svn commit: r230207 - in head/sys: netinet sys

[Gleb Smirnoff]

  Ben,

On Thu, Jan 19, 2012 at 03:12:27PM -0500, Ben Kaduk wrote:
B> > G> It's not a big deal for me, I have no problem with replacing them on
B> > G> FreeBSD. FreeBSD is not even officially supported platform for the
B> > G> product and I build/test on FreeBSD solely for the purpose of avoiding
B> > G> linuxisms and platform specific behaviour. Let's hope it won't break for
B> > G> somebody else :)
B> >
B> > I'm pretty sure it will not break anyone, except for people still living
B> > in Internet with classdful addressing. Do you know any?
B> 
B> Not directly as classful addressing per se, but the interpretation of
B> an address as possibly-classful is encoded in the voting algorithm for
B> the ubik server elections that make our AFS distributed filesystem
B> work.  We support having ubik servers on quite a few different OSes,
B> and we can only change the voting algorithm with an OpenAFS major
B> version number bump because of our interoperability guarantees.
B> I'm not willing to claim that removing classfull addressing will
B> preserve the voting algorithm without an in-depth review that I don't
B> have time to undertake.
B> 
B> When you removed ia_net{,mask} from struct in_ifaddr (with no
B> __FreeBSD_version bump!  And then merged it to 9.0 during the RC
B> stage!), I had to scramble to keep net/openafs compiling and with the
B> same functionality.  It turns out that there is another code path in
B> OpenAFS (used by e.g. Darwin) that instead uses the IN_CLASSA() family
B> of macros, so I could preserve compatibility.  But if you go and
B> remove those too, I am screwed.
B> 
B> Please don't just assume that no one is using classful addressing and
B> remove things ahead of a reasonable deprecation schedule.

I'm sorry for that :( I should have bumped __FreeBSD_version.

I don't plan to remove what currently left from classes: the macros,
and the autoguessing in the case if mask isn't supplied.

-- 
Totus tuus, Glebius.