table-schema using jsonschema native terms items and properties rather than field

[jgunstone]

I'm a big pydantic user and main author of ipyautoui, a package which builds UI forms from a (often pydantic generated) jsonschema.
In ipyautoui a dataframe can be defined as follows

pydantic code to make schema

from pydantic import Field, constr, RootModel, ConfigDict
from ipyautoui.basemodel import BaseModel
from typing_extensions import Annotated
from datetime import datetime


class NestedObject(BaseModel):
    """description in docstring"""

    string1: str = Field(default="adsf", description="a description about my string")
    int_slider1: Annotated[int, Field(ge=0, le=3)] = 2
    int_text1: int = 1
    date: datetime


class RootArray(RootModel):
    root: list[NestedObject]
    model_config = ConfigDict(
        json_schema_extra={"format": "dataframe", "primaryKey": ["date"]}
    )

from jsonref import replace_refs
from pprint import pprint

schema = RootArray.model_json_schema()
# ----
schema = replace_refs(schema)
schema = {k: v for k, v in schema.items() if k != "$defs"}
# ^ for simplicity / illustration purposes I removed the $defs / internal json link

pprint(schema)

{
    "format": "dataframe",
    "items": {
        "description": "description in docstring",
        "properties": {
            "date": {
                "format": "date-time",
                "title": "Date",
                "type": "string"
            },
            "int_slider1": {
                "default": 2,
                "maximum": 3,
                "minimum": 0,
                "title": "Int Slider1",
                "type": "integer"
            },
            "int_text1": {
                "default": 1,
                "title": "Int Text1",
                "type": "integer"
            },
            "string1": {
                "default": "adsf",
                "description": "a description about my "
                "string",
                "title": "String1",
                "type": "string"
            }
        },
        "required": ["date"
        ],
        "title": "NestedObject",
        "type": "object"
    },
    "primaryKey":  ["date"],
    "title": "RootArray",
    "type": "array"
}

note: length of items object above must equal 1 for this to be compliant.

the equivalent following the frictionless data spec would be:

{
    "fields": [
        {"format": "date-time", "title": "Date", "type": "string", "name": "date"},
        {
            "default": 2,
            "maximum": 3,
            "minimum": 0,
            "name": "int_slider1",
            "title": "Int Slider1",
            "type": "integer",
        },
        {"default": 1, "name": "int_text1", "title": "Int Text1", "type": "integer"},
        {
            "default": "adsf",
            "description": "a description about my " "string",
            "name": "string1",
            "title": "String1",
            "type": "string",
        },
    ],
    "primaryKey":  ["date"],
    "title": "RootArray",
}

I think that the are some benefits to the pydantic-style approach:

• jsonschema native terms "type", "format", "items", "properties" used https://www.learnjsonschema.com/2020-12/
• the dataframe itself is defined as a json-compliant array, "type"="array" and "format" = "dataframe".
  • the resulting tabular data is the same as outlined in the spec: https://specs.frictionlessdata.io/table-schema/#tabular-data
• row items (objects) are defined by the items keyword (i.e. no need to introduce additional term "field")
  • the required "name" field is implicitly required as it is the object key
• this allows for validation using standard jsonschema libraries and pydantic

I'm interested if there would be any appetite to support this type of specification - perhaps as an allowed alternative? or possible future version specification?

I'm also interested to know if users here think that this is a good way of defining tabular data, or if there are issues with this suggestion that I've missed?

many thanks