On Sun, 16 Jun 2002, Bob Nestor wrote:

> On Sunday, June 16, 2002, at 08:10 AM, Hauke Fath wrote:
> > At 21:04 Uhr -0500 15.6.2002, Bob Nestor wrote:
> > Well, I assure you I did not play any dirty tricks to get /var on
> > 'd'.  ;)
>
> No, but your comments got me thinking last nite.  I think the approach
> that worked for me on 1.3 doesn't work right now and I found some
> additional problems this morning. I've been able to reproduce the
> problem Frederick mentioned yesterday, so I'm working on a new approach.

What do you think of this idea...? the partitioning code mmap()'s a
small file (or sparse file) containing an actual Apple Partition Map,
and manipulates that, then invokes VNDIOCSET() to configure the file
as a disk, to get the kernel's interpretation. After, the label (map)
could be saved to a file, or copied to the target disk. It sounds more
complicated, but it would be foolproof, it would let you save a label
(map) to a file, and it would be robust against proposed changes in
the kernel disklabel code (to support 16 partitions, LFS and RAID and
so on).

By the way, another problem I recall from earlier experiments:
changing only the mount points still leads to sysinstall trying to
write a new map. If you manage to get your partition map the way you
want it some other way, sysinstall should leave it alone and go on.

Frederick