On Thu, 6 Jun 2002 @ 12:55am (-0400), gabriel rosenkoetter wrote:

gr> On Wed, Jun 05, 2002 at 01:18:16PM -0400, Brian A. Seklecki wrote:
gr> > It can't be _that_ bad if it's the default action >:}
gr>
gr> Sure it *can*, just so happens people get away with it. But if you
gr> manage to get unlucky on the timing, it could cause catastrophe,
gr> especially if your machine is actively doing some production job
gr>

production & -current ?

gr> using the binaries you're updating as you're updating them. Why risk
gr>

...maybe pre-1.6, but the new system seems to circumvent this
freebsd-style insanity by building it's own set of local tools based on
the existing ones.

gr> it? You're going to be rebooting because of the kernel anyway.
gr>
gr> Plus, does a make build overwrite your existing /etc? I've never had
gr> the balls to find out...
gr>

...in my case it did not.  had to run usr/sbin/etcupdate and
usr/src/etc/postinstall.

gr>
gr> (Incidentally, these days I only reboot once; that is, install the
gr> new kernel, immediately boot -s, figure if it got that far not
gr> enough has changed since my last snapshot to matter, and unpack the
gr> distfiles. Also not the safest thing in the world, but it does mean
gr> only the single reboot and it's a *little* safer than installing
gr> over your live system.)

...per Julio Merino's suggestion, it seems that the best route is to build
once with a DESTDIR, verify your source tree builds completely, then build
again w/o DESTDIR but w/ the -u flag, as to not clear objects, but install
the compiled binaries.

...then etcupdate

...then usr/src/etc/postinstall

-lava