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), 5 /*
 164                                                                                                * bytes
 165                                                                                                * per
 166                                                                                                * entry
 167                                                                                                */);
 168     }
 169
 170     public void print(final PrintStream out) {
 171         for (final RowBase row : rows) {
 172             row.print(out);
 173         }
 174     }
 175
 176     private final RAFSerializer<IndexEntry> indexEntrySerializer = new RAFSerializer<IndexEntry>() {
 177         @Override
 178         public IndexEntry read(DataInput raf) throws IOException {
 179             return new IndexEntry(Index.this, raf);
 180         }
 181
 182         @Override
 183         public void write(DataOutput raf, IndexEntry t) throws IOException {
 184             t.write(raf);
 185         }
 186     };
 187
 188     public static final class IndexEntry implements RAFSerializable<Index.IndexEntry> {
 189         private final Index index;
 190         public final String token;
 191         private final String normalizedToken;
 192         public final int startRow;
 193         public final int numRows; // doesn't count the token row!
 194         public List<HtmlEntry> htmlEntries;
 195         private int[] htmlEntryIndices;
 196
 197         public IndexEntry(final Index index, final String token, final String normalizedToken,
 198                 final int startRow, final int numRows) {
 199             this.index = index;
 200             assert token.equals(token.trim());
 201             assert token.length() > 0;
 202             this.token = token;
 203             this.normalizedToken = normalizedToken;
 204             this.startRow = startRow;
 205             this.numRows = numRows;
 206             this.htmlEntries = new ArrayList<HtmlEntry>();
 207         }
 208
 209         public IndexEntry(final Index index, final DataInput raf) throws IOException {
 210             this.index = index;
 211             token = raf.readUTF();
 212             if (index.dict.dictFileVersion >= 7) {
 213                 startRow = StringUtil.readVarInt(raf);
 214                 numRows = StringUtil.readVarInt(raf);
 215             } else {
 216                 startRow = raf.readInt();
 217                 numRows = raf.readInt();
 218             }
 219             final boolean hasNormalizedForm = raf.readBoolean();
 220             normalizedToken = hasNormalizedForm ? raf.readUTF() : token;
 221             htmlEntryIndices = null;
 222             if (index.dict.dictFileVersion >= 7) {
 223                 int size = StringUtil.readVarInt(raf);
 224                 htmlEntryIndices = new int[size];
 225                 for (int i = 0; i < size; ++i) {
 226                     htmlEntryIndices[i] = StringUtil.readVarInt(raf);
 227                 }
 228                 this.htmlEntries = CachingList.create(new AbstractList<HtmlEntry>() {
 229                     @Override
 230                     public HtmlEntry get(int i) {
 231                         return index.dict.htmlEntries.get(htmlEntryIndices[i]);
 232                     }
 233                     @Override
 234                     public int size() {
 235                         return htmlEntryIndices.length;
 236                     }
 237                     }, 1);
 238             } else if (index.dict.dictFileVersion >= 6) {
 239                 this.htmlEntries = CachingList.create(
 240                         RAFList.create((RandomAccessFile)raf, index.dict.htmlEntryIndexSerializer,
 241                                 ((RandomAccessFile)raf).getFilePointer(), index.dict.dictFileVersion), 1);
 242             } else {
 243                 this.htmlEntries = Collections.emptyList();
 244             }
 245         }
 246
 247         public void write(DataOutput raf) throws IOException {
 248             raf.writeUTF(token);
 249             StringUtil.writeVarInt(raf, startRow);
 250             StringUtil.writeVarInt(raf, numRows);
 251             final boolean hasNormalizedForm = !token.equals(normalizedToken);
 252             raf.writeBoolean(hasNormalizedForm);
 253             if (hasNormalizedForm) {
 254                 raf.writeUTF(normalizedToken);
 255             }
 256             StringUtil.writeVarInt(raf, htmlEntries.size());
 257             for (HtmlEntry e : htmlEntries)
 258                 StringUtil.writeVarInt(raf, e.index());
 259         }
 260
 261         public String toString() {
 262             return String.format("%s@%d(%d)", token, startRow, numRows);
 263         }
 264
 265         public String normalizedToken() {
 266             return normalizedToken;
 267         }
 268     }
 269
 270     static final TransformingList.Transformer<IndexEntry, String> INDEX_ENTRY_TO_TOKEN = new TransformingList.Transformer<IndexEntry, String>() {
 271         @Override
 272         public String transform(IndexEntry t1) {
