On 27-Nov-99 Manuel Bouyer wrote:
> Could be a good idea. But if this was the same program some code could
> be avoided (sysinst could call postinst directly with the internal data
> structure, no need to know how to write a file).
> Having 2 separate programs could allow us to same some space on the boot
> floppy, with sysinst pulling postinst from the install media (but this mean
> that it's sysinst job to mount the CD or NFS server, or to pull a file
> from ftp, and this may be annoying).
> 
> For now I would go with a single program, but with 2 really separate area.

Well.. one of my major goals was the ability to parse an incoming datafile and
just skip the whole asking questions phase of the install.  If a combined
sysinst-postinst could do that.. thats fine with me.. and could be separated
out later if desired.

> I'm not sure we should have the notion of offset here. The install tool
> should deal with this by itself.
> I can't see any reason why the user would want to hardwire the partition
> offsets.

You are probably right about the offset not being necc.  I don't remember why I
had that in there.

> Yes, log window ! This is important (we could use different ttyE with wscons,
> but not all ports support wscons yet, unfortunably).

Actually sysinst sorta-allready has code in it for that.  If you used a gui
like CDK (curses development kit) it has a log-window widget with scrollback
and everything.. very slick.

One real big advantage of allowing some sort of wildcarding or fuzzy matching
in the input file specs, would be the ability to supply generic configs for
systems.  Example:

We could ship a basic config.. that has wd/sd0 configured with root, usr, swap
and the rest of the disk 50/50 for /src and /pkg.  Sysinst would then ask for
the IP stuff, and the media gunk, and the system would just install.

An admin could configure a file that got an IP from dhcp or a list or
something, and could mass install netbootable sparcs off the network with one
file, hands-free.

The main reason I wanted sysinst/postinst separated would be to allow it to
work in two stages.. (eventually, not as a first step).. but imagine the
following:

fire up CD, load sysinst.. answer disk size/ip/media questions.  Present user
with a list of packages, pick the packages.  Present user with an rc.conf menu,
and configure.  call postinst.

Postinst would install, put postinst in /etc/rc, and reboot.  Machine boots in
stage two, postinst configures rc.conf, and starts installing packages.  It
finishes, removes itself from rc, and reboots.

Just some future ideas, for why I wanted to see it separated.  You are right,
it wouldn't need to be.. but it might be easier to do.  For example a more
slender postinst would ship on the media, and the full-blown one would be part
of base, and allow the packages, rc, etc..

---
Tim Rightnour <root@garbled.net>
NetBSD: Free multi-architecture OS http://www.netbsd.org/
NetBSD Mailing lists on the web: http://mail-index.netbsd.org/