Subject: Re: Nice to see NetBSD mentioned. However...
To: None <netbsd-advocacy@netbsd.org>
From: Miles Nordin <carton@Ivy.NET>
List: netbsd-advocacy
Date: 01/08/2001 16:25:34
sysinst is not an ``old-style text installer.'' It has sane, efficient
menus and widgets, and it can display a command's output inside a
subwindow. The Twocows pontificator seems to be confusing ANSI-color
and proprietary IBM PeeCee line drawing characters with ``modern
installer.'' These characteristics make no useability/friendliness
difference whatsoever, and they badly mangle output on many terminals
that NetBSD supports. A long time ago, a Bell Labs hacker developed the
``termcap'' idea. Then RedHat broke new ground in the so-called-Unix
community by throwing termcap's basic sanity out the window, in favour of
installers based on old DOS-only ``terminal emulators'' for terminals that
never existed---most of them had a terminal called ``ANSI-BBS'' which is
what the RedHat and Debian colorful/linedrawing installers seem to be
patterned after. It's no wonder that the same people who used to struggle
so hard to get a full-screen editor to work want us to concentrate on the
pretty lines and colors. Speaking as one who has tried to use AlphaBIOS
over a serial console, I can only call these architectural decisions
embarassingly foolish, but not half so foolish as one who looks at a
correctly-functioning system and calls it ``incomplete'' just because he's
never seen a system work correctly before. This type of bogon
design philosophy works great for side-scrolling video games based on
character-cel glyph blocks like ``Legend of Zelda,'' which is a fine
game by the way, but the philosophy has no place in any complex system,
or any system accountable for large amounts of money.
I'd also like to point out that several of the (completely rewritten)
RedHat special-installation-disk-partitioners I've used make off-by-one
errors in determining the start- and end-points of the partitions. Not
only does this make installation difficult, but if an operating system's
culture is incapable of recruiting and training programmers who can
answer basic seventh-grade-math-bowl questions like ``how many posts
are in a six-meter fence with posts every meter?'', do you really want
to trust their programs with valuable data?
Why is it that when these Linux/PeeCee-centric organizations review an
operating system, they dwell almost exclusively on _installation_,
and do not attempt to make observations about the running system from
their experiences as I just did? Is their goal run Unix,
or to install Unix?
Maybe we should all take a step backwards, and try to decide whether we're
here to write a Unix or here to write a Unix installer. If we're here to
write an installer, we may as well follow Linux's lead by halting
innovation on UVM, UBC, LFS, SMP, and ALTQ so that we can devote more
resources to designing a desktop widgetset that includes a WordBASIC
emulator that we can use in our installer. It might also help to put
statically-linked Perl in /bin and crunchide, and use it to rewrite sysinst
from scratch, with some bash and Tcl glue scripts and imlib bindings
thrown in---everyone is talking about Perl these days, and it's huge,
so it _must_ be the language of the future. We can either recode all
of NetBSD in Perl now, or find ourselves left behind when all the
competing distros go to grafical inst's. If NetBSD doesn't start
embracing the new advanced technologies like web-enabled administration
tools with .asp's and .psp's, it won't be able to Compete with the
installers of other Open Source (tm) Operating Systems.
I'll tell you one thing for sure: writing installers seems to pay a
helluva lot better than writing operating systems! Who do we have to
thank for this absurdity?
If installers are really in such demand, it seems to me you'd want to
choose the hardest-to-install Unix you can handle, so as to have the
greatest boasting-rights. Then, if someone installs a Unix which is
different from the one you installed, you should be prepared with remarks
about how that Unix is ``easy'' or ``lightweight'' or ``watered down for
the unwashed masses,'' thus proving your superiority. Given that scenario,
I can understand why Linux is so popular, and I can understand why a
Unix with simpler, less arcane, and more openly-documented installation
procedures is maligned in the same article. But, why is it maligned for
being hard to install, when difficulty-of-installation is the virtue
that earns users in this community their respect?
You know what? It doesn't matter! Why? Because the PeeCee is dead!
Why do you think NetBSD has been concentrating on the inexpensive
interfaces (USB, IDE), on real-time kernel preemptibility (SMP),
on the networking stack (IPsec, ALTQ, IPv6), on the key embedded CPU's
(m68k, ppc, mips, arm, sh3), on a unified ELF toolchain, and on a
cross-compilable build architecture? Twocows is _not_ part of any
sane master-plan. Don't be reactionary about this. There are no
constructive comments in that nonsense article. We do not need that
writer's attention or approval. Trust me: when the shit hits the fan,
NetBSD will be the only open project even close to ready for it.
What I'm really concerned about is, what's going to happen to the
project's culture when people concerned more with making money than
with writing code start wanting to contribute poorly-written junk.
I am almost certain it will fork <cough>, perhaps many times, when code
starts getting rejected. People looking to make a purchase only see
what you have, not how you got it.
Then again, as long as the American tech industry completely collapses,
we should be fine.