[@spec2ts/openapi] [bug] schema with "oneOf" keyword ignores "properties" keyword

[marcogrcr]

I encountered a bug when attempting to generate types for the following JSON schema:

openapi: 3.0.3
info:
  title: example
  version: 1.0.0
paths:
  /example:
    get:
      responses:
        200:
          description: OK
          content:
            application/json:
              schema:
                type: object
                properties:
                  name:
                    type: string
                  value:
                    type: integer
                oneOf:
                  - properties:
                      value:
                        minimum: 0
                        maximum: 5
                  - properties:
                      value:
                        minimum: 10
                        maximum: 15

Expected result:

The library detects that there's an overlap between the properties names of $.oneOf and $.properties and generates a type that satisfies all specifications:

export interface GetExampleResponse {
    name?: string;
    value?: number;
}

Actual result:

The library ignores $.properties when $.oneOf is present. It also does not detect there's an overlap between the elements of the $.oneOf array and generates a type union that's redundant.

export type GetExampleResponse = {
    value?: any;
} | {
    value?: any;
};

[mtrimolet]

I think the reason for this is either:

• due to a design decision made by the initiator of the project (poke @SomaticIT) in which case all of the following should be ignored, or
• scared into the misunderstanding of jsonschema validation/annotation concepts we had when writing the core-parser, which is what I'll try to explicit below.

After diving into jsonschema drafts I get the impression that we are considering it as a "cumulative properties" model or worse a "singleton datatype" model, where it should be considered a "cumulative hierarchical constraints" model for building a datatype, starting from type unknown which "validates anything", narrowing/combining the type for each constraint and circuit breaking with never type which "validates nothing".

This would imply a complete rewriting of the parse function, so I'll dive into it and give it a try but let's talk about it so we can avoid wasting time 😅