LDWG meeting minutes, April 1, 2019

Attendees

Issue: https://github.com/p4lang/p4-spec/pull/743
Pull Request: https://github.com/p4lang/p4c/issues/1771
Proposal is to add a new error InvalidParserArgument to v1model.p4
Mihai and Antonin added support for advance in the parser to BmV2.
P4_16 allows targets to restrict header types to be multiples of 8.
Action Items:
- Standardize jargon "target" and "architecture"
- Add to ChangeLog

Pull Request: https://github.com/p4lang/p4-spec/pull/743
Proposal is to clarify notion of priority for const entries
Discussion focused on whether to define the semantics in a general way, or specifically in terms of `ternary.
Action item: rewrite text to be more general.

Pull Request: https://github.com/p4lang/p4-spec/pull/736
Proposal is a record of an off-line discussion between @hanw and @jafingerhut
There are several open questions, such as what are restrictions on struct and how lookahead should work.
Perhaps there should be other kinds of packet_in and packet_out objects to handle these use cases?
This feature may not be needed; it was developed in preparation for minSizeInBits proposal.
Action Item: We'll let this sit until there is a compelling use case.

** Pull Request:** https://github.com/p4lang/p4-spec/pull/736

Only change since last discussion is to add min before names minSizeInBits and minSizeInBytes
In particular, max operations are not supported; if someone has a compelling use case, they can develop it.
Action Item: Merge into the spec

Issue: https://github.com/p4lang/p4-spec/pull/1694
Pull Request: https://github.com/p4lang/p4c/pull/1704
Issue stems from the converter, which converted intrinsic metadata into user metadata, and hence Bmv2 couldn't find it.
Another way to view the issue is that compilation is not idempotent
The fix maps intrinsic metadata to standard metadata
@hanw wants to check that this would not have impact on a proprietary compiler
Also don't want to lose switch.p4 as a test case for the compiler
Action Item: More work to do. We will revisit.

Pull Request: https://github.com/p4lang/p4-spec/pull/656
Proposal is to allow overlays, which give a different view of underlying state
Mihai would like this feature to be purely syntactic sugar
We believe that the current proposal, with overlays initialized just once, satisfies this property.
We want to make sure that overlays do not make it easy to write programs that do too much copying.
Action Items:
- @mbudiu-vmw will continue to push forward with the overlay proposal
- @antoninbas and @hanw will explore struct-based approach to their hash example