Applications and updating from registries (was: Re: New version, draft-faltstrom-idnabis-tables-02.txt, available)

[John C Klensin]

--On Wednesday, 20 June, 2007 08:39 +0200 Patrik Fältström
<patrik at frobbit.se> wrote:

> On 19 jun 2007, at 23.13, Debbie Garside wrote:
> 
>> But if applications are designed to update their systems from
>> a   Registry of
>> ALWAYS codepoints for the particular script this surely
>> wouldn't   happen?
> 
> That is something that will not happen. People say the same
> whenever a standard is updated, but in real life some
> software/computers are never updated, and not everyone can be
> connected to "the registry". Because of this, 10 year cycles
> for updating software that is deployed on the Internet is not
> uncommon.

While I agree with Patrik, this is also irrelevant.  If one is
an application looking up an FQDN that contains IDN labels, then
the "always" and "maybe" (both "yes" and "no" if they are
separated) categories are equivalent.  The only things that are
rejected by such an application are the "never" category and any
codepoints that are unassigned in the version of Unicode
supported by that application or the libraries it calls.   Such
an application might use the different between "always" and
maybe to condition warnings to users or other actions, but those
are UI issues, not IDNA protocol ones.

The only systems that need to know the difference between
"always" and "maybe" are those for name registration, and they
are a rather different kettle of fish from what we normally
consider "applications".  Part of what goes into that kettle is
an economic incentive to be up-to-date about any scripts (or
other collections of characters) the might want to permit in
registrations, but, for them, unlike applications that look
things up, it doesn't make any difference at all what version of
Unicode they use and when they update to it as long as they
don't move backwards.

On the application side, this is also why things can never move
out of "Never": if applications reject labels containing "never"
characters as bogus, then, if a character is moved out of that
category and registered, whether the label will be resolvable
will depend on the particular software the user is running,
certainly leading to inconsistent behavior across the network
and possibly exhibiting inconsistent behavior on a single
machine as the user shifts applications.

For that reason, I agree with the suggest made by Ken and others
that classification into "always" or "never" must be permanent:
characters can move into those categories from "maybe" but can
never move out.  Obviously that suggests that, if there is any
uncertainty and no appropriate authority to take responsibility,
then we had better be extremely conservative about getting
things out of the "maybe" category/ies into one of the permanent
ones.  If a mistake is made, we will have to live with it...
probably forever or until we reboot the DNS and start over,
whichever comes first.

      john