In message <tkraeiplbs1.fsf@parity-error.sm.sony.co.jp>enami tsugutomo writes
>itojun@iijlab.net writes:
>
>> 	since we ship userland in binary form, we need to always
>> 	enable IPv6 for userland (otherwise, users would need to
>> 	recompile the whole userland when he turns on IPv6).

Yes. True. No argument there. Unfortunately, the state of IPv6
implementatino deployment __today__ is such that if we want to ship
code that works at sites with IPv4, plus "early" IPv6 services,
we also have to ship with IPv6 turned *off*.

I guess that means we need either two kernels, or two userlands: one
with IPv6 on, and one with it off.  Also, dont forget the small-footprint
devices which plausibly cannot afford the bloat of IPv6.


>btw, since now sys_init tries to open v6 sockets and exits if failed
>to open it, the binary won't work on v4 only kernel.


*Sigh*.  Just as much of the world is waking up to the problems
with IPv6, we're ramming it down everyone's throat?