Session Date/Time: 14 May 2025 14:00

CBOR

Summary

The meeting primarily focused on the cbor-packed document, discussing the optimal values for tunables A, B, and C based on new experimental data. A proposal was made to register CBOR simple values for general packing and templating use, rather than exclusively for cbor-packed, leading to a sense of the room supporting an A=16 value for shared simple values. Work on this separate registration document was deferred. The meeting concluded with the cancellation of the next scheduled interim and a call for the working group to review an alternative draft for cbor-preferred-serialization (CDE) to foster broader consensus.

Key Discussion Points

cbor-packed Tunables (A, B, C):
- Vadim presented an update to sopac-15 reflecting tunables A (1+0 simple values for shared item references), B (1+1 tags for straight argument references), and C (1+1 tags for inverted argument references).
- New data suggests B and C should likely be equal, moving towards two primary numbers to determine (A and B/C).
- The argument against setting A=B=C=0 was reiterated; cbor-packed aims for reasonable conciseness without excessive complexity, unlike some "binary JSON" formats.
Experimental Data for Tunables:
- Zip Distribution: Indicated that B=8 provides a good balance between conciseness and complexity, costing about half a byte more per reference than the original B=32.
- SBOM File Analysis: A large SBOM file with structural redundancy showed that the first 8 argument table elements were used appreciably more often, supporting B=8 for such structured data.
- IoT DNS Messages (for Tunable A): An analysis of DNS responses from an IoT testbed suggested that while many messages are simple, a value of A=12 (out of 20 available 1+0 simple values, after reserving for indefinite length encoding and false/true/null/undefined) appears reasonable based on current data, offering benefits to over 95% of messages in the corpus. The original cbor-packed design considered A=16.
- Comparison to Zstandard confirmed that cbor-packed serves a different purpose than general-purpose compression.
- More corpus data from the community was requested to refine these numbers further.
Proposal for Shared Simple Value Registration:
- Vadim proposed that CBOR simple values (e.g., all 16 available 1+0 values) should be registered in IANA for general packing and templating purposes, rather than being exclusively assigned to cbor-packed. This would allow other specifications (e.g., cbor-seat) to use these values without being dependent on cbor-packed.
- Carsten noted that cbor-packed is designed with an extension point, allowing new table setup mechanisms to define their own meaning for these simple values and argument reference tags within their own context (e.g., via a specific tag). Editorial clarification in the cbor-packed document might be beneficial.
- The discussion revolved around whether explicit separate registration or clearer wording in cbor-packed was the appropriate mechanism for this flexibility.
DNS Draft Review:
- Vadim briefly mentioned issues with examples in the current DNS cbor-packed draft, specifically regarding tag combinations and optimized compression, suggesting further explanation or fixing is needed.
cbor-preferred-serialization (CDE) Alternative Draft:
- Lawrence presented his work on an alternative draft for cbor-preferred-serialization, aiming for a different structural approach that is shorter, clearer, and introduces fewer new ideas than the current CDE draft.
- He intends to publish this draft and called for the working group to review it, focusing on its overall structure and orientation, to foster stronger positive consensus on this important document.

Decisions and Action Items

A sense of those present indicates no objection to setting tunable A=16 in cbor-packed if the range of simple values can be shared for general packing/templating by other mechanisms.
Work on Vadim's proposal for a separate document registering generic simple values (and cbor-seat more broadly) is deferred until the three core cbor-packed related documents are finished.
The interim meeting scheduled for May 28th is canceled due to chair unavailability.

Next Steps

Continue discussion on the mailing list regarding the final values for cbor-packed tunables (A, B, C) and how to best address the shared use of simple values (e.g., via editorial clarifications in cbor-packed or a separate registration document).
Working group members are requested to review Lawrence's alternative draft for cbor-preferred-serialization (CDE) once published.
Engage in broader discussion on the mailing list regarding the direction and consensus for the cbor-preferred-serialization topic.
The next interim meeting is scheduled for June 11th.