Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 15 Dec 1999 10:57:36 +1030
From:      Mark Newton <newton@internode.com.au>
To:        Ryan Thompson <freebsd@sasknow.com>
Cc:        "Daniel C. Sobral" <dcs@newsguy.com>, Peter Jeremy <peter.jeremy@alcatel.com.au>, current@FreeBSD.ORG
Subject:   Re: sysinstall: is it really at the end of its lifecycle?
Message-ID:  <19991215105736.A467@internode.com.au>
In-Reply-To: <Pine.BSF.4.10.9912141705440.76991-100000@sasknow.com>
References:  <3856BD33.5DE1AB48@newsguy.com> <Pine.BSF.4.10.9912141705440.76991-100000@sasknow.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Dec 14, 1999 at 05:37:09PM -0600, Ryan Thompson wrote:

 > Daniel, here, sees the X install as being "user-friendly".  Is the text
 > based install not? 

Let's not get too fixated on the visual aspect of installing the OS:
That's just a sideline (an important one, but a sideline nonetheless).
There's a whole swag of structural details which the current sysinstall
fails miserably in.

For example, has anyone noticed how virtually every OS on the market
except the *BSD's build up their distributions in the same file format
and with the same package database machinations as their third-party
add-on packages?  If I'm on a Solaris box, or an IRIX box, or a SCO 
box, or a Redhat box, or essentially anything else except BSD I install
the base operating system using the same tools I'd use for any other
software.

This provides enormous benefits.  Worried about bloat?  Define what
you mean by "Base system install" at the actual time that you're installing
the system.  Don't need a nameserver?  Don't install it.  Don't need 
lpd?  Don't install it.  Do you need Fortran?  Fine, install it, even 
though it isn't part of the default installation set (ooooh, I'm gonna
get flamed for that :-)

Upgrades are another issue:  At the moment, patching parts of the base
system is utterly hopeless.  Consider what happens whenever there's a 
security advisory:  We release a source-code patch to CERT, and say to
everyone, "Install the patch if you have the sources installed, but
if you don't have the sources you're going to have to upgrade the entire
god-damned operating system!"  And once someone has upgraded by patching
the source code, they suddenly have a "base distribution" which is 
subtly different from what would have been described as the "base 
distribution" the day before they patched it, so future bug reports become
a shot-in-the-dark type of problem.

Wouldn't it be easier to say, "pkgpatch named-8.8.2p2857" (or something -
I've pulled that example out of my butt) and have it md5 the files it's
about to replace to make sure that you have the faulty version it's
attempting to upgrade, back up the old files, install a new binary, and
patch the sources if they happen to be installed, AND RECORD THE FACT
THAT THIS HAS BEEN DONE IN THE PACKAGE DATABASE?  And if you don't like
the patch?  Back it out.

This is something other OS's find trivial:  To continue the example of
patching named, every other UNIX I can think of has named in a stand-alone
package as part of the base install.  If you want to upgrade it, you
install a more recent version of the package, and the fact that you've
done it gets recorded in a "this has been patched" section of the package
database.

Note that I haven't mentioned user-interfaces once in the discussion above:
The problems with sysinstall have very little to do with user interfaces.

 > To take this a step further, why not keep (or keep something similar to)
 > the current sysinstall, but have an option to fetch, install, configure
 > and run X and another GUI installer distribution, then start the X server
 > and continue the installation process from there? 

This discussion is orthogonal to the one we're actually happening, which
is about the structural problems in sysinstall which has lead Jordan to
place the "this has a limited lifespan" comment in the sources.  You can
do what you're proposing whether we end up with a new installer or not.

Anyway, don't think about user-interfaces -- They're the easy bit.  Re-read
Jordan's (very lucid) message on the topic from a few hours ago and think
about the problems described therein and the solutions that have been 
proposed;  you can slot your favourite user interface (even one that
looks the same as the one we're using now!) into that picture later once
the background issues have been dealt with.

   - mark

-- 
Mark Newton                               Email:  newton@internode.com.au (W)
Network Engineer                          Email:  newton@atdot.dotat.org  (H)
Internode Systems Pty Ltd                 Desk:   +61-8-82232999
"Network Man" - Anagram of "Mark Newton"  Mobile: +61-416-202-223


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19991215105736.A467>