I suspect that may be a cc problem.
I, too, have had that exact same problem, and tracked it down to if I
built /usr/src/linux with gcc 2.7.2.1, it would get really confused
about module symbols (to the point where McAfee's WebShield WOULD NOT
LOAD at a customer site in Norway when I was doing debugging there; I
had to telnet to my lab, at the end of a long and slow 64k line, edit
and build the modules there, then telnet them back, because 2.7.0
would happily generate the modules.) And for the life of me I don't
know if it's because of the compilers or because of the dratted
mutability of every single Linux application level tool [1].
>Heck, I'd settle for a *sane* /usr/include/netinet and /usr/include/net
>on Linux without the mbuf's.
Simple solution to that: Port BSD libc over to Linux. I'd cheerfully
shelf my plan to do just that and use your port [2]. But this isn't
really a kernel issue, but a userland problem, and it's probably not
germaine to this mailing list.
____
david parsons \bi/ Mr. conservative.
\/
[1 the solution to this, of course, is to NEVER upgrade. My standard
distribution is already about as distant from the Linux distribution
baseline as it is from FreeBSD, but it works and I only need to spend
about 10 minutes a week on administration chores. At least newer
kernels still work on most of my hardware -- I don't know what I'd
do if a kernel came out that wouldn't work properly with libc's 4.8.0
and 5.0.9.]
[2 No compatability with baseline distributions, but see note 1]