[Libbt-devel] News on the Mac front - LibBT 1.03

[Dakidd <da...@so...>]

As of about an hour ago ago, I've got an error and warning free compile,
and functionality of both btcheck.c and btlist.c on a PowerMac 7500 hopped
up to a G4@350 running MacOS 9.1. Testing has been done using
locally-stored .torrent files that I've actually used for live transfers
using the "official" Python implementation of BitTorrent (note 1) in sizes
ranging from a couple hundred K to a couple-few-hundred megs, both
single-file and "directory" modes.

I'm pleased to report that, with the exception of a cosmetic glitch (Looks
like an "LL" suffix is being used, rather than an "L") in a printf format
string (From the debugger, I can see that the correct data is in the
variable, it's just getting printed wonky), execution has been flawless.
And speedy! As an example, hashing torrent pieces - I have a fileset
(.torrent and target file - Target file size just a shade over 175 megs)
that takes well over an hour and a half to hash and decide whether it is or
isn't complete when using the "Official" Python-based client. The same file
going through btCheck (which I'm building to use a slightly adapted public
domain SHA1 implementation called "ReidSHA", since OpenSSL hasn't been
ported to Classic MacOS (note 2) that I know of) is hashed and verified as
complete or not after only 51 seconds! That's with *ZERO* optimizations
being applied - No "hand-crafting ML byte-strings", no compiler
optimizations or use of special processor features, just a straight-up "as
it is written, so shall it be compiled". Once I decide to crank up the
compiler's brainpower by turning on some of the compile optimizations, I
expect the speed will improve even more as it starts sequencing and
pipelining the generated code and making use of currently unused processor
features like predictive branching or the G4's "Velocity Engine"... But I
digress...

The LibBT 1.03 tarball was my base code, and despite my bitching about
Kevin's coding *STYLE*, I have to be fair and say that I can find nothing
to bitch about in his *CODE* once it's been hammered into readability. A
couple of gaffes here and there, which is only to be expected in a project
like this, where the spec is a moving target, and the codebase is evolving
to hit it, but overall, nice and tight. It's just that the voyage to
readability can be frustrating! But, once it gets there, it's a sweet
codebase, with a lot of headroom, and pretty much ALL the toys that should
be needed to follow the protocol wherever it goes. And if it comes up
short? Well, that's the beauty of open source... SOMEBODY will get miffed
at the lack, and fix it! A million monkeys, with a million keyboards,
typing for a million years, just might up and (re)write Linux, too! Or
worse... Windows!!! AUGH!!!! Start shooting monkeys - QUICK! :)


Next step:
Hacking the bejeezus out of the stuff that tries to talk to the 'net. CURL
- FLUSH! Gonna use Apple URLAccess instead. That'll get me talking to the
tracker. Then it's on to try to change from *nix/Windows style networking
using the nice, simple, freindly, reliable, Sockets TCP/IP API, to Mac's
god-awful train-wreck of a TCP/IP implementation called "OpenTransport" so
that my servent can talk to others.

If I'm not back in 6 months, call the mounties and tell 'em I went
thattaway! And tell 'em to bring extra rope - I'll probably be be found
clinging to a knot tied at the end of mine! :)

BTW: Update, since I see I bailed out of Eudora before it could actually
post this message -

It's now several hours later, and the glitch in the printf (turned out to
be exactly what I thought it was: LL in a format string instead of L) has
been fixed. Operation and functionality of both btCheck.c and btList.c are
now 100% with all torrent files I've thrown at them (about 40, all of them
previously used to control actual transfers) so far. Making copies and
moving them around, then mangling the target data of the torrent in some
way to simulate damaged/missing pieces has been detected and reported as
present down to the single changed bit level, so the "Something's wrong
with the file" mode is working just as well as the "Yep, the file's all
there" functionality. I'm officially declaring this portion of the port to
Classic MacOS a success!

Now to get the networking to happen so I can build a client using a REAL
language instead of being stuck with that godforsaken Python interpreter.

Interpreter... <snork> I'd rather compile it! :)

<Ponders Applesoft (Apple IIe BASIC) or Apple Integer BASIC on a Mac>

<flees, screaming for tranquilizers, at the thought>

Anyway...

I *THINK* I've managed to conditionalize my changes in such a way that
Windows/*nix folks can compile my changed code by simply substituting their
own config.h file with the last line before the final #endif reading
"#define TARGET_MACOS 0" or "#undef TARGET_MACOS". My intent was to be sure
that my changes are completely transparent, but I cannot claim with
certainty that I've managed it.

(note 1) Painfully slow, written in an interpreted language (Python) whose
only available (to my knowledge) Mac-hosted interpreter keeps crashing on
me. Python is also a language which bears the second curse of being
"foreign" to me - I don't grok Python (and to be blunt, I don't WANT to
grok Python - it makes my brain itch) so I can't fiddle with it to try to
make it faster, despite having the source code staring me in the face.
Hence, my motivation to get a C-based BitTorrent "toolkit" operational on
the Mac. To make things worse, the interpreter has its own bugs and
glitches, so that about half the time, I wonder whether the most recent
problem I've encountered has been with BitTorrent or a problem with Python.

(note 2) For you non-Mac folks, "Classic" distinguishes the older MacOS
(9.2.x and below) from "The cats" - MacOS X - AKA 10.x.x AKA
Jaguar/Panther/Etc. (I think we're officially up to Tiger now - Or does
that release next week? It's either just released, or releasing very soon)
MacOS X is based on *nix - a sub-flavor of NetBSD, if I recall rightly -
while MacOS 9.x and earlier are now grouped as "Classic", and despite
looking pretty similar on the screen, are pretty significantly different
than X when you crawl under the hood and start banging the bits directly
against the silicon.

Don Bruder -  da...@so...      <--- Preferred Email - unmunged
I will choose a path that's clear: I will choose Free Will! - N. Peart