Use P4 annotations instead of heuristic in the code generator

[KrisNey-MSFT]

The current SAI API code generator https://github.com/Azure/DASH/blob/main/dash-pipeline/SAI/sai_api_gen.py reads the P4Info (derived from dash-pipeline.p4) and emits SAI header (.h) files. Some of the transformations are a bit involved and contain hidden assumptions, and the P4 code does not explicitly show the relationships between P4 and SAI headers, for example when is a SAI attribute mandatory. Using annotations would allow the P4 code to explicitly declare, for example, the SAI table name, or identify a mandatory SAI attribute. Using attributes would also greatly simplfy the code generator, since it would only have to do simple transformations without lots of special-case code.

This idea was originally suggested here: #240 (comment)

A sample experimental annotation is provided below. You can compile the P4 code using make p4 and read the resulting P4Info file dash-pipeline/bmv2/dash_pipeline.bmv2/dash_pipeline_bmv2_p4rt.txt in your build environment.

https://github.com/chrispsommers/DASH/blob/4ce46efa65bf49bd342efdf43f4b08d454cdb408/dash-pipeline/bmv2/dash_pipeline.p4#L39

[yusefMS06]

@marian-pritsak Does this issue need to stay in the queue?

[chrispsommers]

@marian-pritsak Does this issue need to stay in the queue?

We need to keep this item open. It was agreed this is the long term solution.

Note, this is a great project for another volunteer, it mainly requires appropriate annotations to P4, and Python changes to https://github.com/Azure/DASH/blob/main/dash-pipeline/SAI/sai_api_gen.py which will become simpler as a result. It could be done incrementally.

[KrisNey-MSFT]

Annotations for tables, keys, action parameters - Marian will prepare a PR for next week.
Modify code generator for annotations in parallel.
Very straightforward, lookup key in annotation, pass to dictionary, then to templates.
0 logic.

[KrisNey-MSFT]

hi @chrispsommers - can this one be closed then, considering we have 473?

[chrispsommers]

No, as @r12f mentioned, the old way to name tables is still in place. This is a step towards the eventual solution.

[r12f]

Yea, sorry. I probably gave my commit a really bad name... I have already started the work moving the annotation to Sai.

[r12f]

The first PR is #477 . More PRs will come later, since I am taking today oof... :(

[KrisNey-MSFT]

Hi @r12f - should we close this one as finished, in favor of #477 ?
TY, Krisney-MSFT

[chrispsommers]

There are a series of MRs which all contributed to this, but I agree this can be closed. Perhaps @r12f can list the MRs for posterity in a closing comment. Thanks!

[KrisNey-MSFT]

Closing