To Add PNA support into BMv2. about behavioral-model HOT 5 OPEN

In general, I would clean up the original PRs before we can even start to make suggestions on how to break them down. This means removing nanomsg and thrift.

from behavioral-model.

Ping @antoninbas @jafingerhut @thantry @drobinkent

from behavioral-model.

First, thank you @rupesh-chiluka-marvell for your effort on open sourcing this work!

I'm not the most qualified person to give suggestions here, but just want to share some high-level thoughts and hopefully stimulate more discussions that could lead to better and more detailed solutions.

Regarding PR splitting, I think a test-driven approach might work for this process. The idea is to split into a series of PRs with each making a single P4 test program work. For example, the first PR could introduce minimal changes to BMv2 and p4c to enable a minimal P4 PNA program (e.g. single state parser + empty control block) to compile with p4c and run with BMv2. Each subsequent PRs will make the test P4 program a little bit more complicated.

from behavioral-model.

Also to give a better context to everyone, I'd like to attach @antoninbas's suggestion from the email thread:

We will have a better idea of the actual size of the bmv2 PR once the nanomsg files are removed. But yes, there should be potential for splitting up the PR after that, e.g., externs could be added in a subsequent PR.

from behavioral-model.

Created a new PR with just skeleton of pna_nic: #1255

from behavioral-model.