 273             return t1.token;
 274         }
 275     };
 276
 277     public IndexEntry findExact(final String exactToken) {
 278         final int result = Collections.binarySearch(
 279                 TransformingList.create(sortedIndexEntries, INDEX_ENTRY_TO_TOKEN), exactToken,
 280                 getSortComparator());
 281         if (result >= 0) {
 282             return sortedIndexEntries.get(result);
 283         }
 284         return null;
 285     }
 286
 287     public IndexEntry findInsertionPoint(String token, final AtomicBoolean interrupted) {
 288         final int index = findInsertionPointIndex(token, interrupted);
 289         return index != -1 ? sortedIndexEntries.get(index) : null;
 290     }
 291
 292     public int findInsertionPointIndex(String token, final AtomicBoolean interrupted) {
 293         token = normalizeToken(token);
 294
 295         int start = 0;
 296         int end = sortedIndexEntries.size();
 297
 298         final Collator sortCollator = sortLanguage.getCollator();
 299         while (start < end) {
 300             final int mid = (start + end) / 2;
 301             if (interrupted.get()) {
 302                 return -1;
 303             }
 304             final IndexEntry midEntry = sortedIndexEntries.get(mid);
 305
 306             final int comp = sortCollator.compare(token, midEntry.normalizedToken());
 307             if (comp == 0) {
 308                 final int result = windBackCase(token, mid, interrupted);
 309                 return result;
 310             } else if (comp < 0) {
 311                 // System.out.println("Upper bound: " + midEntry + ", norm=" +
 312                 // midEntry.normalizedToken() + ", mid=" + mid);
 313                 end = mid;
 314             } else {
 315                 // System.out.println("Lower bound: " + midEntry + ", norm=" +
 316                 // midEntry.normalizedToken() + ", mid=" + mid);
 317                 start = mid + 1;
 318             }
 319         }
 320
 321         // If we search for a substring of a string that's in there, return
 322         // that.
 323         int result = Math.min(start, sortedIndexEntries.size() - 1);
 324         result = windBackCase(sortedIndexEntries.get(result).normalizedToken(), result, interrupted);
 325         return result;
 326     }
 327
 328     private final int windBackCase(final String token, int result, final AtomicBoolean interrupted) {
 329         while (result > 0 && sortedIndexEntries.get(result - 1).normalizedToken().equals(token)) {
 330             --result;
 331             if (interrupted.get()) {
 332                 return result;
 333             }
 334         }
 335         return result;
 336     }
 337
 338     public IndexInfo getIndexInfo() {
 339         return new DictionaryInfo.IndexInfo(shortName, sortedIndexEntries.size(), mainTokenCount);
 340     }
 341
 342     private static final int MAX_SEARCH_ROWS = 1000;
 343
 344     private final Map<String, Integer> prefixToNumRows = new LinkedHashMap<String, Integer>();
 345
 346     private synchronized final int getUpperBoundOnRowsStartingWith(final String normalizedPrefix,
 347             final int maxRows, final AtomicBoolean interrupted) {
 348         final Integer numRows = prefixToNumRows.get(normalizedPrefix);
 349         if (numRows != null) {
 350             return numRows;
 351         }
 352         final int insertionPointIndex = findInsertionPointIndex(normalizedPrefix, interrupted);
 353
 354         int rowCount = 0;
 355         for (int index = insertionPointIndex; index < sortedIndexEntries.size(); ++index) {
 356             if (interrupted.get()) {
 357                 return -1;
 358             }
 359             final IndexEntry indexEntry = sortedIndexEntries.get(index);
 360             if (!indexEntry.normalizedToken.startsWith(normalizedPrefix)) {
 361                 break;
 362             }
 363             rowCount += indexEntry.numRows + indexEntry.htmlEntries.size();
 364             if (rowCount > maxRows) {
 365                 System.out.println("Giving up, too many words with prefix: " + normalizedPrefix);
 366                 break;
 367             }
 368         }
 369         prefixToNumRows.put(normalizedPrefix, numRows);
 370         return rowCount;
 371     }
 372
 373     public final List<RowBase> multiWordSearch(
 374             final String searchText, final List<String> searchTokens,
 375             final AtomicBoolean interrupted) {
 376         final long startMills = System.currentTimeMillis();
 377         final List<RowBase> result = new ArrayList<RowBase>();
 378
 379         final Set<String> normalizedNonStoplist = new LinkedHashSet<String>();
 380
 381         String bestPrefix = null;
 382         int leastRows = Integer.MAX_VALUE;
 383         final StringBuilder searchTokensRegex = new StringBuilder();
 384         for (int i = 0; i < searchTokens.size(); ++i) {
 385             if (interrupted.get()) {
 386                 return null;
 387             }
 388             final String searchToken = searchTokens.get(i);
 389             final String normalized = normalizeToken(searchTokens.get(i));
 390             // Normalize them all.
