In the last snapshot grabbed(970801), I find that rpc.bootparamd
was failing because in /usr/src/usr.sbin/bootparamd.c, haddr was
declared as

long    haddr;

instead of some sort of inet_addr_t which is the length of an ipv4
host address, 32 bits.  I changed it in this program to u_int32_t
and the bug was corrected, but I was surprised to find that
/usr/include/arpa/inet.h also uses longs instead of some sort of
UINT32.  (The bug was when the Alpha long was passed to gethostbyaddr,
the resolver returned an internal error, and the host could not be
looked up.)

Perhaps we need a systematic sweep for this type of problem in the
source code?  And the introduction of an inet_addr_t?