Does multicast replication preserve local metadata in v1model/simple switch?

[smolkaj]

Empirically speaking, the answer is yes, on BMv2. However I don't see this documented anywhere currently. Maybe @jafingerhut would know?

@zhenk14 for visibility.

[jafingerhut]

As far as I know, multicast and unicast are equivalent in the v1model implementation for metadata that is preserved from ingress to egress.

Note: In case you have not heard me say it before, unless the P4_14 specification says something explicitly (and on many such detailed questions like this, it is often silent on the topic), I consider the behavioral-model implementation of v1model to be the effective definition of the v1model architecture. Yes, occasionally we have changed things in this implementation because it seems wrong, but not many such things have been changed since behavioral-model was written.

[smolkaj]

Treating unicast & multicast uniformly makes sense.

Treating BMv2 as the definition also makes sense at this point, given that the P4_14 spec is very incomplete. A search for "multicast" in the spec surfaces nearly no information at all.

[jafingerhut]

If you have any suggested PR for this documentation file that would help clarify the BMv2 v1model behavior here, I'd be happy to review it: https://github.com/p4lang/behavioral-model/blob/main/docs/simple_switch.md

[smolkaj]

Will send a PR after https://github.com/p4lang/behavioral-model/pull/1204/files goes in to avoid merge conflicts.

[jafingerhut]

Sorry for the long delay. I just approved and merged PR #1204

[github-actions[bot]]

This issue is stale because it has been open 180 days with no activity. Remove stale label or comment, or this will be closed in 180 days

[smolkaj]

Reminder to myself: need to send a PR to clarify this.