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