 391             searchTokens.set(i, normalized);
 392
 393             if (!stoplist.contains(searchToken)) {
 394                 if (normalizedNonStoplist.add(normalized)) {
 395                     final int numRows = getUpperBoundOnRowsStartingWith(normalized,
 396                             MAX_SEARCH_ROWS, interrupted);
 397                     if (numRows != -1 && numRows < leastRows) {
 398                         if (numRows == 0) {
 399                             // We really are done here.
 400                             return Collections.emptyList();
 401                         }
 402                         leastRows = numRows;
 403                         bestPrefix = normalized;
 404                     }
 405                 }
 406             }
 407
 408             if (searchTokensRegex.length() > 0) {
 409                 searchTokensRegex.append("[\\s]*");
 410             }
 411             searchTokensRegex.append(Pattern.quote(normalized));
 412         }
 413         final Pattern pattern = Pattern.compile(searchTokensRegex.toString());
 414
 415         if (bestPrefix == null) {
 416             bestPrefix = searchTokens.get(0);
 417             System.out.println("Everything was in the stoplist!");
 418         }
 419         System.out.println("Searching using prefix: " + bestPrefix + ", leastRows=" + leastRows
 420                 + ", searchTokens=" + searchTokens);
 421
 422         // Place to store the things that match.
 423         final Map<RowMatchType, List<RowBase>> matches = new EnumMap<RowMatchType, List<RowBase>>(
 424                 RowMatchType.class);
 425         for (final RowMatchType rowMatchType : RowMatchType.values()) {
 426             if (rowMatchType != RowMatchType.NO_MATCH) {
 427                 matches.put(rowMatchType, new ArrayList<RowBase>());
 428             }
 429         }
 430
 431         int matchCount = 0;
 432
 433         final int exactMatchIndex = findInsertionPointIndex(searchText, interrupted);
 434         if (exactMatchIndex != -1) {
 435             final IndexEntry exactMatch = sortedIndexEntries.get(exactMatchIndex);
 436             if (pattern.matcher(exactMatch.token).find()) {
 437                 matches.get(RowMatchType.TITLE_MATCH).add(rows.get(exactMatch.startRow));
 438             }
 439         }
 440
 441         final String searchToken = bestPrefix;
 442         final int insertionPointIndex = findInsertionPointIndex(searchToken, interrupted);
 443         final Set<RowKey> rowsAlreadySeen = new HashSet<RowBase.RowKey>();
 444         for (int index = insertionPointIndex; index < sortedIndexEntries.size()
 445                 && matchCount < MAX_SEARCH_ROWS; ++index) {
 446             if (interrupted.get()) {
 447                 return null;
 448             }
 449             final IndexEntry indexEntry = sortedIndexEntries.get(index);
 450             if (!indexEntry.normalizedToken.startsWith(searchToken)) {
 451                 break;
 452             }
 453
 454             // System.out.println("Searching indexEntry: " + indexEntry.token);
 455
 456             // Extra +1 to skip token row.
 457             for (int rowIndex = indexEntry.startRow + 1; rowIndex < indexEntry.startRow + 1
 458                     + indexEntry.numRows
 459                     && rowIndex < rows.size(); ++rowIndex) {
 460                 if (interrupted.get()) {
 461                     return null;
 462                 }
 463                 final RowBase row = rows.get(rowIndex);
 464                 final RowBase.RowKey rowKey = row.getRowKey();
 465                 if (rowsAlreadySeen.contains(rowKey)) {
 466                     continue;
 467                 }
 468                 rowsAlreadySeen.add(rowKey);
 469                 final RowMatchType matchType = row.matches(searchTokens, pattern, normalizer(),
 470                         swapPairEntries);
 471                 if (matchType != RowMatchType.NO_MATCH) {
 472                     matches.get(matchType).add(row);
 473                     ++matchCount;
 474                 }
 475             }
 476         }
 477         // } // searchTokens
 478
 479         // Sort them into a reasonable order.
 480         final RowBase.LengthComparator lengthComparator = new RowBase.LengthComparator(
 481                 swapPairEntries);
 482         for (final Collection<RowBase> rows : matches.values()) {
 483             final List<RowBase> ordered = new ArrayList<RowBase>(rows);
 484             Collections.sort(ordered, lengthComparator);
 485             result.addAll(ordered);
 486         }
 487
 488         System.out.println("searchDuration: " + (System.currentTimeMillis() - startMills));
 489         return result;
 490     }
 491
 492     private String normalizeToken(final String searchToken) {
 493         if (TransliteratorManager.init(null)) {
 494             final Transliterator normalizer = normalizer();
 495             return normalizer.transliterate(searchToken);
 496         } else {
 497             // Do our best since the Transliterators aren't up yet.
 498             return searchToken.toLowerCase();
 499         }
 500     }
 501
 502 }