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