator(
 526             swapPairEntries);
 527         for (final Collection<RowBase> rows : matches.values()) {
 528             final List<RowBase> ordered = new ArrayList<>(rows);
 529             ordered.sort(lengthComparator);
 530             result.addAll(ordered);
 531         }
 532
 533         System.out.println("searchDuration: " + (System.currentTimeMillis() - startMills));
 534         return result;
 535     }
 536
 537     private String normalizeToken(final String searchToken) {
 538         if (TransliteratorManager.init(null, null)) {
 539             final Transliterator normalizer = normalizer();
 540             return normalizer.transliterate(searchToken);
 541         } else {
 542             // Do our best since the Transliterators aren't up yet.
 543             return searchToken.toLowerCase();
 544         }
 545     }
 546
 547 }