Re: [idn] proposed i18n naming rules

["Eric A. Hall" <ehall@ehsco.com>]

Mark.Andrews@isc.org wrote:
> 
> >
> > Does this meet the consensus:
> >
> >
> > Internationalized domain names have the following attributes:
> >
> >   sequence of labels:
> >
> >       labels contain any valid UCS character code, regardless of
> >         whether or not a character representation is assinged to that
> >         character code
> >
> >       character codes are explicit, not required to be normalized in
> >         any way
> >
> >       minimum label length of 1 UCS character code
> >
> >       maximum label length of 63 UCS character codes
> >
> >   maximum FQDN length of 255 UCS character codes (including the
> >     label separators when the IDN is written out)
> >
> > Internationalized host identifiers are a subset of internationalized
> > domain names, and have the following attributes:
> >
> >   sequence of labels:
> >
> >       only those UCS characters from the "safe-set"
> >
> >          [version of] http://www.unicode.org/Public/UNIDATA/Scripts.txt
> >
> >          U+002D (HYPHEN-MINUS)
> >
> >       MUST be normalized through nameprep prior to processing
> >
> >       leading and trailing characters MUST NOT be HYPHEN-MINUS
> >
> >       minimum label length of 2 UCS characters
> >
> >       maximum label length of 63 UCS characters
> >
> >   maximum FQDN length of 255 UCS character codes (including the
> >     separator characters when the IHI is written out)
> >
> >   U+002E (FULL STOP) is ONLY valid for use as a separator when
> >     the IHI is written out
> >
> 
>         Does any one see that this *will just not work*.  This has
>         unnormalised sequences of labels in the query trying to be
>         matched against normalized labels in the resulting from
>         delegation process.

The syntax rules above say nothing about queries versus responses. The
clarification at the bottom should make it clear that systems which are
issuing lookups for the purpose of connecting to a host would be using the
hostname ruleset, which requires normalization.

Are you referring to lookups of IDNs (as opposed to IHIs) that would not
be normalized?

> > The suggestion would be that the IHI rules would apply to all encoding and
> > decoding systems which knew that an IDN specifically referenced a host.
> > This would mean applications that were issuing a lookup for the purpose of
> > establishing a connection, protocol operations that would use a domain
> > name for a host-specific function (such as message routing, HTTP HOST
> > headers, NS RRs in DNS where the NS explicitly referenced a host, and so
> > forth), and processes which decode structured data (such as email
> > addresses and URLs) for display.
> >
> > All other IDNs would fall under the IDN rules. This includes SRV entries
> > in the DNS, email addresses stored in SOA and RP RRs (which can contain
> > FULL STOP, among others), and so forth.
> 
>         Mailboxes after the first label are subject to the same contraints
>         as hostnames.  I would also argue that SRV ownernames are also
>         subject to the same constraints as hostnames once you remove the
>         first two labels.  In otherwords you can't assume that one set
>         of rules about what is legal applies to all labels in a domainname.

Yeah, good point.

-- 
Eric A. Hall                                        http://www.ehsco.com/
Internet Core Protocols          http://www.oreilly.com/catalog/coreprot/