[P4-design] rough notes from 11/7 meeting
jnfoster at cs.cornell.edu
Mon Nov 7 19:01:38 EST 2016
* Our goal is to deal with all remaining issues today, leaving the
November 21st date to verify that all solutions, some of which may
require additional work on prototyping, are adequate.
* The initial release of the spec to the P4.org community will be
labeled as a draft. There will still be time to improve the packaging
of the spec and even to fix any show-stopping issues that come up
during public review.
* GitHub Issues
- #76 more explicit control plane parameters
Issue is how to specify data plane and control plane arguments in
actions. We seem to need both. The issue is how to make the syntax
explicit (so the control-plane interface is clear) and yet subtle (so
it's not too clunky for programmers).
Action item: tentatively go with two braces, but add syntactic sugar
for common cases. Edwin will review to see if he is happy.
- #81 support for local externs
Issue is how to deal with local externs -- e.g., a use of a checksum
unit in a parser. We observe that we can use @atomic annotations to
group sets of operations involving an extern together, which seems to
handle our current use cases.
Action item: we'll take the proposed change.
- #59 header stacks have non-standard behavior with valid bits
Action item: Andy, Nate, and Mihai to see if the current spec can be
massaged into a clean stack in the parser + array elsewhere idiom. For
example, make direct access to elements in the parser illegal.
- #18 / #58
- #99 annotations
P4_16 annotations are less powerful than say Java's. Do we want to
allow a richer annotation language? For example, lists of annotation
elements? name = expression?
Action item: Chris to implement and we are likely to adopt this in the spec.
- #98 action profile
Action item: go out with profile approach for now but possibly revisit
in the future.
Action item: we will adopt the proposed relaxations and will add some
text to the spec to allow implementations some flexibility.
* 1.1 spec
We agreed to publish all released versions of the spec on the web
page, possibly tagged as "unsupported."
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the P4-design