Missing names sanitization cause conflicts

[andreaTP]

Description

If the specification contains an endpoint called models ( most notably OpenAI ).

Apparently, there are issues with things getting not generated or overwritten ...

Repro

A minimal repro looks like:

openapi: 3.0.0
info:
  title: Test
  version: 1.0.0
  description: something
paths:
  /api/something/v1:
    post:
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: "#/components/schemas/Test1"
      responses:
        "200":
          description: "something"
          content:
            application/json: {}
  /models:
    post:
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: "#/components/schemas/Test2"
      responses:
        "200":
          description: "something"
          content:
            application/json: {}
components:
  schemas:
    Test1:
      description: this is a first test
      properties:
        test:
          description:
          type: string
      type: object
    Test2:
      description: this is a second test
      properties:
        test:
          description:
          type: integer
      type: object

Resolution

I'm not sure what would need to be escaped properly. Let me know if you need a further narrowing down of the issue.

[baywet]

🤦‍♂️ we hardcoded that namespace name and didn't make it reserved...
I'm guessing the best approach would probably be to project a "models" root path segment to "modelsRequests" or something like that in the builder (before any refiner). Thoughts?
Is this something you want to take on?

[andreaTP]

I'm guessing the best approach would probably be to project a "models" root path segment to "modelsRequests" or something like that in the builder (before any refiner). Thoughts?

I agree that this should be done before the refiner phase.
Not sure how this works now, but we can probably add models to the reserved words for all of the languages.
Additionally we can make the models namespace customizable by the user, wdyt?

Is this something you want to take on?

My backlog in this project is growing a little too fast compared to my availability those days 😅
I'll feel more comfortable if we keep it unassigned and I can pick it up, eventually, if no one else resolves it before me.

[baywet]

Not sure how this works now, but we can probably add models to the reserved words for all of the languages.

This would lead to the "hardcoded" models namespace to be considered conflicting and get renamed in the current design. This is why I was suggesting to do that earlier.

Additionally we can make the models namespace customizable by the user, wdyt?

I'm torn on this one. On one hand we want to keep as few parameters as possible (simplicity), on the other we allow configuration of the root namespace name and the client class. I think that even if we allowed configuration, we'd still need to handle potential collisions. So I'd consider configurability as an orthogonal aspect to this one.

My backlog in this project is growing a little too fast compared to my availability those days

I feel you... I'll try to have a look at it on the next "sprint", let's consider it assigned to me for now and ping me on this issue if you want to take it over before I get to it.

[andreaTP]

I think that even if we allowed configuration, we'd still need to handle potential collisions. So I'd consider configurability as an orthogonal aspect to this one.

Agreed.

I'll try to have a look at it on the next "sprint", let's consider it assigned to me for now and ping me on this issue if you want to take it over before I get to it.

Deal 🙂

[baywet]

Authored #3479 to address the problem