Hi, Bob.

BL>  BL> Isn't that normal? If you zip a file ten times you get ten
BL>  BL> subtly different zipped-sizes, all with the same CRC. It's a
BL>  BL> bloody nuisance,

BL>  FM> Are you sure about that? Ten different times with the same h/w
BL>  FM> & s/w congig?

BL>   I checked it again and Paul is right. But I tested it once before
BL> when I was writing a QWK reader trying to use the zipped size to
BL> identify the QWK packet... and I kept getting a different size. God
BL> knows why.

I think OLX uses a similar strategy, but probably records the size after
re-zipping so it doesn't run into that problem.

BL>  FM> See my message to Paul. If the compression algorithm can use
BL>  FM> more memory to keep a larger table, it can compress more
BL>  FM> effectively. Here's an example, loosely based on Richard
BL>  FM> Brent's compression algorithm...

BL>   Ahhh... that's possible. It was in VB and I was calling PKZIP.
BL> Perhaps Windows gave me different memory each time! It's certainly not
BL> doing it now, as a straight zip with lots of cache.

Cache won't matter, it'll be the amount of memory it can grab.

BL>  FM> Hopefully I haven't made any arithmetic errors there, in
BL>  FM> calculating the offsets backwards into the file. If I have, you
BL>  FM> should get the idea anyway.
BL>  FM> Oh, I've just noticed that I could have replaced the
BL>  FM> "he" string near the end with a pair
too, can't be
BL>  FM> bothered recalculating.

BL>   I see it all. Thanks for going to the trouble. Compression is on my
BL> list of things to learn about (after I've conquered zmodem).

Compression is a weird and wonderful subject. There are many different
techniques, some of which you've probably encountered already like the
fairly simple RLE compression which PCX uses. The "2-D RLE with
tokenising" which we discussed elsewhere (but I lost all the code I'd
written) is a variant.

Huffman coding is perhaps the most complex thing you'll have to get your
mind around, and is essential - it's part of every commercial
compression implementation I've studied. And just when you thought you
knew it all, some guru comes up with something new. There's a new
technique which promises to be even better than the top popular programs
like ZIP, LHARC & ARJ at least for text, which was published in Dr.
Dobbs earlier this year and is available from the author's web site
including source. Let me know if you want it. Of course the code is in
'C' which is some sort of low-level programming language and is
incomprehensible, but the text description gives a very well written
(IMHO) explanation.

Regards, fIM.

 * * Poured spot remover on my dog, now he's gone
@EOT:

---