RNC ProPack

From MultimediaWiki
Jump to navigation Jump to search

RNC Pro-Pack is a general data compression library (like zlib), RNC is acronym for author's company name, Rob Northen Computing.

Packer DOS binary, along with assembly decoding modules (barely useable nowdays) can be freely downloaded here.

There are several known variants in the wild, version 1 was employed to compress data files in various games, like Beneath a Steel Sky or Dungeon Keeper, version 2 seems to be used in ETV codec.

Header

All numbers are stored as big-endian.

 3 bytes   signature "RNC"
 1 byte    version
 4 bytes   unpacked data length
 4 bytes   packed data length
 2 bytes   CRC-16 of packed data
 2 bytes   CRC-16 of unpacked data
 2 bytes   overlap size (used for inplace unpacking)

Compression

As an extra protection measure against curious hackers compressed data can be encrypted with secret 16-bit key (supplied to decompression routine by programmer.) In that case all byte literals decoded as follows:

 byte ^= key; key = _rotr(key, 1);

Version 0

This is uncompressed data.

Version 1

Decoder source is in ScummVM repository.

Bits are packed into bytes and those are interspersed with byte values (like in KMVC for example).

Data is packed with LZ77 scheme into blocks, each block has three Huffman tables for literal and copy lengths and offsets.

Huffman tables:

 5 bits - number of entries (0-16)
 4x<number> bits - code lengths

for codes reconstruction and tree reading pleas refer to decoder code.

Block starts with three tables description and 16-bit number of (literal, copy) pairs.

 while (!end) {
   read literal length table
   read offset table
   read copy length table
   blocks = get_bits(16);
 
   while (blocks--) {
     read_length = get_value(literal_table);
     while (read_length--)
       *dst++ = get_byte();
     if (blocks) {
       offset = get_value(offset_table);
       length = get_value(copy_len_table);
       memmove(dst, dst - offset, length);
       dst += length;
     }
   }
 }

Version 2

This seems to be designed to be faster on decoding so it does not use Huffman tables (but it's still LZ77 scheme with the same bitreader).

 get_bits(2); // unused
 while (!end) {
   if (!get_bit()) {
     *dst++ = get_byte();
   } else {
     if (!get_bit()) {
       length = 4 + get_bit();
       if (get_bit()) {
         length = (length - 1) * 2 + get_bit();
         if (length == 9) {
           length = (get_bits(4) + 3) * 4;
           for (i = 0; i < length; i++)
               *dst++ = get_byte();
           continue;
         }
       }
       offset = get_offset();
     } else {
       if (get_bit()) {
         if (get_bit()) {
           length = get_byte() + 8;
           if (length == 8) {
              get_bit();
              continue; //dunno why
           }
         } else {
           length = 3;
         }
         offset = get_offset();
       } else {
         offset = get_byte() + 1;
       }
     }
     memmove(dst, dst - offset, length);
     dst += length;
   }
 }

Offset is coded like that:

 value = 0;
 if (get_bit()) {
   value = get_bit();
   if (get_bit()) {
     value = value * 2 + 4 + get_bit();
     if (!get_bit())
       value = value * 2 + get_bit();
   } else if (value == 0)
     value = get_bit() + 2;
 }
 offset = (value << 8) + get_byte() + 1;