new interfaces code (bug fix + questions)

[Andrej Borsenkow]

[[ I copy it to Harlan Stenn as well ]]
>
> 1) I try to auto-detect whether the size is returned at the start of
>    the buffer so we don't need an ifdef. Let me know if you think what
>    I've done is valid.
>

Harlan, what do you think of it? The code tries to autodect
SIZE_RETURNED_IN_BUFFER that is hardcoded in ntpd and relevant part looks like

/* we can ignore the possible sizeof(int) here as the resulting
   number of interface structures won't change */
n = strioctl.ic_len / sizeof(struct ifreq);

/* we will assume that the kernel returns the length as an int
   at the start of the buffer if the offered size is a
   multiple of the structure size plus an int */
   if (n*sizeof(struct ifreq) + sizeof(int) == strioctl.ic_len) {
      ifr = (struct ifreq *)(buff + sizeof(int));
   } else {
      ifr = (struct ifreq *)buff;

It looks fine for me ... but I do not have access to any system that is using
(or needs) STREAMS TLI.

May be, if it works, you could use it for ntpd as well?

>
> The new code is at ftp://samba.org/pub/tridge/misc/interfaces.c
>
> > How do you intend to choose the method? I think, that IFCONF must be
> > selected if at all possible.
>
> In autoconf I'll try HAVE_IFACE_AIX first, then HAVE_IFACE_IFCONF then
> HAVE_IFACE_IFREQ. We need to try AIX first as IFCONF works on AIX but
> only gives some of the interfaces.
>

Again, hard to tell. Immediate con is the case of compiling on a system without
network :-) Is probably rare. The worst case is a system, that supports both,
but returns less intefaces with IFCONF (sockets) than with IFREQ (STREAMS TLI).

/andrej