add method to update sub-schema in simple allOf case

[tharwan]

This PR will try to merge the allOf sub-schema with the surround schema in a simple way. This addresses issues of otherwise missing properties like defaults. For example:

  {
    "default": "A",
    "allOf": [
      {
        "title": "all of enum",
        "description": "An enumeration.",
        "enum": [
          "A",
          "B",
          "C"
        ],
        "type": "string"
      }
    ]
  }

this would not render a dropdown with a default selection of A.

[dokempf]

Did I understand correctly, that this code is happening only iff the number of subschemas for the allOf rule is 1? Does this happen in your use case? And would it be possible to just drop it from the input schema? I am asking because I am hesitant to add complex logic that alters schemas to enable use cases that can easily prevented. I was always prepared to flag some JSONSchema features as not supported for this project (and the allOf rule is quite a contender...)

[tharwan]

Our use case is to display json schemas that originate from pydantic. Which uses allOf in many cases to refer to sub-models (compare with pydantic/pydantic#2592).

I think pydantic is an extremely common way in the python eco-system to deal with json and json schemas, so it would make sense to support it.

I am very much open to all suggestions how to simplify this.

Edit:
Example

class Status(Enum):
    A = "A"
    B = "B"
    C = "C"

class Model(BaseModel):
    status: Status = Status.A

Model.schema()

results in

{
  "title": "Model",
  "type": "object",
  "properties": {
    "status": {
      "default": "A",
      "allOf": [
        {
          "$ref": "#/definitions/Status"
        }
      ]
    }
  },
  "definitions": {
    "Status": {
      "title": "Status",
      "description": "An enumeration.",
      "enum": [
        "A",
        "B",
        "C"
      ]
    }
  }
}

which the above case is derived from. Note that the reference can be resolved prior with https://pypi.org/project/jsonref/

[dokempf]

I am fine with supporting this as you have a valid use case.