PRECIS Framework: Preparation, Enforcement, and Comparison of Internationalized Strings in Application Protocols
draft-ietf-precis-framework-23

After discussion, I'm moving this to a non-blocking comment.  I still think that the working group and the responsible AD did not handle this right, and strongly urge not repeating this mechanism in future.

The registry created in Section 9.1 is very odd indeed.  I guess IANA is expected to assume that the format of the registry will look like the non-normative table in the appendix, but Section 9.1 doesn't say what the format of the registry will be.  In the IANA review, it's clear that they're not sure what's going to happen here.

But the real oddity here is that the specification of the registry involves an *enormous* startup cost for the designated expert, *and* requires that the DE be appointed and start her work immediately.  Normally, IANA takes the required actions as soon as the document's approval is announced, but in this case they will have to wait for the DE to be appointed and to derive the entire content of the registry.

It seems to me that the right way to have handled this would have been for the working group to have engaged the appropriate experts and made the table Appendix A *be* the initial contents of the registry, rather than explicitly denouncing Appendix A and leaving it as a seemingly quite onerous startup task that will delay the IANA actions indefinitely.

Why was it done this way, and what is the plan to get the registry content specified in a reasonable time?  Should approval of the document wait for that content to be specified?  Or are we really expecting to approve the document with the content of the registry left open?

UPDATE AFTER DISCUSSION:
The response to my final questions says that IDNAbis did it this way, the intent is to use the same expert as for that registry, and it's not expected to be terribly burdensome.

It's good to hear that it won't be burdensome, but it still seems that the working group produced an incomplete document.  The right way to have handled this would have been to involve the appropriate experts near the end of the working group's process, and to get the table in Appendix A to be the initial registration data, already vetted by the expert.  That way, the instructions to IANA would be clear, and the IETF and the IESG would be reviewing the complete picture.  When the document is approved and IANA creates the registry, they will contact the authors to confirm that it's all correct, at which point the authors would ask the expert to check it again.  It's unlikely that there'd be any changes needed in the roughly four weeks between IETF last call and document approval, and given that the expert you intend to recruit is also our liaison to the Unicode Consortium, he could confirm that they are not working on any updates just now.

As it stands, what the working group seems to be saying is that they don't have the expertise to get this right, and can't get the right experts involved... which, in any other context, we would push back on very hard, indeed.

The following are all items I mentioned in my AD Eval, but we decided none was a show-stopper and all could be held until the end of Last Call. The only substantive point is on 4.1.5, and the world will not end if I end up in the rough on this point. (We don't expect a whole lot of feedback during Last Call; the document got a good deal of review by a lot of experts, but the topic is pretty esoteric for anyone else to have much of a strong opinion.)

Throughout: Change "Informational Note:" to "Note:". I don't see any of them for which it makes a difference.

3.1: I would move the first paragraph down further in the section. And I would delete the parenthetical at the end of "Contextual Rule Required"; no need to introduce undefined terms here.

3.2.4 and 3.3.4: The SHALLs in here seem weird to me. Above, you don't say that a string with a Disallowed character "SHALL be rejected". If it were me, I'd simply say:

   Any code points that are not yet designated in the Unicode character
   set are considered Unassigned for purposes of the XXXClass, and such
   code points are to be treated as Disallowed.

4.1:

Change "MUST register" to "are registered". MUSTs for registration seem silly. (If you want to say, "Implementations MUST NOT use unregistered classes", you could, but I don't think you want to do that.)

Change "It is RECOMMENDED for profile names to be of the form" to "The naming convention for profile names is to use the form".

4.1.5: I'm not thrilled with this section in general, but in particular I'm not sure what "mixed-direction strings are not supported" means. We do know how to process strings that contain characters with a mix of directionality. Such strings are sometimes a visual challenge, but not a processing challenge: The RFC 5893 Bidi Rule exists because IDNs want to be visually stable in either an LTR or RTL context, and because IDNs use "." as a label separator yet want to have text display consistent in a context that is unaware of labels. There are perfectly reasonable cases where none of these hold true, so I think this section could be softened so that it's not overtly pushing that protocols never allow mixed direction text or that they should always lean toward using RFC 5893. 

4.2: A bit of ABNF neatening:

OLD
      fullname = namepart [1*(1*SP namepart)]
      namepart = 1*(idpoint)
NEW
      fullname = namepart *(1*SP namepart)
      namepart = 1*idpoint

9.2: It might be nice to add section numbers (3.2 and 3.3) to the registrations for IdentifierClass and FreeformClass.

9.3: It might be nice to note the naming convention from 4.1 in the template.