[arch-releng] aif -p automatic (WAS:the future of quickinst)

[Dieter Plaetinck]

On Tue, 10 Mar 2009 09:32:54 +0100
Dieter Plaetinck <dieter at plaetinck.be> wrote:

> On Mon, 9 Mar 2009 20:01:35 -0500
> Dan McGee <dpmcgee at gmail.com> wrote:
> 
> > On Mon, Mar 9, 2009 at 5:29 PM, Thomas Bächler
> > <thomas at archlinux.org> wrote:
> > > Aaron Griffin schrieb:
> > >>
> > >> On Mon, Mar 9, 2009 at 2:55 PM, Dieter Plaetinck
> > >> <dieter at plaetinck.be> wrote:
> > >>>
> > >>> * aif -p automatic -c /path/to/configfile:
> > >>> the recommended way: you have
> > >>> the ability to specify partitioning/filesystems/package
> > >>> lists/hostname etc in a config file format which i'll try to
> > >>> keep as simple as possible.
> > >>> that, and the ability to define custom hooks and whatnot to
> > >>> implement custom logic in your config
> > >>
> > >> This seems the most ideal here. If we're assuming you're
> > >> mass-installing or something of the sort, the installer needs to
> > >> know all this information for each machine.
> > >>
> > >> I think we should even skip the initial "quickinst" copy and just
> > >> go this route - though providing a sample config file would be
> > >> necessary though
> > >>
> > >
> > > I agree.
> > >
> > > Someone could easily create a custom installation medium with a
> > > custom aif automatic configuration and use that for mass
> > > installation.
> > 
> > My only thought is whether this is even necessary- I'm guessing most
> > people used quickinst as just a part of a full automated install
> > (e.g. shell script that first does partitioning, mounting, etc).
> > 
> > I wouldn't mind a tool that takes care of this for me, but if it
> > isn't useful for a wide variety of cases, it won't be used. In
> > addition, any config file format we write is surely not as flexible
> > as a shell script would be.
> > 
> > -Dan
> 
> Aha ! This is actually the very reason I started this all.
> I want for myself to have a script that does all this cool stuff.
> (the port of /arch/setup is just a little side project as far as I'm
> concerned ;-)
> 
> And you're right that a file format adds another layer of complexity,
> but you get some free stuff in return such as the rolling back in case
> of failure, no need to program sfdisk input (which is quite ugly).
> If no one will use this particular part of aif.. well that's fine by
> me. we'll see :)
> 
> Dieter


Hey dudes.
I got rid of the quickinst stuff and implemented the automatic
procedure.
In my tests, it produces working systems.
One thing to note: when the process is done, and you type reboot,
you'll get squashfs and i/o errors on the live /.  didn't figure that
one out yet.

You can install it by installing aif-experimental-git package from
AUR.
Invocation:
aif -p automatic -d -c /usr/share/aif/examples/generic-install-on-sda

To get an immediate idea of the code and sample config (which are
actually very simple), see:
http://projects.archlinux.org/?p=aif.git;a=blob;f=src/core/procedures/automatic;h=b0e327aa3a5b453a0da8598f5295677da416a862;hb=experimental
http://projects.archlinux.org/?p=aif.git;a=blob;f=examples/generic-install-on-sda;h=9d257763bfa4ff5b39a25d3598ea4296ef8f85a2;hb=experimental

Have fun!

Dieter