[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: [desktop] Unix configuration nightmare

On Tue, Oct 22, 2002 at 09:54:39AM -0400, Jaldhar H. Vyas wrote:
> I think it would be more useful if this discussion stayed focused on
> simple, concrete things we could do to improve the Debian desktop right
> now rather than pie in the sky ideas like redoing the entire Unix way of
> configuration.

Work on the install tool. http://gluck.debian.org/debian-cd/ has bootable
(for i386) sarge CDs, regenerated daily based on debian-installer. A _lot_
of work is needed.

Note that at this point there is *no* prospect of PGI being available
as an install option on official CDs, so don't assume that the work
done there (in the UI, in autodetection, in _anything_) will have any
affect on sarge's installer. Before anything like that will happen it'll
require a lot of effort from someone to either port PGI's functionality
across to d-i, or to restructure PGI so that images can be easily and
automatically constructed, and so that it supports the entire range of
machines and installation methods we wish to support.

d-i's getting close to it's second birthday, and even with the fairly
active development it's enjoyed the last three months, it still hasn't
really reached the point of "functional". At the rate we're going,
we're anywhere up to a year and a half away from having something we
can test and build upon, and that impacts the quality of the rest of
our distribution by making it absolutely impossible for anyone to do
installation testing, which is our main line of defence from packaging
bugs [0].

Some concrete suggestions:

	* Help clean up and fix debian-installer. There are huge numbers
	  of problems; try an install, find something obnoxious, fix it,
	  rinse, repeat.

	* Rework the boot-floppies system into udebs. debian-installer is
	  suffering from two independent sets of problems at the moment:
	  one is that the udeb system is completely untried, and it's
	  difficult getting everything needed in udeb form. The other
	  is that debconf is completely untried for the install tool,
	  and that throwing up some design issues. Having boot-floppies
	  as udebs would allow us to shake out one set of problems,
	  without having to worry about the others, and would give us a
	  chance at a fallback position if our patience runs out before
	  debian-installer is done.

	* Rework PGI into udebs. Similar benefits, similar difficulties,
	  but additionally: PGI has only ever worked on a couple of
	  architectures, essentially requires a CD, and requires a
	  significant number of packages that aren't already udeb'ed;
	  and on the upside, is reportedly much more pleasant than
	  boot-floppies or debian-installer.

Note also that we'll need to ensure that sarge works well with the next
Linux kernel (2.6/3.0) -- we're _highly_ unlikely to release before it's
stable, even given the usual teething problems stable kernel releases
have, and that it's only just approaching its (first?) feature freeze.

Seriously, having a crappy installer -- or worse, not even having a
crappy installer -- really is hurting Debian's effectiveness, both at
"world domination" and at simply meeting our own needs and expectations.


[0] Upgrade testing is great, except for two things: it's incredibly
    difficult to give repeatable reports, since both your system and the
    archive are moving targets; and our biggest problem isn't in handling
    upgrades, it's in handling new installs. Hands up anyone who's done a
    few woody installs and _not_ come across any unnecessary annoyances,
    or any bugs.

Anthony Towns <aj@humbug.org.au> <http://azure.humbug.org.au/~aj/>
I don't speak for anyone save myself. GPG signed mail preferred.

 ``If you don't do it now, you'll be one year older when you do.''

Attachment: pgpvqiKRKOgIt.pgp
Description: PGP signature

Reply to: