Re: [xmpp] [Technical Errata Reported] RFC7622 (5769)

[Florian Schmaus <[e-mail]>]

On 09.07.19 15:14, Florian Zeitz wrote:
>>> The intend of 7622 (in particular Section 3.2.1) as I recall it was that
>>> A-labels would generally be valid in JIDs.
>>
>> The question here is: Do you mean A-label or "a string which is (also) a
>> valid A-label"?
>>
> Maybe you want to enlighten me what the distinction is you make here,
> and how it is useful?

Of course. As I wrote in a previous mail, I want to avoid PRECIS for the
domainpart to require anything more heavyweight besides what is most of
the time a codepoint-by-codepoint validation routine. Otherwise
implementations have to perform a semantic analysis of the given string
in order determine if they are dealing with a DNS Name consisting of
only A-labels or an IPv4/v6 address, or something which is potentially
forbidden. I think that adds unnecessary complexity and is overkill, as
validating XMPP implementations would perform that on every inbound JID
they encounter.

> I can only imagine that it is one of intent. Yet, I'm having a hard time
> imagining people creating labels that consist of 'xn--' followed by a
> valid punycode encoding of IDNA-valid codepoints, but insisting these
> are not an A-label.
> 
>>> However, software would be
>>> required to process (user) input containing A-labels so that any
>>> JID-slots would only contain U-labels on the wire.
>>
>> Exactly. This is an important insight: Any string which is a valid
>> A-label may be valid in the domainpart of a Jid. However, if this string
>> got derived from an U-label, then you did something wrong. As JID
>> domainparts constructed from DNS names which contain U-label(s) should
>> conserve the U-label presentation (and not use the ACE/A-label form).
>>
> It seems to me that you are implying that software should not convert
> something that looks like an A-label to a U-label either?
> I.e. each label of a domainpart always has to remain in the
> representation provided by the user?

Not at all. I am only talking about what is allowed on the wire. We are
also in the fortunate situation that domainparts do not require ASCII
compatible encoding (ACE), and users are used to enter the IDNA/U-label
representation of a DNS name in their XMPP (client) software. Whether or
not software checks if the entered string into its UI is an A-label, is
not part of what I am considering. Although if I would, then I'd
probably argue that users do usually not enter ACE / A-label into UIs.
How often have you entered xn--ber-goa.de into your browser? Of course,
it potentially can not hurt if the software tries to convert those to
U-labels. But that is not a concern of the wire protocol which RFC7622
describes, although a note about this would probably be helpful.

- Florian