transient sclass SingleTextWordIndex {
  S regexp = "\\w+";
  new ElementInstanceMatrix<Int, S> wordMatrix;
  int length;
  
  *(S text) {
    init(text);
  }
  
  *(S *regexp, S text) {
    init(text);
  }
  
  void init(S text) {
    length = l(text);
    wordMatrix.numberToInstance = wordMatrix.instanceToNumber = i -> i;
    for (IntRange r : regexpFindRanges(regexp, text))
      wordMatrix.add(r.start, ll(upper(substring(text, r))));
    wordMatrix.doneAdding();
  }

  LPair<S, Int> wordsAndOffsets(S text) {
    ret map(regexpFindRanges(regexp, text),
      r -> pair(upper(substring(text, r)), r.start));
  }
  
  // assumes word boundaries left and right of query
  int[] preSearch(S query) {
    ret indicesOfWordCombination(wordsAndOffsets(query));
  }
  
  int[] indicesOfWordCombination(LPair<S, Int> wordsWithOffsets) {
    int n = l(wordsWithOffsets);
    if (n == 0) null;
    if (n == 1) ret intArray_minus(first(wordsWithOffsets).b, wordMatrix.instancesContainingElement_intArray(first(wordsWithOffsets).a);

    // get entries for words, exit when a word is unknown
    ElementInstanceMatrix.Entry[] entries = new ElementInstanceMatrix.Entry[n];
    for i to n: {
      ElementInstanceMatrix.Entry e = wordMatrix.index.get(wordsWithOffsets.get(i).a);
      if (e == null) null;
      entries[i] = e;
    }
    
    // go through words again, shift & AND-combine all bit sets
    BitSet bs = leftShiftBitSet(wordsWithOffsets.get(0).b, cloneBitSet(entries[0].bitSet()));
    for (int i = 1; i < n; i++)
      bs.and(leftShiftBitSet(wordsWithOffsets.get(i).b, entries[i].bitSet()));

    ret bitSetToIntArray(bs);
  }
}