On Mon, 4 Sep 2000, Manuel Bouyer wrote:

> On Mon, Sep 04, 2000 at 09:12:05AM +1100, Robert Elz wrote:
> >     Date:        Sun, 3 Sep 2000 17:44:34 +0200
> >     From:        Manuel Bouyer <bouyer@antioche.lip6.fr>
> >     Message-ID:  <20000903174434.A517@antioche.eu.org>
> > 
> >   | with 8 partitons, wd0a has minor 0, wd0h minor 7 and wd1a minor 8
> >   | (see the DISKUNIT, DISKPART and DISKMINOR macros). With 16 partitons,
> >   | miror 8 becomes wd0i and wd1a has minor 16.
> > 
> > Only if the mapping from minor(dev) to partition remains the simple form
> > that it currently takes (the low N bits).   It doesn't have to.  It would
> > not be totally unreasonable to have the partition number taken from the
> > low 3 bits of the minor dev, then skip (say) 10-12 bits for the device
> > number, then use the remaining 5-7 bits for the rest of the partition number.
> > That would allow between 256 and 1024 partitions/device, which is way
> > more than can possibly fit in a label anyway (or be named using the XdNx
> > naming scheme).
> 
> Sure, there are ways to handle this. But the consensus was that this was not
> the rigth way of doing it, and I think it still has compat problems (there
> will be, for ports which have already 16 partitions :)

> The result of the discussion was that the rigth thing was to go to a new,
> machine-independant sheme (we also have to change the disklabel to allow
> more than 16 paritions because with only 16 parts, we'll have the problem
> again soon).
> 

If I'm not mistaken, FreeBSD has a different partitioning scheme.  They
use "slices" 1-4 for the four partitions, and then "slices" 5 and above
for extended partitions.  Have you considered that as an option, or is
there some problems with that also ?  And I don't think they have any 8/16
partitions limit.

Rakhesh