>However, IMO, edlabel should stay simple ... all of the "special" cruft 
>that's currently in disklabel(8) should go into the appropriate 
>machine-dependent portions of the kernel, and the disklabel ioctl suite 
>re-thought.

re-thinking the ioctl disklabels sounds cool. But  relocating the
port-specific cruft from disklabel(8) into machine-dependent kernel
code makes it effectively impossible to "port" disklabel to label
a disks for one port from another machine architecture.

I'd actually like to see one (non-distribution) binary version
of disklabel that's capable of labelling a disk and installing
bootblocks for as many ports as practicable:  i386, sparc, alpha,
pmax, ...  

The rationale for this is to support installation by building a
ready-to-go disk for an architecture X machine on an architecture
Y machine.  Jason's suggestion sounds cool, except for apparently
ruling out this installation route.