Markdown Version | Session Recording

Session Date/Time: 01 Jun 2022 14:00

CBOR

Summary

The CBOR working group met to discuss outstanding administrative items, future directions for C-DDL, and a new design proposal for CBOR Packed. A decision was made regarding an errata report for RFC 8610. The bulk of the session focused on a refined proposal for CBOR Packed, including a merged table for shared data, argument references, and a flexible function tag mechanism for custom processing. Future meeting plans and next steps for the CBOR Packed draft were also outlined.

Key Discussion Points

Administrative Issues

• Errata Report 6575 on RFC 8610 (C-DDL): Carsten Bormann noted this errata, reported a year ago, concerns a missing internal cross-link in RFC 8610 regarding how a specific effect mentioned in Section 2.2.3 is supplied in Section 3.6. A previous mailing list discussion suggested "Hold for Document Update" as the resolution.

C-DDL Evolution

• Syntax for Semantical Definitions: Carsten Bormann highlighted the limitations of the current hash 7.25 syntax in C-DDL, particularly when dealing with generics or computed tag numbers. This syntax, which uses additional information rather than the whole argument, has proven problematic in several drafts.
• Wider Applicability of C-DDL: A point was raised about C-DDL's potential use beyond CBOR and JSON, for example, in defining data models for HTTP Structured Field Values. Currently, C-DDL is approximately 90% suitable for this, but lacks a clear mechanism to base definitions on different generic data models. This suggests a need for C-DDL updates to support broader data model definitions.
• This topic is on the agenda for future updates to C-DDL, prompting the working group to consider if C-DDL should explicitly support defining data models for other formats or remain focused on CBOR/JSON applications.

CBOR Packed (Draft Discussion)

Carsten Bormann presented a refined design for CBOR Packed, incorporating feedback from a previous meeting.

• Design Overview:
  • Single Merged Table: The design moves away from separate prefix and suffix tables to a single merged table. This allows using the same string for both prefixes and suffixes, using different numerical indices for distinction.
  • Reference Types:
    • Sharing Reference: A tag (e.g., Tag 51) that directly points to an item in the shared table, extracting it unchanged.
    • Argument Reference: A tag that includes both its own data (the argument) and a reference into the merged table. A bit indicates whether it's a prefix or suffix reference.
  • Processing Model: The accessor-in-place model (also known as pointer tracing) is the primary model for defining meaning during unpacking. The unpacker model can be described in terms of the accessor-in-place model.
• Argument Processing:
  • When both the reference and the "rump" (unpacked content) are strings, string concatenation occurs.
  • If both are arrays, array concatenation.
  • If both are maps, map merging.
  • Function Process: For all other cases, particularly when a tag is involved, a "function process" is invoked. This is a new extension mechanism.
• Function Tags as Extension Point:
  • A function tag on either the reference or rump value defines custom processing. A flag (the prefix vs. suffix bit) determines which of the two parameters (reference or rump) contains the function tag.
  • This allows defining application-specific processing beyond basic concatenation or merging.
  • A generic unpacker can identify unknown function tags and defer processing to the application.
  • Example: A "midfix" tag was presented, which takes an array, places the first element, then the referenced data, then the last element.
• Extension Points:
  • New table setup tags can be defined.
  • New function tags can be defined.
  • Avoid New Reference Tags: It was proposed that the set of reference tags themselves (e.g., Tag 51) should ideally not be extended to maintain forward compatibility for existing implementations.
• Open Questions/Considerations:
  • Example Function Tag: Should the CBOR Packed document include at least one concrete example function tag (like the "midfix" example) to encourage implementation?
  • Error Handling: Are there extension points needed for error handling?
  • Tag 6 Ambiguity: Current Tag 6 (major type 6) for sharing/argument references could become ambiguous if an integer argument is supplied to a function tag. A proposal was made to reserve another tag (e.g., Tag 224) as a reference tag to avoid this.
  • Complexity/Profiling: The need for profiles (e.g., "sharing-only" for pointer tracing vs. "function tag processing") and explicit tags to identify specific profiles was discussed.
• A sense of those present indicated no immediate objections to the proposed design, and further feedback was encouraged on the mailing list.

Decisions and Action Items

• Errata 6575 on RFC 8610: The working group's advice to the ADs is to close errata 6575 as "Hold for Document Update," with notes explaining the missing internal cross-link is resolved by existing content.
  • Action: Carsten Bormann will send an email to the ADs with this recommendation.
• CBOR Packed Draft: Carsten Bormann will integrate the discussed design changes into the CBOR Packed draft.
  • Action: Once a new revision of the CBOR Packed draft is available and internal WG comments are addressed, the chairs will forward it to the ADs for Working Group Last Call.

Next Steps

• The CBOR working group plans to hold a one-hour session at IETF 118 in Philadelphia. An agenda for this meeting will be developed over the coming month.
• The assumption is that the CBOR Packed document will have a completed reaction to working group last call comments before the I-D deadline for IETF 118, allowing for discussion at that meeting if needed.
• Carsten Bormann will post a message to discus@ietf.org to suggest improved sorting of Jabber users in the Medico interface.