> Ah, it begins to make sense. You're sharing the disk between NetBSD
> and some other OS? If so, the same rule applies. Partition c still
> maps the whole drive. That way, NetBSD is able to get at things like
> the boot blocks, DOS FDISK table and the like. It also allows access
> to the disk label and lets NetBSD repair bad blocks.

In other words, c is saying "this is what you can access" and d is
saying "this is what is there" ?

> PS: It's usually a good idea to have any DOS partition represented by
> a UNIX partition too. That way, your UNIX system can mount the DOS
> stuff, giving it handy access to the DOS partition's filesystem.

Good point.

Just some comments:

In reality, I have an adaptec 1542b with its 1Gb limit plugged into a
4.5Gb quantum atlas 2. I originally stuck all the NetBSD together
above the limit thinking NetBSD doesn't bother with such limits, but
of course it does need to go through the bios to boot, so I was
wanting to just change the disktab to make a bit near the beginning
the root partition, and call the original root partition f on /var.
Then have whatever is left of the 1Gb for DOS.

I thought the 'must go through the bios' bit mean that
/usr/mdec/biosboot.sym came into it, so I still don't understand why
(on i386) a kc disk from src/distrib/i386/floppies/kc doesn't boot a
compressed kernel (tells me bad format) but blocks stuck onto a floppy
using installboot do. Also there's the bit about disklabeling floppies
which doesn't seem to work for me. The plan was: compile kernel for
new machine, create an inst and kc for it (all this is in -current),
boot up & install.

Anyway, thanks for the clarification,

 Patrick