On 1030735606 seconds since the Beginning of the UNIX epoch
Greywolf wrote:
>

>Those "pointless trivial examples" are exemplified if just ONE block goes
>bad under /rescue (the new situation) while it's much less likely that
>you're going to lose blocks under Every Single Binary - or even a good
>portion of them - in /bin or /sbin

I'm not convinced by the `one bad block in /rescue' argument.  If we
want to analyse the potential failure cases, then we should also consider
that:

	1.  a crunchgen'ed binary is substantially smaller and hence
	    a random bad block is less likely to affect it at all, and
	2.  NetBSD provides demand loading of text pages and so one
	    bad block in the crunchgen'ed binary may not actually
	    stop any particular program from working.

Also, it is [as has been pointed out on this thread] very possible
to set up other mechanisms to ensure recoverability in the case of
problems.  My favourite is copying the install kernel onto the root
partition somewhere as this provides a tested kernel+userland setup
which has the tools that you need to install/upgrade the system.

It is not terribly difficult to make your own MD-image kernel that
has all of your favourite tools in it, and have a specific kernel
for the machine.  In this case, you can probably get the required
disk space down to under 2 or 3 MB.

--
    Roland Dowdeswell                      http://www.Imrryr.ORG/~elric/