>> Here's a suggestion to avoid these problems: Change disklabel's idea
>> of the `source of truth' to be the MBR, not the operator.

>I disagree, it should do what I want.
>1.2 install is almost perfect for me.

Lucky for *you*.

The i386 1.2 and 1.2.1 installation is unacceptable to me.  It
has wiped out every single hard disk I have installed it on.
The bugs (and installation problems)  here are well known
Besides, having to use a pencil and paper is mororic.



> I can see only possible improvement:
>give operator default start and length of NetBSD portion of disk, if
>there is NetBSD fdisk partition already defined (but don't force
>the operator to use them! ).

You seem to not be listening.

If I am installing NetBSD onto a machine that already has Linux, and I
need to keep the Linux partition, then it's vital to me that NetBSD
obey what the MBR says.

If I do that from Linux, and set up an empty MBR slice for NetBSD, the
1.2 installation *trashes the disk*.

[snip]

>I hate computer lying to me.

You have not understood my suggestion. I am not suggesting anywhere
that we make the computer `lie to you'.  I am suggesting we change the
primary way *you* tell NetBSD where your i386 NetBSD partition is on
the hard disk to be the MBR.

I think the reasons for that are obvious: the hardware often requires
that we lie to the BIOS or OS about the disk geometry already, and the
currently-recommended method is too error-prone even if the bugs are
fixed.