Notes from 2023-Mar-17 meeting on data-plane-modifiable per-table-entry state

[Andy Fingerhut]

I modified the last slide a bit on ideas for moving forward with some idea.

The basic idea: Don't change the language spec now.  Instead, create a prototype implementation in open source p4c that multiple vendors use for their own back-end implementation.  The main challenge would be doing it in a way that can be easily kept up to date with main branch updates to p4c as time goes on, until and unless it becomes part of the language spec.

Notes here: https://docs.google.com/presentation/d/15vZJETWr0Cyht-dcQFnat28A3c9FoY7Z/edit?usp=share_link&ouid=113165450274035133953&rtpof=true&sd=true

Andy

[hem...@mnkcg.com]

If an if-statement is used to write to an action parameter, programmers may still use the if-statement to contain no writes to any action parameter. Also, as Andy has pointed out, if there are 17 lines before an action parameter is written to, the issue gets interesting. But if a new tag is used with action parameter , e.g., rmw, then compiler doesn’t care about 17 lines. All p4c has to enforce is, only a rmw parameter is allowed as left hand side of an assignment statement.

 

Hemant

--
You received this message because you are subscribed to the Google Groups "p4-arch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to p4-arch+u...@lists.p4.org.
To view this discussion on the web visit https://groups.google.com/a/lists.p4.org/d/msgid/p4-arch/CAKvLtDZUG_dS5zJo1XEOvZqyWSy%2BEP9GOuszMWqQMMxwqv%2Bw%2BA%40mail.gmail.com.

[hem...@mnkcg.com]

I wonder why we discussed any solution today in action body when I had sent this email below? Do folks have an earthquake issue with rmw type for an action parameter? Using rmw parameter type does not need any change in action body.

 

Thanks,

 

Hemant

--
You received this message because you are subscribed to the Google Groups "p4-design" group.
To unsubscribe from this group and stop receiving emails from it, send an email to p4-design+...@lists.p4.org.
To view this discussion on the web visit https://groups.google.com/a/lists.p4.org/d/msgid/p4-design/04a601d959c5%244f4f3280%24eded9780%24%40mnkcg.com.

[Andy Fingerhut]

See my other email and the summary of the two primary proposed approaches, which I hope accurate captures the current concerns: p4-guide/data-plane-writable-per-table-entry-state at master · jafingerhut/p4-guide (github.com)

Earthquake issue?  No.

Concerns from a language spec that strongly prefers to avoid backwards-incompatible changes, now and in the future?  Yes.

Andy