Re: tlug: Re: djb [was: ibm.net with LINUX (Red Hat)]

["Stephen J. Turnbull" <turnbull@example.com>]

>>>>> "Karl-Max" == Karl-Max Wagner <karlmax@example.com> writes:

    Karl-Max> Seconded. Qmail IS good code - just look at the clearcut
    Karl-Max> structure of it.

XIM has clear structure too, but it sucks, or did back in X11R5.  It
was easy enough to understand where shortcuts were being taken.

    Rex> code because he rarely uses standard idioms and libraries
    Rex> (witness his "substdio" library) but the stuff he writes is
    Rex> secure and rock solid reliable.

    Karl-Max> Actually he gives reasons why he uses his own stuff. And
    Karl-Max> weeding out a library to the basics wil debloat it - and
    Karl-Max> code bloat very often ensues in reliability and security
    Karl-Max> risks.

Granted.  But you need to trust the author(s).

    >> When you don't have time to grok code (read "lazy" if you
    >> want), and the code doesn't use standard idioms, reliability is
    >> hard to judge.  "It's my code so it's good" doesn't build
    >> confidence until I have independent corroboration....

    Karl-Max> Actually, there's lots of qmail supporters - people of
    Karl-Max> known quality like Russ Nelson.

Sure.  But Russ may support it because it's easy to use.  All I know
about Russ (besides the legendary Crynwyr packet drivers) is how
painful he found it to run the DJGPP (and other) mailing lists using
sendmail 6.something + USCD listserv.  If that was where I was coming
from, I'd consider Smail heavenly!  :-P  Point is, I don't know that he
has grokked the code, either.

And at the time I encountered qmail, I had nothing to go on except Jim 
Tittsler's statement that he thought qmail could be configured to
prevent spam relaying, and Bernstein's glowing review of his own
program, and disparagement of protocols other than QMTP.  That feature 
was not well-documented, and Bernstein's evident disdain for sendmail
left me with worries that it would not be a "drop in" replacement for
smail.  I was in a hurry, as you'll see....

So I blew off qmail, and spent about 4 hours finding out that the
newest beta of smail had the necessary feature---and fixing it so it
actually worked.  The code was not pretty, but it was modular enough
that I could localize the problem and fix it.  An unreliable
kludge---I didn't bother to submit it 'cause I knew it would only work
for me, and the next version of Smail completely rewrote the subsytem
anyway.

The spammers took about three weeks to stop trying to relay through my 
box.

And that's one reason why people don't make radical changes even
though they know their legacy system isn't up to snuff---the legacy
system works and they don't want to chance downtime because the new
wonderful stuff tickles a bug in existing software.

    Rex> His approach to protocol design is equally dependable --
    Rex> rather than develop protocols that work correctly only if
    Rex> everyone follows the spec to the T, he advocates designing

    Karl-Max> Not only that. He simply sees the problems with some
    Karl-Max> protocols - and there are a lot of Internet protocols
    Karl-Max> that are simply outmoded and ought to be replaced. He

Yup.  There are still a few 7-bit mail gateways out there.  They ought
to be replaced.  Millions of Windows machines use Shit-JIS.  _They_
ought to be replaced.  Your point is...?

Arrogance does not consist in being right and knowing it loudly.
Arrogance consists in being right and thinking that justifies
demanding that others correct their mistakes, at their expense.

    Karl-Max> simply states that and in the same time already provides
    Karl-Max> a better solution.

It's not unqualifiedly better unless it's backward compatible, too.
But that's impossible, because (most of what I've seen I agree with
him _in theory_) he _is_ right, after all. :-)

    Rex> protocols that work correctly even if implementors just
    Rex> follow examples and code to "common practice" (ref:
    Rex> http://pobox.com/~djb/proto/design.html).

    Karl-Max> To put it in a nutshell: they are reliable and secure -
    Karl-Max> no "ivory tower" designs but designs made for the real
    Karl-Max> world where things often go wrong.

Um, I'm lost.  Either it's fuzzy and works with common practice, or
it's reliable and secure.  You don't get both.

    >> Problem with that approach is that protocols interact, and
    >> there are legacy systems issues as well.

    Karl-Max> It's seldom a good idea to stick to legacy technology it
    Karl-Max> that is known to fall short of present
    Karl-Max> requirements. That kind of thinking actually stifles
    Karl-Max> technical progress - and this is dangerous.

Good enough is good enough.  Stifling technical progress is dangerous
only to employment of technicians.  :-)

Seriously, who do you think is sticking to legacy technology for the
sake of the technology?  I'd love to have a 400MHz Alpha or an SGI on
my desk, but I'm stuck with legacy technology.  I guess that makes me
dangerous.

Or take spam.  PDAtropos told me that it was impossible to implement a
spam filter for a system like AOHell's.  They could write one, they
just didn't dare install it without thorough testing.  And it wasn't
just AOHell that was forwarding spam, earthlink still does as far as I
can tell.  AOHell came around, but not for at least 12 months after
that plaintive cry from their Postmaster.  (And by that time he was
most definitely on the side of the angels, unlike his attitude when
AOL sudenly became a plague upon Usenet.)

Or take Cobol.  Ed Yourdon claimed that as of the early 90s more lines 
of code were being written in Cobol than in any other language
family.  Few of them were new programs, of course.  But they were
lines of code.  Cobol!  Just like that purple dinosaur, it won't go
away.

    I love you, you love me
    In Cobol you can count to three
    But it takes ten lines of seven tokens each
    Because it looks like human speech.

Well, sort of kind of.

    Karl-Max> happens, common engineering practice tells us that it is
    Karl-Max> better to start a new design from the ground up and to
    Karl-Max> discard the old one. Which is exactly what djb did.

And so did the guys who wrote smail, and the guy who wrote exim.
And the guys who created TrueType and NetBIOS and Windows NT.

Who among those got it right, and how can you tell by listening to
them?

    Karl-Max> What it comes down to: Far sighted people that
    Karl-Max> agressively point out deficiencies others are unable to
    Karl-Max> see are often considered arrogant.

The problems with RFCs 821 and 822 are well-known.  But for heaven's
sake we can't even make the move to ISO-Eurodate (whatever the number
is, you know what I'm referring to), let alone ditch the CRLF
convention.  It's not inability to see the problem that delays change.

Aggressively pointing out deficiencies that others are unable to see
is not arrogant in itself.  Justifying the aggression by the
presumption that others don't see as well as you do, is.

    Karl-Max> Actually, it saved ME a lot of hassles. Since then I am
    Karl-Max> always careful with the tag "arrogant" with people that
    Karl-Max> have unpopular views.

Being arrogant doesn't increase the likelihood of being wrong, it just 
increases the probability of being irrelevant.

-- 
University of Tsukuba                Tennodai 1-1-1 Tsukuba 305-8573 JAPAN
Institute of Policy and Planning Sciences        Tel/fax: +1 (298) 53-5091
--------------------------------------------------------------
Next Nomikai: 18 September, 19:30 Tengu TokyoEkiMae 03-3275-3691
Next Meeting: 10 October, Tokyo Station Yaesu central gate 12:30
--------------------------------------------------------------
Sponsor: PHT, makers of TurboLinux http://www.pht.co.jp