Re: [radext] WGLC #1 for draft-ietf-radext-nai-03

[Sam Hartman <[e-mail]>]

>>>>> "Alan" == Alan DeKok <[e-mail]> writes:

    Alan>   Bernard and I had a long talk today about the NAI document.
    Alan> There are some clear points, but there is still one point
    Alan> which is unclear.

    Alan>   It's not clear if the process of normalizing a realm name is
    Alan> independent of Unicode version.  If it is, then proxies can
    Alan> happily normalize what they get, and we don't care what
    Alan> endpoints produce.

    Alan>   If it's dependent on Unicode version, then the only system
    Alan> capable of normalizing the NAI is the endpoint which produces
    Alan> it.  And they don't normalize NAIs now.

I think this thinking is bogus.
We're trying to create interoperability.
If we improve the fraction of the time things work, we've made forward
progress even if there are corner cases where it doesn't work.

Endpoints MUST (BUT WE KNOW yOU WON't) normalize because there are some
cases where it's unicode version dependent and a proxy with an older
version has no hope of succeeding.

However, in practice, proxies can do a lot, and so they should.
First, the proxy may not have things stored in Unicode.
Or it may have its routing tables stored in a database that has its own
approach to I18N (say LDAP or some SQL database).
In practice, people will sometimes let that database do the
normalization, and if it increases the chance that you'll route to the
realm then you should encourage that.

It seems bogus to me to assume you need a perfect solution, to assume
normalization can happen in one place or to ignore the long picture as
we migrate things 10-15 years from now.

--Sam