Phil Nelson <phil@steelhead.cs.wwu.edu> writes:

Just one more point here...


>This fits in very nicely with my original philosophy for sysinst ...
>
>"If a program exists that does a job, just run that program, don't
>put the functionality in sysinst."
>
>So if the pkg_* tools could easily unload the distributions sets,
>sysinst would need only to call pkg_add to do the job.  It does look
>like that pkg_add would need a new flag to provide the root directory
>of the install tree for sysinst's use.  The current "-p prefix"
>changes the idea of where to install in the full tree and overrides
>the precomputed prefix.  sysinst would want to keep the standard
>/usr/pkg prefix (or whatever the prefix was) but put that relative to
>the /mnt directory.  [..snip..]

Sure.  Call the new flag -R ('root') (or `-P' for `prefix'?).
It provides a prefix which is prepended to the existing precomputed
prefix, like doing a `chroot'.

That could be used outside of sysinst, too. If you have a diskful
server for diskless clients (an i386 server for Sharks) you can
install pkgs on the i386, into the /usr/pkg directory shared by the
Sharks, using -R /path/to/diskless/root, and all the internal paths in
the installed pkg would Just Work on the clients.

Plus, if you cut up the `base' set into pkg(s) that includes /sbin,
/bin, and /etc (or at least a minimal one) and no more, say,
<rootpkg>, then doing a

     pkg_add  -R /path/to/diskless/root  <rootpkg>

would meet Charles' request for setting up `sandboxes' for
network-boot installs.  modify as necessary for getting <rootpkg> out
of the pkgset it comes in., Or simply unpack all of `base' into the
diskles-root, if you want to demo NetBSD to people at a conference/
meeting via netbooting their own laptops.

(Jim: could you note that down somehere safe, please?)