Date:        Mon, 16 Jun 2003 09:15:05 +0100
    From:        David Laight <david@l8s.co.uk>
    Message-ID:  <20030616091505.G3322@snowdrop.l8s.co.uk>

  | Yes, but is is dangerous to be using space that isn't alocated in the mbr.

Yes, of course - but what matters is what is eventually in the MBR,
when the whole process is finished, not what is there at the instant
that the user says "allocate this much space".

  | IMHO sysinst should install 'sane' systems where filesystems are either
  | part of a netbsd partition or partition in their own right.

Yes, that is more or less what I was trying to say.

  | You can't do that, there are restriction on what the extended partition
  | can describe.  You have to create the mbr extended partition and then
  | transfer the info to the netbsd disklable.

No, you don't - you can do it in either order, once it is all written
no-one can tell which block got put there first.   Obviously things need
to conform to the various constraints imposed (by MBR rules on i386,
cylinder boundaries on suns, etc) but that's different from "create
the MBR first".

  | Indeed, also sysinst doesn't need to be able to create all of the setups
  | that will work with netbsd.  Just ones that will work given most start
  | positions.

Agreed.

  | For i386, the Linux 'one filesystem one partition' scheme isn't stupid!

It isn't, if one assumes that the MBR should rule.   But BSD (and
unix in general before that) has been using labels (though they used to be
in the driver, rather than on the drive) to control the world since before
MBRs were invented.   In the true unix world, the label rules.

What's more, to retain the tradition of minimising (hiding) the differences
between ports as much as possible, that is the way it should stay.  Only
a few ports use MBR labeling, all of them use the disklabel.

kre