The purpose of this document is to detail the format of "squashed"
files created by PKARC version 2.0 or later.  This document assumes
some basic knowledge of existing ARC formats and various compression
techniques.  For more information consult the references listed at the
end of this document.

The general format for an ARC file is:

[[archive-mark + header_version + file header + file data]...] + 
archive-mark + end-of-arc-mark

The archive-mark is 1 byte and is the value 1A hex.  The file header
can be defined by the following 'C' structure, and is 27 bytes in size.

typedef struct archive_file_header
  { char name[13];		/* file name */
    unsigned long size;		/* size of compressed file */
    unsigned short date;	/* file date */
    unsigned short time;	/* file time */
    unsigned short crc;		/* cyclic redundancy check */
    unsigned long length;	/* true file length */
  };

The name field is the null terminated file name.

The size is the number of bytes in the file data area following the
header.

The date and time are stored in the same packed format as a DOS
directory entry.

The CRC is a 16-bit CRC on the file data area based on a CRC polynomial
from the article by David Schwaderer in the April 1985 issue of PC
Technical Journal.

The length is the actual uncompressed size of the file.



The header versions are defined as follows:

Value	Method		Notes
-----	--------	-----------------------------------------------------
  0	  -		This is used to indicate the end of the archive.
  1	Stored		(obsolete) (note 1)
  2	Stored		The file is stored (no compression)
  3	Packed		The file is packed with non-repeat packing.
  4	Squeezed	The file is squeezed with standard Huffman squeezing.
  5	crunched	The file was compressed with 12-bit static Ziv-Lempel-
			Welch compression without non-repeat packing.
  6	crunched	The file was compressed with 12-bit static Ziv-Lempel-
			Welch compression with non-repeat packing.
  7	crunched	(internal to SEA) same as above but with different
			hashing formula.
  8	Crunched	The file was compressed with Dynamic Ziv-Lempel-Welch
			compression with non-repeat packing.  The initial
			ZLW code size is 9-bits with a maximum code size
			of 12-bits (note 2).  An adaptive reset is used
			on the ZLW table when it becomes full.
  9	Squashed	The file was compressed with Dynamic Ziv-Lempel-Welch
			compression without non-repeat packing.  The initial
			ZLW code size is 9-bits with a maximum code size
			of 13-bits (note 3).  An adaptive reset is used
			on the ZLW table when it becomes full.

Note 1:
  For type 1 stored files, the file header is only 23 bytes in size,
  with the length field not present.  In this case, the file length
  is the same as the size field since the file is stored without
  compression.

Note 2:
  The first byte of the data area following the header is used to
  indicate the maximum code size, however only a value of 12 (decimal)
  is currently used or accepted by existing ARC programs.

Note 3:
  The algorithm used is identical to type 8 crunched files with the
  exception that the maximum code size is 13 bits - i.e. an 8K entry
  ZLW table.  However, unlike type 8 files, the first byte following
  the file header is actual data, no maximum code size is stored.



References
----------

Source code for ARC 5.0 by Tom Henderson of Software Enhancement Associates,
usually found in a file called ARC50SRC.ARC.

Source code for general Ziv-Lempel-Welch routines by Kent Williams, found
in a file LZX.ARC.  Kent Williams work is also referenced in the SEA
documentation.

Source code and documentation from the Unix COMPRESS utilities, where most
of the ZLW algorithms used by SEA originated, found in a file called
COMPRESS.ARC.

Ziv, J. and Lempel, A. Compression of individual sequences via
variable-rate coding. IEEE Trans. Inform. Theory IT-24, 5 (Sept. 1978),
530-536.

The IBM DOS Technical Reference Manual, number 6024125.


	- Phil Katz, 12/27/86