Question: Of what use are constants in BMv2 JSON field_lists?

[jafingerhut]

simple_switch can read without error arbitrary mixes of type "hexstr" and "field" inside of the "elements" list, as shown in the example below:

  "field_lists" : [
    {
      "id" : 1,
      "name" : "tuple_0",
      "source_info" : {
        "filename" : "resubmit-constant-prop-bug-p414.p4",
        "line" : 66,
        "column" : 0,
        "source_fragment" : "field_list resubmit_FL { ..."
      },
      "elements" : [
        {
          "type" : "hexstr",
          "value" : "0x0003"
        },
        {
          "type" : "field",
          "value" : ["scalars", "tmp"]
        },
        {
          "type" : "field",
          "value" : ["scalars", "tmp_0"]
        }
      ]
    }
  ],

I found the commit to behavioral-model where the capability to accept "hexstr" elements was added, but it wasn't clear to me from reading that why it was added.

Does P4_14 allow constants in a field_list definition?

It makes sense that P4_16 allows constants in a list of fields given as an arg to a call like a hash or checksum calculation. Was that the motivation?

Separate from that question, does it seem to you like it would always be a mistake of some kind if a field list used for recirculate, resubmit, or clone3 primitives of simple_switch contained an element with type "hexstr"? I cannot think of any reason why that would make sense. The current p4c can sometimes generate such a field_list, but I am guessing that it is always a sign of a bug in p4c when it does so.

[jafingerhut]

Answering one of my own questions, which I should have looked up before asking: The P4_14 spec does allow constant values in field_list definitions. Probably primarily for the purpose of hash/checksum calculation input lists, would be my guess.

The other question about whether constant values make sense in a recirculate/resubmit/clone3 primitive is still open, in my mind.

[jafingerhut]

I am pretty sure that the answer to the question: "Does it make sense for a recirculate/resubmit/clone primitive to contain constant values?" is "No, that is a sign of a bug in p4c bmv2 back end."

I have a Python program that can detect this and a few other similar problems in such field lists, written to see how prevalent this issue is in p4c: p4lang/p4c#1669

The Python program that reads BMv2 JSON files and looks for possible problems is here: https://github.com/p4pktgen/p4pktgen/blob/master/tools/bmv2-json-check.py