David Vernet wrote: > On Tue, Apr 02, 2024 at 10:50:02PM -0400, Suresh Krishnan wrote: > > At the recently concluded IETF119 bpf WG meeting, I had asked a > > question to Dave about the Provisional registrations for BPF > > instruction conformance groups. Section 5.1 of draft-ietf-bpf-isa-01 > > talks about Provisional registrations, but does not elaborate further. > > Specifically, I think it would be good to cover the following cases > > > > * Do we allow conversions from Provisional to Permanent? If so how? > > Would you mind please pointing to examples of other RFCs we can look at to > see how this is typically specified? My assumption was that we would just > initiate a standards action or IESG review to change the state from Provisional > to Permanent (meaning, that it was sufficient to simply define the registration > policies for Permanent and Provisional), but it sounds like we need to be more > explicit in our language. It seems that > RFC8126 section 4.13 doesn't specify a standard method for converting > between states: > > 4.13. Provisional Registrations > > Some existing registries have policies that allow provisional > registration: see URI Schemes [RFC7595] and Email Header Fields > [RFC3864]. Registrations that are designated as provisional are > usually defined as being more readily created, changed, reassigned, > moved to another status, or removed entirely. URI Schemes, for > example, allow provisional registrations to be made with incomplete > information. > > Allowing provisional registration ensures that the primary goal of > maintaining a registry -- avoiding collisions between incompatible > semantics -- is achieved without the side effect of "endorsing" the > protocol mechanism the provisional value is used for. Provisional > registrations for codepoints that are ultimately standardized can be > promoted to permanent status. The criteria that are defined for > converting a provisional registration to permanent will likely be > more strict than those that allowed the provisional registration. > > If your registry does not have a practical limit on codepoints, > perhaps adding the option for provisional registrations might be > right for that registry as well. > > Hmm, and looking at RFC 7595 [0] section 7.3 Change Control as a possible > xample, it specifies the following: > > 7.3. Change Control > > Registrations can be updated in the registry by the same mechanism as > required for an initial registration. In cases where the original > definition of the scheme is contained in an IESG-approved document, > update of the specification also requires IESG approval. > > 'Provisional' registrations can be updated by the original registrant > or anyone designated by the original registrant. In addition, the > IESG can reassign responsibility for a 'provisional' registration > scheme or can request specific changes to a scheme registration. > This will enable changes to be made to schemes where the original > registrant is out of contact or unwilling or unable to make changes. > > Transition from 'provisional' to 'permanent' status can be requested > and approved in the same manner as a new 'permanent' registration. > Transition from 'permanent' to 'historical' status requires IESG > approval. Transition from 'provisional' to 'historical' can be > requested by anyone authorized to update the 'provisional' > registration. > > [0]: https://www.rfc-editor.org/rfc/rfc7595#page-9 > > Dave, what do you think? I guess we should add a paragraph(s) explaining the > processes for this state machine? The ISA document says Permanent requires "Standards action or IESG Review" (the latter is a typo, should say "IESG Approval" to match RFC 8126 section 4.10 terminology). So converting to Permanent from nothing, or converting to Permanent from Provisional is currently the same... Standards action or IESG Review. Yes I can copy language from RFC 7595 (which I was also the editor of) to make it explicit. > > * Do Provisional registrations timeout after a while if they are not > > made Permanent? > > Dave? I'm not sure if this has been discussed or what the norm would be. It's up to us, there examples that time out, and examples that don't. (URI schemes being an example that don't.) There is no requirement in RFC 8126 to have any discussion of timeout. The lack of such discussion at in the document at present means that there is currently no timeout, i.e., like provisional URI schemes. If we did want a timeout, we'd have to add language to say that. Documents that are labeled as "experimental" are supposed to discuss timeouts, but things that are "provisional" generally do not. My current recommendation is to not have a timeout, but I don't feel strongly either way. > > * How do we remove Provisional registrations? Are the codepoints freed up? > > Also not sure if this has been discussed or what the norm should be. Absent language saying otherwise, currently you can convert a Provisional registration to Historical via the process for Historical. In the ISA spec this is currently "Specification required". In the RFC 7595 example, this is instead stated otherwise: "Transition from 'provisional' to 'historical' can be requested by anyone authorized to update the 'provisional' registration." Here I would definitely recommend copying the RFC 7595 precedent, which makes more sense process wise. Dave
