Subject: Re: proposed mods to config(8) and the kernel build process
To: Matthew Jacob <firstname.lastname@example.org>
From: Greywolf <email@example.com>
Date: 12/11/2000 12:38:49
On Mon, 11 Dec 2000, Matthew Jacob wrote:
# Date: Mon, 11 Dec 2000 12:11:35 -0800 (PST)
# From: Matthew Jacob <firstname.lastname@example.org>
# To: Luke Mewburn <email@example.com>
# Cc: NetBSD Kernel Technical Discussion List <firstname.lastname@example.org>
# Subject: Re: proposed mods to config(8) and the kernel build process
# > Is there any reason not to provide this functionality until a separate
# > kernel configuration mechanism exists? (which hasn't even been
# > discussed let alone had an implementation started or completed :)
# No, of course not. I'm just pointing out that before one adds a feature one
# should step back and say "Will we want this in two years?".
# > ``oh no, don't do <foo>, we'll have feature <bar> to do that better''
# > two years go by, no sign of <bar>, we've missed out on being
# > slightly better for our users because there was no <foo>
# > either :-|
# > Can someone point me in the direction of docco or what part of FreeBSD
# > does this? If we do decide we like the functionality and their scheme
# > is reasonable, we probably should look at using it.
# I think I mentioned this at BSDcon- there was some talk of parts of this
# as part of the kernel loader talk- I don't know how detailed it is and
# how well documented other than in source.
# But the basic gist is that in committing to a (in FreeBSD's
# implementation, froth) comman loader as your second stage bootstrap you
# can then provide a consistent boot interface to all platforms. The
# fallout of this is that things like ports and irqs for configurable
# devices like ISA devices are now a set of hints sitting in a file off of
# /boot (e.g., see below) that can be overridden.
# This makes the notion of a stored config file not as useful or desirable
# as it used to be. Possibly compilation options still, but if the intent
# is to ultimately have it all be knobs, why go to the effort.
What if /boot/ is spuriously corrupted, either as an fs or as a dir?
I think the idea of config information being stored exclusively in
a file instead of in the kernel is asking for trouble, as if it is in
an irretrievable file, one will not even be able to arrive at single-user
mode unless one has an "emergency repair disk" (uck).
...or are my paleolithic roots showing, again?
# This may not be desirable (or practical) for NetBSD. I was just mentioning it
# so that before the establishment of an invasive feature, even a minor one,
# that some thought was applied.
This is not to toss the idea completely; it just makes *me* twitch a bit.
*BSD: Twice the Bits-Clean of other Leading OSes.