Class FTIndex

basex-core/src/main/java/org/basex/index/ft/FTIndex.java:53–515

This class provides access to a fuzzy full-text index structure stored on disk. Each token has an entry in sizes, saving its length and a pointer on ftdata, where to find the token and its ftdata. The three database index files start with the prefix DataText#DATAFTX and have the following

 * @author Christian Gruen
 */
public final class FTIndex extends ValueIndex {
  /** Minimum fixed size for each token entry. */
  private static final int ENTRY = 9;

  /** Cached texts. Increases used memory, but speeds up repeated queries. */
  private final IntObjectMap<byte[]> ctext = new IntObjectMap<>();
  /** Levenshtein reference. */
  private final Levenshtein ls = new Levenshtein();

  /** Index storing each unique token length and pointer
   * on the first token with this length. */
  private final DataAccess dataX;
  /** Index storing each token, its data size and pointer on the data. */
  private final DataAccess dataY;
  /** Storing PRE and POS values for each token. */
  private final DataAccess dataZ;

  /** Cache for number of hits and data reference per token. */
  private final IndexCache cache = new IndexCache();
  /** Token positions. */
  private final int[] positions;

  /**
   * Constructor, initializing the index structure.
   * @param data data reference
   * @throws IOException I/O Exception
   */
  public FTIndex(final Data data) throws IOException {
    super(data, IndexType.FULLTEXT);
    // cache token length index
    dataX = new DataAccess(data.meta.dbFile(DATAFTX + 'x'));
    dataY = new DataAccess(data.meta.dbFile(DATAFTX + 'y'));
    dataZ = new DataAccess(data.meta.dbFile(DATAFTX + 'z'));
    positions = new int[data.meta.maxlen + 3];
    final int pl = positions.length;
    for(int p = 0; p < pl; p++) positions[p] = -1;
    for(int is = dataX.readNum(); --is >= 0;) {
      final int p = dataX.readNum();
      positions[p] = dataX.read4();
    }
    positions[pl - 1] = (int) dataY.length();
  }

  @Override
  public synchronized IndexCosts costs(final IndexSearch search) {
    final byte[] token = search.token();
    if(token.length > data.meta.maxlen) return null;

    // estimate costs for queries which stretch over multiple index entries
    final FTOpt opt = ((FTLexer) search).ftOpt();
    return IndexCosts.get(opt.is(FZ) || opt.is(WC) ? Math.max(1, data.meta.size >> 4) :
      entry(token).size);
  }

  @Override
  public synchronized IndexIterator iter(final IndexSearch search) {
    // current search token
    final FTLexer lexer = (FTLexer) search;