I'd just like to lament that for someone who professes to know a
bit about the partition/disklabel/install/boot process (see my home
page), it still is full of pitfalls.  I think it is because there
are no visual cues about what disklabel (et. al.) do.  In other
words, it requires that you set up and maintain a substantial mental
model of exactly what is happening to the disk, and where.  However,
I am open to other explanations (e.g., I am an idiot).

Case in point; I tried most recently to convert a system from
another OS to NetBSD.  I believe that I disklabeled, edited the
disklabel to leave a full cylinder instead of just one track before
the NetBSD stuff, then newfs'd.  Then I installed the boot blocks
(via installboot), installed a substantial number of programs, and
tried to boot.  Nothing worked; I then remembered that I needed to
change the "hard" partition table (via fdisk).  However, this
appeared to fry the disklabel.  I fiddled around and managed to
muck everything up to the point where I decided to start over.  At
work here was the kernel and its fake disklabels, the disklabel
itself installing to "somewhere", and the boot blocks being installed
to "somewhere".  In both cases the "somewhere" is relative to either
the (possibly kernel-fabricated) disklabel or the partition table,
and both the origin and the coordinates are not communicated in
terms the user can understand (if at all).

I'm not sure why I am writing this.  If you have had problems like
these installing NetBSD, I suppose you may consider yourself in
more company than before.  I guess I am also interested in comments
about having these programs provide more feedback about what they
are doing, and where (ideally in C:H:S or LBA addresses; if not,
perhaps origin, offset and extent).
--
"1935 will go down in history! For the first time a civilized nation has full
 gun registration! Our streets will be safer, our police more efficient, and
 the world will follow our lead in the future!" -- Adolf Hitler