BACKASCII

				John Osudar

			   Electronics Department
			Argonne National Laboratory
				  205 A-051
			   9700 South Cass Avenue
			   Argonne, IL 60439-4837

			       Phone numbers:
			       FTS: 972-7505
			      (312) 972-7505

			 Electronic mail addresses:
			 Bitnet:  B35049 at ANLCMT
			    MFENET:  B35049@AN2

BACKASCII consists of a pair of utility programs, BACKPACK and BACKUNPACK.
These programs were originally designed to convert BACKUP saveset files to and 
from a packed ASCII format, so that such files could be sent over 
communications links (e.g. networks such as Bitnet) which do not accept, or
incorrectly translate, eight-bit binary data, and possibly even some seven-bit 
printable characters.

BACKPACK works by counting the frequency of each byte value in each record of 
a file, writing a header indicating which values have been replaced by 
substitute characters, and then ASCIIfying and compressing the record.  
Identical consecutive values may be replaced by repeat counts, and the most 
frequently occurring values are replaced by single printable characters from 
the set "G"-"Z" and "a"-"z".  Less common values are represented by their
two-character hexadecimal equivalents.  The resulting "packed" file has 
variable-length records of at most 80 characters, and contains only upper-case 
and lower-case letters, along with digits, periods, and commas.  Options may 
be specified to restrict the use of lower-case letters as well, for use in 
cases where only upper-case data can be transferred, or to expand the set of
replacement characters to include all non-blank printable ASCII characters 
(except the hex digits, comma, period, and dollar sign) for cases where all
those are known to transmit correctly.  (The dollar sign and space are never
included in any packed file.  The reason is that the dollar sign might occur
in column 1, preventing the "distribution" file from working -- the line would
appear to be a DCL command -- and some communications software does strange
things with spaces, such as delete trailing ones.) 

BACKUNPACK reads the packed file, and recreates the original file.  The 
original filename and extension are written into the packed file by BACKPACK, 
so the unpacked file will have the same name as the original.  BACKUNPACK 
knows how to unpack upper/lowercase, uppercase-only, and "extended" format 
packed files.

Although BACKASCII was designed to process BACKUP savesets, which are 
fixed-length record files with a large recordsize, it can be used to process 
any file of fixed-length or variable-length records, e.g. a VMS executable
image, a graphics metafile, etc.  (Note that, once the recordsize is large
enough that there aren't enough replacement characters to handle all
replaceable byte values, increased recordsize reduces efficiency.  However, 
small records tend to have more unreplaceable byte values, i.e. values that 
produce replacement savings of less than three bytes.) 

When used with BACKUP savesets created with /BLOCK=2048, packed files that are 
10% to 20% smaller than the original saveset have been seen.  With graphics 
metafiles averaging over a block per record, packed files were up to 30% 
larger than the original.  Packed EXE files vary, from around 10% smaller than 
the original to as much as 20% larger.  Using the "extended" format, an
additional 2% to 5% size reduction was seen on savesets, while graphics
metafiles were only 20% to 25% larger than the original file. 

BACKASCII is the command procedure that drives both utilities.  It is executed 
with parameters as follows:

	@BACKASCII   command   inputfile   {outputfile}

"command"  may be:	PACK  to produce a packed upper/lowercase output file;
			PACK/U to produce a packed uppercase-only output file;
			PACK/X to produce a packed full-ASCII output file;
			UNPACK to unpack a packed file

"inputfile" is the name of the file being processed

"outputfile" is required on the PACK commands, and optional on UNPACK.  If 
		specified on UNPACK, it should include the device and/or
		directory where the file is to be written; the filename and
		file type of the original file (as contained in the packed
		file header) will be retained.

BACKDIST is a template command procedure that includes the (uncommented)
source of BACKUNPACK, and provides a place to insert a packed file.  By
editing BACKDIST and inserting a packed file where indicated, the resulting
command procedure can be sent to another VMS site, where executing it will
produce a copy of the original file.  (Fortran is required on the receiving
node in order to make use of this.) 

Note that BACKUNPACK is written entirely in VAX Fortran, and calls no system 
services or other VMS-specific routines directly.  It should be relatively 
easy to convert BACKUNPACK to work with a non-VMS Fortran-77 compiler, in 
which case it would be possible to ship packed files of binary data (e.g. 
graphics metafiles) to non-VMS hosts and unpack them into original form at the 
destination.

Also note that the packing and unpacking processes require a fairly large 
amount of CPU time, although this version has been optimized to work a lot 
more efficiently than the original version.  (On a 785, typical packing and 
unpacking rates of 50 (unpacked) blocks per CPU-second have been observed -- 
i.e. a 6000-block saveset might require two MINUTES of 785 CPU time to pack 
or unpack!)