I supped and compiled a kernel on the ol' hp340 yesterday, and if_le.c
   couldn't compile because of a bogus check involving a tailq in leinit(). 
   I removed that piece after looking at sys/dev/isa/if_le.c, but i'm both
   not convinced that it was the Right Thing to do and not convinced that my
   machine won't die with some obscure panic someday. 

Ethernet devices *always* have at least one address in their address
lists -- the link-level (hardware) address.  So the test was, since at
least Net/2, superfluous at best.  It's conceivable that the interface
should refuse to turn on the IFF_UP bit if there are no protocol
addresses defined, but if so, this should be handled elsewhere in some
generic fashion, and should report an error to the user.