Evaluate support of Pydantic-v2 generated JSON Schemas

[tcompa]

When testing the schemas in fractal-analytics-platform/fractal-tasks-core#737, we should be aware of how Pydantic V2 changes handling of Optional attributes (which includes the use of anyOf, and the fact that it seems that attributes are always included). The new behavior is sometimes confusing, see:

• JSON Schema for optional fields is incorrect. pydantic/pydantic#7161
• Allow optional properties to be truely optional pydantic/pydantic#8394

It's likely that this produces issues in the form generator.
By taking the example in pydantic/pydantic#7161 (comment), we end up with this JSON Schema:

{
  "properties": {
    "field": {
      "anyOf": [
        {
          "type": "string"
        },
        {
          "type": "null"
        }
      ],
      "default": null,
      "title": "Field"
    }
  },
  "title": "Model",
  "type": "object"
}

If we try it out in https://rjsf-team.github.io/react-jsonschema-form/, we find this behavior:

which IMO is a suboptimal UX, because it forces a two-steps process: first I need to decide whether to include the argument, and then I can decide whether/how to fill it.

---

TLDR We'll need to keep an eye on this, when experimenting with schema-based form generators.

[tcompa]

This issue was a FYI, but the plan is rather to make sure that fractal-tasks-core produces appropriate JSON Schemas (that is, without the additional anyOf pattern for nullable field) even with Pydantic v2.

Closing.