BA 1.01 beta - 2000 revision 5,  3 October 2000.

This is still experimental software under constant development.
As usual, use it at your own risk!

It's a Burrows and Wheeler transform implementation.
I'm pushing it forward to give better compression and get faster,
in that order. With better models, estimations of contexts that is
getting better and better, fragmentation etc. compression is getting
better in a steady pace. I expect that results should end up in the
same neighbourhood as of the best ppm implementations but with the
same obvious advantages as earlier:
    speed and efficient use of precious memory.

This new version has improvements in compression too but mainly
higher speed in sorting.
I'll continue to release new versions of ba as it's getting better.
They are demos of what is currently possible with the bwt algorithm.
In the future when I feel content about its performance, I will
release the source code in some form.

I've implemented Sadakane's idea about only sorting buckets that
actually needs sorting.
I've also fixed two bugs. Blocks of size < 20 bytes got the last
symbol wrong when decoded. I also forgot to set the symbol at 
start[-1] that could cause a faulty sort in rare cases.
Also some minor adjustments of the analyzing part has been done.

I have tried to simplify usage in this version.
You don't need to care about the different flags, it's designed
to estimate best possible compression by it self.
What you need to know though is:
-v  to make it display results.
-nr where nr is a number from 1 to 50, 1 to choose 100k blocks.
    Default size is filesize unless filesize > 900k.
-z  to adjust blocksizes to contents, good for most binary files
    sometimes even on texts in a tar archive.

ba -2vz file1 file2 file3 ...
will pack file using 200k blocks, be verbose and try to adjust blocksizes
to contents.
It's using 5n + C of memory. E.g. a 200k block will need
5 * 200k + 100k = 1100k (something like that).

Some results (with the -z option):
calgary.tar     815 kb
canterbury.tar  475 kb
book1           213 kb
geo             49 kb
kennedy.xls     90 kb  (72k if ba -1p)
act-txt.tar     939 kb (from archive compression test by Jeff Gilchrist)

Note that on canterbury.tar, geo, kennedy.xls and act-txt.tar compression
is already superior to ppm. Ppm still has an advantage on small text-files
because of excellent control and predictions on context.
I'm working to change this though.

If you have ideas, opinions or bugs to report, you're welcome to write
to me at mikael.lundqvist@spray.se
