It occurred to me that Rick Kelly wrote in gmane.os.netbsd.current:
> Pkgsrc is limited. Install NetBSD 2.0 and use the pkgsrc for NetBSD 2.0, or
> live with getting your /usr/pkg trashed by a newer pkgsrc trying to upgrade
> everything that you already have installed.

Pkgsrc, as fine as it is, is completely inadequate for having everything as a
package. If we're going to make libc a package, for example, how on earth
are we going to deal with dependencies and updating? The Gentoo way of
doing this seems to be "Ah, just overwrite it, user will reinstall the
system if we break things horribly now".

The second issue is the one of support. We currently have specific, but
well defined, versions that we can support. There's only ONE NetBSD-1.6.2, 
and we support that version. There's ONE NetBSD 2.0, and we support that
too. If you make everything package based, every box gets to be unique
in it's own right, since no two boxes will have the same version of 
everything, making support virtually impossible (Again, look at gentoo,
whose support is abysmal and seems to be "Oh, did you update everything?".
This may not be possible for several reasons, not in the last place because
I really don't feel like compiling more than I really should on my next68k
machine. And with everything as a package, there won't be cross-compiling
either)

I've been tracking NetBSD-current for a few years now. I never ever broke
my machine to the point that a complete new reinstall was the only feasable
way to get started again. With Gentoo, the expected lifetime of an install
is just under three months.

I really don't see the *point* in discarding the current distribution scheme,
just like I really don't see the *point* in having no MTA in base. Every
NetBSD box needs to have *some* kind of local delivery agent at the very
least - otherwise cron (including the weekly/monthly/daily security stuff)
will break. So shall we remove cron too, then? 

There's nothing to be gained, and a lot to be lost.