[kotlin] Add AnyOf/oneOf to multiplatform

[chizhanov]

This PR adds support for AnyOf and OneOf with the generateOneOfAnyOfWrappers flag in Kotlin Multiplatform.

I removed some checks to compare with the original anyof_class.mustache and oneof_class.mustache because multiplutform is more limited(kotlinx.serialization only).

For testing, you can use the following code:

openapi: 3.0.0
info:
  version: 1.0.0
  title: pets

paths:
  /petsOneOf:
    patch:
      requestBody:
        content:
          application/json:
            schema:
              oneOf:
                - $ref: '#/components/schemas/PetByAge'
                - $ref: '#/components/schemas/PetByType'
      responses:
        '200':
          description: Updated

  /petsAnyOf:
    patch:
      requestBody:
        content:
          application/json:
            schema:
              oneOf:
                - $ref: '#/components/schemas/PetByAge'
                - $ref: '#/components/schemas/PetByType'
      responses:
        '200':
          description: Updated

components:
  schemas:
    PetByAge:
      type: object
      properties:
        age:
          type: integer
        nickname:
          type: string
      required:
        - age

    PetByType:
      type: object
      properties:
        pet_type:
          type: string
          enum: [Cat, Dog]
        hunts:
          type: boolean
      required:
        - pet_type

It would be great if someone added testing for these cases, but I can't do that because I don't understand how they are organized here.

@dr4ke616 (2018/08) @karismann (2019/03) @Zomzog (2019/04) @andrewemery (2019/10) @4brunu (2019/11) @yutaka0m (2020/03) @stefankoppier (2022/06) @e5l (2024/10)

@wing328

[wing328]

thanks for the PR

is it correct to say that you've tested this fix locally to confirm it works for your use cases?

[chizhanov]

thanks for the PR

is it correct to say that you've tested this fix locally to confirm it works for your use cases?

Yes, that’s correct. I verified it with the example from the first post.
In addition, I tested it against a large OpenAPI specification containing AnyOf and OneOf.
The generated code worked correctly in that case as well. While the result is not very convenient to use, it is functionally correct.

[wing328]

let's give it a try. i'll try to add some tests later

[wing328]

While the result is not very convenient to use,

can you please elaborate a bit more?

[chizhanov]

While the result is not very convenient to use,

can you please elaborate a bit more?

Yes. In the generated code(ResponseInner) our model is stored in the actualInstance field, which always has type Any. This is not ideal, because it loses type-safety.
It would be much better if the generator could produce a sealed class (or a similar construct) representing the possible types. That way the compiler could help ensure correct usage instead of relying on runtime checks.