>While what you're saying is true, and devices are in a tree structure
>most of the time, this is no longer true with multi-pathed devices.

It never has been. :-)  In that autoconf paper that I never seem to
get around to working on, multi-path devices are acknowledged as a
"hole".  The problem I had (and still have) is how one is to
identify "device X" uniquely in any sort of generic fashion, so
that searching "bus B" and finding "device of description D" can
match it against "existing device of same description found on other
path".

Chris