Had a shower thought this morning. I am primarily interested in making sysinst
work well, while not breaking anything else.

Solving the problem of mapping BIOS drives to NetBSD devices is certainly
sufficient, but it is not necessary when using the self-booting installer.

It would be sufficient for sysinst's purposes if we could just ensure that
any disk it will be installing onto has valid MBR geometry information.
As long as that is correct, we don't really care about the other disks.

So when an install floppy boots, we could have it examine the fixed disks and
try to decide if the kernel+sysinst will intuit a usable geometry for each
one. We summarize the results and offer to "prepare" disks for installation.
Of course we'd have to print information about the disks like how big they
are, to help people deduce which disk is the one they planned to install on.

What I mean by "prepare" is writing a blank MBR with the BIOS geometry in it.
Later, when sysinst runs, it can put a lot more faith in the geometry returned
by fdisk(8). (Actually, I'd be happy if sysinst simply displayed the geometry
at some point but did not ask for explicit confirmation; experts can cancel
sysinst, shell out and run fdisk(8) manually. Why duplicate the functionality
in sysinst when it's really fdisk(8) that could use a better interface ?)

The rationale for this scheme is that on a joe-average user system, all disks
will either have a valid MBR on them already or they will be empty/trash.
Especially in the majority case where there is only one disk that does not
already have valid MBR data on it, it is pretty easy for the user to okay
the writing of a valid MBR.

Todd Whitesel
toddpw @ best.com