Re: Re: Some fun with tcp_wrappers

[wietse () porcupine org (Wietse Venema)]

Tomas Hoger:
> On Wed, 15 Apr 2009 14:53:22 -0400 (EDT) wietse () porcupine org (Wietse
> Venema) wrote:
>
> > Wietse Venema:
> > > > "test-hostsctl servicename unknown IP unknown" is what some
> > > > applications do expecting tcp_wrappers to resolve IP to hostname.
> > >
> > > I think that it would be a mistake to change a documented API that
> >
> > On the other hand, if you could add a new function under a new name
> > that does have the expected behavior, then there would be no
> > confusion, no risk of cross-platform applications breaking, and I
> > would withdraw my objection.
>
> That does not sound like a viable alternative and is likely to damage
> portability lot more, let me explain:
>
> - Application upstreams will not (should not) use any API that is
>   vendor-specific extension and not included upstream.  Even if there
>   is some new upstream version, it might take years to get into wide
>   enough use to applications to use new API, and result in
>   incompatibility with old systems.
>
> - Applications change would be required.  If that is done, there's
>   little reason to change to new API instead of existing hosts_access.

Linux-specific modifications to a 20-year old API break cross-platform
software. May I remind you that not all the world runs Linux.

On the other hand, no cross-platform software will break if one
adds a Linux-specific extension to the API.

I haven't yet pointed out the problems with automatic hostname
lookups from hosts_ctl(). One problem is that it breaks programs
that must not do hostname lookups such as portmappers.

Programs like the portmapper should not do hostname lookups, because
that would result in infinite recursion when host lookups involve
SUNRPC services such as NIS.

Again, not all the world is Linux. Making platform-specific changes
to a 20-year old API breaks cross-platform software in unexpected ways.

        Wietse