!include once #1027304 // Eclipse Collections

import org.eclipse.collections.impl.map.mutable.primitive.*;

// compressor that takes input byte by byte and compresses
// as soon as possible.
// byte mode+case-sensitive only
// We now may have literals and pairs in any order
final sclass LCStreamingCompressor {
  int safety = 0;
  
  abstract sclass Chunk {
    abstract S text(L<Chunk> chunks);
    IntPair getPair() { null; }
  }
  
  srecord CPair(int i1, int i2) > Chunk {
    CPair(IntPair p) { i1 = p.a; i2 = p.b; }
    
    S text(L<Chunk> chunks) {
      ret linesLL_rtrim(chunks.get(i1).text(chunks), chunks.get(i2).text(chunks));
    }
    
    IntPair getPair() {
      ret IntPair(i1, i2);
    }
  }
  
  srecord CPrim(S s) > Chunk {
    S text(L<Chunk> chunks) { ret s; }
  }

  bool verbose = false, verboseCompressionSteps = false;

  new LineComp_PairIndex pairIndex;
  new L<Chunk> chunks;
  Map<Char, Int> literalIndex;
  new LongIntHashMap existingPairsIndex; // formerly Map<IntPair, Int>

    for (S line : allUniqueLines)
      chunks.add(new CPrim(line));
    primChunks = l(chunks);
    LS _lines = collect s(chunks);
    lineIndex = caseInsensitive ? ciListIndex(_lines) : listIndex(_lines);
    _lines = null;
    
    // simple encoding (only direct line references)
    simpleEncodings = mapValues(textIDToLines,
      (IF1<LS, L<Int>>) (lines -> wrapIntArrayAsImmutableList(mapToIntArray(lines, line -> lineIndex.get(line)))));
    //printAndCheckEncodings(simpleEncodings);
    if (verboseCompressionSteps) print("Have simple encodings");
    
    finalEncodings = compressPairs();
    if (verbose || verify) printAndCheckEncodings(finalEncodings);
  }
  
  void saveAsTextFile(File f) {
    S out = exportEncoding(finalEncodings);
    saveTextFile(f, out);
    
    if (verify) checkDecompression(f, textIDToLines);
  }
  
  void checkDecompression(File file, Map<S, LS> textIDToLines) {
    temp BufferedReader reader = bufferedUtf8Reader(file);
    LineCompReader lcr = new(reader);
    assertEquals(keysList(textIDToLines), asList(lcr.versions()));
    for (S version : keys(textIDToLines))
      assertEquals(lcr.textForVersion(version), lines_rtrim(textIDToLines.get(version)));
    if (verbose) print("Decompression OK for " + nVersions(textIDToLines));
  }
  
  S asText() { ret exportEncoding(finalEncodings); }
  
  S exportEncoding(Encodings encodings) {
    new LS buf;
    buf.add((byteMode ? "BYTECOMP " : "LINECOMP ") + primChunks); // magic signature
    for (Chunk c : chunks) {
      if (c cast CPair)
        buf.add(c.i1 + " " + c.i2);
      else
        buf.add(((CPrim) c).s);
    }
    for (S id, L<Int> l : encodings)
      buf.add(id + "=" + joinWithSpace(l));
    ret lines_rtrim(buf);
  }
  
  Encodings compressPairs() {
    pairIndex.verbose = verboseCompressionSteps;
    pairIndex.init();
    for (S version, L<Int> l : simpleEncodings)
      pairIndex.newChain(version, l);
    pairIndex.haveChains();
    simpleEncodings = null; // forget

    ret compressChains(keys(pairIndex.chains));
  }
  
  Encodings compressChains(Cl<S> files) {
    NotTooOften nto = new(printStatsEveryNSeconds*1000);
    IntPair toCompress;
    // Compress only most popular pair in one step
    while ping ((toCompress = pairIndex.mostPopularDuplicate()) != null) {
      if (safety > 0 && --safety <= 0) fail("safety");
      ++round;
      if (verboseStats)
        if (printStatsEvery > 0 && (round % printStatsEvery) == 0
          || round <= printStatsInFirstRounds
          || nto!)
          printStats();
      int count = pairIndex.getCount(toCompress);
      Int idx = makeCPair(toCompress);
      int dups = pairIndex.numberOfDuplicates();
      if (verboseCompressionSteps) print("Compressing pair " + toCompress + " (count=" + count + ") -> " + idx + ", " + (dups-1) + " remaining");
      pairIndex.replacePair(toCompress.a, toCompress.b, idx);
    }
    if (verboseStats) printStats();
    
    // reconvert to normal list
    ret mapToValues(f -> pairIndex.chains.get(f).toList(), files);
  }
  
  // TODO: check if pair is already there
  Int makeCPair(IntPair p) {
    long l = intPairToLong(p);
    S text = null;
    if (fullIndex != null) {
      text = itemToString(p.a) + itemToString(p.b);
      Int idx = fullIndex.get(text);
      if (idx != null) ret idx;
    }
    int idx = existingPairsIndex.getIfAbsent(l, -1);
    if (idx < 0) {
      idx = addAndReturnIndex(chunks, new CPair(p));
      existingPairsIndex.put(l, idx);
      if (fullIndex != null)
        fullIndex.put(text, idx);
    }
    ret idx;
  }
  
  void printAndCheckEncodings(Encodings encodings) {
    for (S id, L<Int> encoded : encodings) {
      if (verbose) print(id + ": " + joinWithSpace(encoded));
      S x = myFromLines(textIDToLines.get(id)), y = decode(encoded);
      if (caseInsensitive)
        assertEqualsIC(x, y);
      else
        assertEquals_quote(x, y);
    }
  }
  
  S decode(L<Int> encoded) {
    LS l = lambdaMap chunkText(encoded);
    if (byteMode) l = lines(lines(l));
    ret myFromLines(l);
  }
  
  // for non-byte mode
  S chunkText(int idx) {
    ret chunks.get(idx).text(chunks);
  }
  
  void printStats() {
    print("Round: " + round);
    pairIndex.printStats();
  }
  
  // byteMode only
  S itemToString(int idx) {
    new StringBuilder buf;
    itemToString_step(buf, idx);
    ret str(buf);
  }
  
  void itemToString_step(StringBuilder buf, int idx) {
    Chunk c = chunks.get(idx);
    if (c cast CPair) {
      itemToString_step(buf, c.i1);
      itemToString_step(buf, c.i2);
    } else {
      S s = ((CPrim) c).s;
      buf.append(byteMode ? charFromHex(s) : s + "\n");
    }
  }
  
  int nLiterals() { ret primChunks; }
  int nItems() { ret l(chunks); }
  
  IntPair getPair(int idx) {
    ret chunks.get(idx).getPair();
  }
}