Re: suggested virtual package name: dns-server
On Thu, Jan 09, 2003 at 12:51:50PM -0500, Michael Poole wrote:
> It seems dodgy to have packages conflict solely because they provide
> service on a well-known port -- at least if the servers can be
> configured to use other ports.
I've done similar discussions with other developers, too, and I also
happen to agree with you, in contrast to most other people I've
spoken to anyway. Afaik most if not all SMTP servers conflict with
each other as per dependency, yet most of them can be configured to
use different ports to listen on, at least when used from inetd
(just to name another example).
> For example, Apache (say, configured
> as a reverse proxy) should not conflict with another web server on the
> same machine just because they both default to using port 80.
> Were you referring to conflicts beyond TCP/UDP port 53?
Yes and no. The default value for the port number is 53, but one can
not only have most of them on a different port, but also on different
addresses. Eg. the standard bind9 installs in a way to listen only
on the loopback interface by default (!) and should therefore conflict
with no other server that was made to serve the world ;-)
Ok, I guess that having such setups should/could be handled with
alternatives, but simply conflicting in random ways with each other
and/or installing and maybe even running without looking at possible
conflicts with other packages is an undesirable situation I'd like
to see resolved. Catering for all possible (weird?) setups is nigh
impossible and a way to break the dependency system because it will
become very difficult to decide if dependencies are met or not.
Btw, I've encountered similar situations with web servers, and with
RADIUS servers, too.