Re: Are we losing users to Gentoo?
On Mon, 25 Nov 2002 19:07:07 +0100
Eduard Bloch <edi@gmx.de> wrote:
> #include <hallo.h>
> * Jim Lynch [Mon, Nov 25 2002, 09:54:10AM]:
>
> > > What we need to accept is there is a (percieved??)
> > > problem, or problems, with Debian as it stands today,
> > > these being (mainly)
> > >
> > > Hard to install (rubbish obviously)
> >
> > Nono, this is true, and primarily due to boot-floppies. One problem is
>
> Are you kidding? He compares the Gentoo cludge (manuall bootstrapping)
> with semi-intelligent boot-floppies setup.
I'm comparing boot-floppies with the potential I see in debian-installer;
I don't have a machine to install onto at the moment, so I have not
performed an install lately. With boot-floppies, I experienced so many
problems building that I'm not likely -ever- to do it again. And -work-
on it? Me? Not gonna happen. I'm not working on messes if there are better
alternatives available. And debian-installer is not available for all
arches yet, but now it seems the problems are smaller than I first
thought, which included the issues of building the whole thing under
the other arches. Having seen some of the code, it looks portable to
me. The problems of portability extend really to physical device access.
In joeyh's opening goals document, he's intelligently looked at
different divisions of work, and commented. I gather that it would be
exceedingly difficult to do things like alter the workflow of
boot-floppies; it looks like debian-installer could be rewired at will,
and further, it looks even like that ability was designed in from the
start. Overall, the codebase looks far more attractive to work on and to
come up to speed on. As far as I'm concerned, the faster debian-installer
can be made to work on all arches to the degree that installation is at
least possible on each, the better; from that point, it's just simple
tuning and polishing to get things working well.
But I have performed many debian installs with the boot floppy setup,
and I found that it still suffers from problems. One problem faced
by all dists is that of teaching people about partitioning and backing
up. At least the installer says "don't do this unless you're backed up".
debian-installer might solve that problem by offering to make all the
partitioning decisions.
dselect, for all its use once a person gets used to it, is not suitable
for a new person. Its interface is hostile in friendly clothes as well
as being antiintuitive. One of its biggest problems is that it does not
allow the user to do any reasonable thing at any point, something that
its interface belies that is possible. The best thing to have ever
happened to dselect is arguably apt-get.
To summarize, I think the potential of debian-installer is very large
and wide: built in is the ability to change wholly how it presents itself
to the user, even to the point of plugging in a graphical install -after-
everything else is working well. Of course, the development of the
graphical installer could proceed in parallel, but I think it's far more
important to get things working for every arch we support, and to ensure
that new arches can be easily included. The more eyes on this job, the
better, and the earlier debian-installer could be made generally
available.
By comparison, boot-floppies looks like kludges atop and beneath other
kludges, and I get the impression this is not easy to change without
affecting other aspects of the installer. I think that unless boot-
floppies gains major flexibility really fast, it should be supported
only until debian-installer has proven it can facilitate installs on
all arches. Once it has done so, boot-floppies should be shelved.
> > that it's quite difficult to work on boot floppies and then get
> > everything uptodate on the image and everything pointing right, etc.
>
> What? What do you mean? When did you install Debian the last time, in
> Slink days?
I didn't say install, I said work on. Read: it's right there, 7 lines
above this one.
> > It's also difficult to build.
>
> Compared to which alternative?
Compared to debian-installer.
> > debian-installer is much easier to build, is modular and also decouples
>
> FUD. Go and try first. From scratch!
Not FUD. I -did-. The only difficulty with debian-installer was that I
had to build the individual udebs, since they weren't in woody archives.
I would have to cd into each directory capable of producing udebs, and
typing dpkg-buildpackage, and copying the resulting udebs into a dir meant
for udebs that would have otherwise come from the archives across a net.
But they all built! And easily so! Things didn't all the way work at the
time (2-3 months ago?) but I'm hearing from the list that people are
reporting some success in installing debian on i386. If debian-installer
is as portable as it appears, problems ought to disappear quickly once
more eyes are on them.
How about yours?
> Gruss/Regards,
> Eduard.
-Jim
Reply to: