kre@munnari.OZ.AU (Robert Elz) writes:

>But attempting to infer that this should be done, merely because of some
>change made to an interface (disabling it: "ifconfig down", deleting an
>address from it, or removing the hardware completely) is simply wrong.

That's where I can't agree. We already infer that some other things
should be done, mereley because of some change made to an interface.

That's bad from an API point of view, because of the mixup, but good
from user interface point of view. When I rip of an interface I expect
things to break that attach to that interface. The single 'uncertainty'
might be how we declare a change to be permanent vs temporary.

For example, a program that has a handle on my USB serial adapter
will receive an error condition when I plug off the cable. I would
also expect the same from "shutting down" the USB port which is
the logical expression of plugging off the cable.

N.B. I'm not sure if an 'ifconfig down' actually has to terminate
connections. We were talking about the case where the socket address
becomes invalid. That's not an arbitrary 'ifconfig down' or address
change but a system-wide condition saying that there is no interface
that could possibly send from the previous address.

-- 
-- 
                                Michael van Elst
Internet: mlelstv@serpens.de
                                "A potential Snark may lurk in every tree."