> >It seems to me that it's a lot easier to fix the daemons to behave
> >reasonably on inserts and removals than it is to teach the kernel to
> >behave reasonably if inserts and removals don't detach interfaces.
> 
> I'd agree, except Linux seems to manage fine, and they do throw the
> state away.  (Maybe its becuase they bind the device name, in the
> userlevel daemon, via the tuple space and/or MAC address?)

We're in violent agreement, Jonathan.  Itojun was saying that the
routing daemons currently behave badly if card removal causes the card
configuration to disappear, and I'm suggesting that although I believe
this is a problem, I suspect that it wouldn't be all that hard to
fix.   Using the BSDI-style system where cards that don't exist are
listed by ifconfig and state for those cards is allocated at initial
device probe time causes problems that I think are more serious for
some uses, such as mosquitonet, and it would be hard to fix these
problems without making card insertion and removal events cause the
interface to appear and then go away.

			       _MelloN_