Re: [exim] building exim on OSX against *external* bind9 (*n…

hi greg,

>> well, for a variety of reasons. suffice it to say that other parts of my
>> project rdirectly equire some features of bind9, and it's best for me to
>> have everything building against the same sets of libs (db, bind, etc.).
>
> Note that BIND-9's named is the only thing with unique features here.

well, there *is* llwres, which i believe is unique to bind9 ...

> The resolver library is an entirely different, and completely separate,
> component. The protocol between the resolver library and named is DNS,
> pure and simple. You can run BIND-9's named and still use pretty much
> any other resolver library with other DNS-using applications.

ah, that's your point. sorry. agreed.

> Note also that on OS-X there are other significant reasons why you might
> want to avoid even using any third-party resolver, not the least of
> which is that it does some special things for certain local services.

fair enuf. tho' i'm not _replacing_ the local resolver, and leaving OSX to "do
its own thing with its own stuff" ...

> Of course a mailer that'll handle internet mail to and from the public
> network could be linked with any compatible and sufficiently bug-free
> resolver library.

in principle, yes. that's the hope here for exim. the postfix crwod was a
mite less polite/accommodating abt the whole business ... =)

> If your goal is to use the most ccomplete, bug-free, implementation of a
> compatible resolver library then the BIND-8 resolver library that's
> included in the BIND-9 distribution, is indeed a good choice (since it's
> probably the best maintained version of such code), but it may not be
> any better than the "standard" one that comes with your system.

not looking for 'better' as much as one source, under my control, separate from
the system.

> Indeed the Mac OS-X resolver code may even be derrived from BIND-9,
> though I don't know for certain. Apple have been very good at keeping
> up-to-date with the current releases of all the sources they use.

well, at least better now thatn they used to ...

> On the other hand, for example, the resolver library still used in
> NetBSD was derived from BIND-4. However it's been maintained fairly
> well and is more or less as good at doing DNS queries as the code from
> BIND-8, or the BIND-8 code from BIND-9. One can link applications on
> NetBSD with other resolver libraries, but using the one supplied in libc
> has advantages since there are system specific changes that integrate
> better with other aspects of the rest of NetBSD. This is similar to,
> but not exactly the same as, the way OS-X has its own system-specific
> features added to its resolver implementation.
>
> A version of an SMTP MTA such as Exim may behave slightly differently if
> linked against different resolver libraries, especially for "local"
> hostnames. How differently depends entirely on your local system
> environment. Perhaps there wouldn't be any perceivable difference.
>
>
>> i think i've just learned something. the "standard UNIX resolver library"?
>
> I think he meant "the resolver library that comes standard with your
> system distribution".
>
> It's also "standard" because it has the same API as most unix-based
> software expects to use in order to interface to the DNS. See the
> resolver(3) manual page and related manual pages for part of the
> definition of this API.

ok.

> The key thing here is that most DNS-using applications, Exim included,
> require the use of the resolver(3) API and they simply will not compile
> and link with BIND-9's new lwres library, and adapting them to use lwres
> is a _far_ more difficult task than you seem to think.

i've not targeted using llwres ... my intention, as per my earlier email, has
been simply to link against my external bind9's libbind.

>> do you mean that that "std lib" is *not* "from" bind, i.e. ? now that you
>> mention it, that wouldn't make any licensing sense, would it?
>
> Most system DNS resolver libraries, especially on unix-like systems, are
> indeed derived from the code supplied with BIND. That code is freely
> reusable by all system vendors.

thx for your comments!

i still have the goal of getting exim to link against the _my_ libbind. as per
some of exim's OS/Makefile, there's references *to* using libbind, but i gather
its on a system where it's the native resolver.

the challenge, at the moment, is exim + bind9's libbind + OSX & its
BIND_8_COMPAT. i haven;t understood anything as yet to indicate that it
*can't* be done ...

cheers,

richard