[Openpgp-dt] Design Team meeting notes: 2022-05-13: packet type grammar edition

[Daniel Kahn Gillmor <[e-mail]>]

Hey folks--

today, the DT met:

 - stephen
 - justus
 - daniel huigens
 - dkg

We talked the entire session about what to do about the packet grammars,
in particular questions about strictness or leniency for consumers of
the grammars, and how to characterize "criticality" for packet types.

This is the issue where this came up:

    https://gitlab.com/openpgp-wg/rfc4880bis/-/issues/121

We discussed it in light of a proposal that I wrote:

 https://gitlab.com/openpgp-wg/rfc4880bis/-/merge_requests/189

   - morphs padding packet into non-critical packet (that can be
     padding)

   - defines all other packets as implicitly "critical" -- if they show
     up someplace that is not defined by the grammar in use, the
     implementation should explicitly reject the entire surrounding
     sequence.

status quo (in 4880) doesn't really distinguish here, & we'd like to add
more guidance as implementations have done varying things, which leads
to a risky situation, for example:

 - one packet sequence can masquerade as another;

 - one encrypted message might decrypt to a different ciphertext
   depending on implementation


- everyone (except SF:-) offers to try create an MR with their least-bad
  option for such guidance

- dkg offered !189
- TODO: justus to write a strict proposal with a partition of remaining packet tag space between critical/non-critical
- TODO: danielh to write a proposal defines skipping surprising packets, and rejecting the stream when encountering an unknown/unparseable "contentful" packet

Talk to y'all Tuesday!

     --dkg