Re: Message Numbers and Disconnect Codes

[nisse%lysator.liu.se@localhost (Niels Möller)]

Henrick Hellström <henrick%streamsec.se@localhost> writes:

> The only reason for specifying a STANDARD way of adding locally
> defined codes and features to an implementation, is to make it
> possible for STANDARD implementations to inter operate even if they do
> not implement the same locally defined extensions.

I disagree. The primary reason for reserving a space for locally
defined names and numbers, is to make it possible to define your own
experimental names and numbers without getting into conflict with
future *standardized* features. This applies to protocol design in
general. There are no guarantess that your local stuff doesn't collide
with somebody else's, unless you somehow agree about it case by case.

In the ssh protocol, we have a way to define *names* in a way that
gives everybody his own namespace, which makes the situation a little
better (but also more complex) than for many other protocols. But we
don't do that for the numbers in the protocol, and it's not feasible
to do so either, imo.

If you feel strongly that reason codes or other numbers in the
protocol need separate namespaces for local extensions, then the right
way to do that is to replace the numbers with names, using the
extension mechanism we use for all other names. But it's too late for
that, imo, so I'll try to stay away from arguing about such a change
in detail.

Regards,
/Niels