feat: [DevOps] Poc RPT

[rpanackal]

Context

AI/ai-sdk-java-backlog#350.

PoC for supporting the latest SAP RPT (Tabular AI) service

Feature scope:

• New foundationmodels module sap-rpt introduced
• Using spec obtained from sap-rpt-1-public-content
• Unit tests and e2e test in place
• Introduce new RptClient and RptModel class

Definition of Done

• Functionality scope stated & covered
• Tests cover the scope above
• Error handling created / updated & covered by the tests above
• Aligned changes with the JavaScript SDK
• Documentation updated
• Release notes updated

[rpanackal]

PoC RPT

Important Mentions:

• The PoC do not support operation on path /predict_parquet for now. I have explicitly disabled this in pom.xml

  1. There is no documentation about endpoint.

  2. The related spec is poorly defined for generation strategy. The prediction_config field in request payload is defined as type string but it should be a JSON of schema PredictionConfig. So, currently, we can't offer any type safety without adding convenience layer on top.

     Note: The exclusion of path /predict_parquet does not exclude generation of model classes used by the path eg: BodyPredictParquet.

• The provided spec file defines components with dynamic properties which vary based on input column names. Example below.

  Column in Request Payload

  {
    "columns": {
      "anyOf": [
        {
          "additionalProperties": {
            "items": {
              "anyOf": [
                {
                  "type": "string"
                },
                {
                  "type": "number"
                },
                {
                  "type": "integer"
                },
                {
                  "type": "null"
                }
              ]
            },
            "type": "array"
          },
          "type": "object"
        },
        {
          "type": "null"
        }
      ],
      "default": null,
      "description": "Alternative to rows: columns of data where each key is a column name and the value is a list of all column values. Either \"rows\" or \"columns\" must be provided.",
      "title": "Columns"
    }
  }

• Some confusing bits in the latest spec (its still beta anyway)

  • Incorrect use of anyOf, for example "columns" field in request, "prediction" field in response. Should have been a oneOf
  • Description states that either rows or columns must be provided on request which sounds like another oneOf usecase, but current schema allow both to be null.
  • Fields that can be omitted are given a default value of null to indicate optionality.

[newtork]

(Question)

Can we use the new generator without spring dependencies?

[rpanackal]

Yes. I am working on the related cloud sdk item to fix spring optionality first. Then, later make use of the same in the new module client

[newtork]

(Major)

The PR title says "PoC", therefore the module shouldn't be released at this level of confidence.

E.g.

    <profile>
      <id>non-release</id>
      <activation>
        <property>
          <name>!release</name>
        </property>
      </activation>
      <modules>
        <module>sample-code/spring-app</module>
+       <module>foundation-models/sap-rpt</module>
      </modules>
    </profile>

[rpanackal]

I believe the spec is currently missing a tag, so the generated api class gets a default name DefaultApi.

[Jonas-Isr]

Mostly minor issues. Looks pretty good :)

[Jonas-Isr]

(Minor/Question)

This is a very non-descriptive class name (which does not come from the spec itself as far as I can see). Does it make sense to rename that when generating?