Extra keys in fields are missing within oneOf / anyOf structures

[simontaurus]

Describe the bug field_include_all_keys=True should store all extra schema keys in Field attributes. However, this does not work for schemas that contain oneOf or anyOf

To Reproduce

Example schema:

        {
            "id": "example",
            "title": "Example",
            "type": "object",
            "properties": {
                "type": {"type": "string", "default": ["example"]},
                "prop1": {"type": "string", "custom_key": "custom_value"},
                "prop2": {
                    "custom_key": "custom_value",
                    "properties": {
                        "subprop0": {"type": "string", "custom_key": "custom_value_0"}
                    },
                    "oneOf": [
                        {
                            "title": "Subschema1",
                            "type": "object",
                            "properties": {
                                "subprop1": {"type": "string", "custom_key": "custom_value_1"}
                            }
                        },
                        {
                            "title": "Subschema2",
                            "type": "object",
                            "properties": {
                                "subprop2": {"type": "string", "custom_key": "custom_value_2"}
                            }
                        }
                    ]
                }
            }
        }

Used commandline:

$ datamodel-codegen --input-file-type jsonschema --input example.json --field-include-all-keys

Actual Result:

# generated by datamodel-codegen:
#   filename:  example.json
#   timestamp: 2025-05-18T05:34:14+00:00

from __future__ import annotations

from typing import Optional, Union

from pydantic import Field


class Subschema1(BaseModel):
    subprop0: Optional[str] = None
    subprop1: Optional[str] = None


class Subschema2(BaseModel):
    subprop0: Optional[str] = None
    subprop2: Optional[str] = None


class Example(BaseModel):
    type: Optional[str] = ["example"]
    prop1: Optional[str] = Field(None, json_schema_extra={"custom_key": "custom_value"})
    prop2: Optional[Union[Subschema1, Subschema2]] = Field(
        None, json_schema_extra={"custom_key": "custom_value"}
    )

Expected behavior

Custom keys contained in Fields:

# generated by datamodel-codegen:
#   filename:  example.json
#   timestamp: 2025-05-18T05:26:34+00:00

from __future__ import annotations

from typing import Optional, Union

from pydantic import Field


class Subschema1(BaseModel):
    subprop0: Optional[str] = Field(
        None, json_schema_extra={"custom_key": "custom_value_0"}
    )
    subprop1: Optional[str] = Field(
        None, json_schema_extra={"custom_key": "custom_value_1"}
    )


class Subschema2(BaseModel):
    subprop0: Optional[str] = Field(
        None, json_schema_extra={"custom_key": "custom_value_0"}
    )
    subprop2: Optional[str] = Field(
        None, json_schema_extra={"custom_key": "custom_value_2"}
    )


class Example(BaseModel):
    type: Optional[str] = ["example"]
    prop1: Optional[str] = Field(None, json_schema_extra={"custom_key": "custom_value"})
    prop2: Optional[Union[Subschema1, Subschema2]] = Field(
        None, json_schema_extra={"custom_key": "custom_value"}
    )

Version:

• OS: Win11
• Python version: 3.11
• datamodel-code-generator version: 0.28.2, 0.30.1

Additional context

The issue occures when serializing the schema here: https://github.com/koxudaxi/datamodel-code-generator/blob/a6acd5214a4eb823241a1b3f5d8d8a1564a595f2/src/datamodel_code_generator/parser/jsonschema.py#L642 and assigning it to a new constructure: https://github.com/koxudaxi/datamodel-code-generator/blob/a6acd5214a4eb823241a1b3f5d8d8a1564a595f2/src/datamodel_code_generator/parser/jsonschema.py#L652

=> The value of the extra field of the property schemas is lost (somehow it is stored under the key #-datamodel-code-generator-#-extras-#-special-# instead).

A workaround is to explicitely copy the extra attribute afterwards:

def parse_combined_schema(
        self,
        name: str,
        obj: JsonSchemaObject,
        path: list[str],
        target_attribute_name: str,
    ) -> list[DataType]:
        base_object = obj.dict(exclude={target_attribute_name}, exclude_unset=True, by_alias=True)
        # key "extra" is renamed to '#-datamodel-code-generator-#-extras-#-special-#' by alias export
        combined_schemas: list[JsonSchemaObject] = []
        refs = []
        for index, target_attribute in enumerate(getattr(obj, target_attribute_name, [])):
            if target_attribute.ref:
                combined_schemas.append(target_attribute)
                refs.append(index)
                # TODO: support partial ref
            else:
                # original code
                # combined_schemas.append(
                #     self.SCHEMA_OBJECT_TYPE.parse_obj(
                #         self._deep_merge(
                #             base_object,
                #             target_attribute.dict(exclude_unset=True, by_alias=True)
                #         )
                #     )
                # )

                # fixed code
                so =  self.SCHEMA_OBJECT_TYPE.parse_obj(
                    self._deep_merge(
                        base_object,
                        target_attribute.dict(exclude_unset=True, by_alias=True)
                    )
                )
                # copy the 'extra' attribute
                if hasattr(so, "properties") and hasattr(obj, "properties"):
                    for k, v in so.properties.items():
                        if k in obj.properties:
                            if obj.properties[k].extras:
                                v.extras = self._deep_merge(v.extras, obj.properties[k].extras)
                        x_of_properties = getattr(target_attribute, "properties", {})
                        if k in x_of_properties:
                            if x_of_properties[k].extras:
                                v.extras = self._deep_merge(v.extras, x_of_properties[k].extras)
                combined_schemas.append(so)

        parsed_schemas = self.parse_list_item(
            name,
            combined_schemas,
            path,
            obj,
            singular_name=False,
        )
        common_path_keyword = f"{target_attribute_name}Common"
        return [
            self._parse_object_common_part(
                name,
                obj,
                [*get_special_path(common_path_keyword, path), str(i)],
                ignore_duplicate_model=True,
                fields=[],
                base_classes=[d.reference],
                required=[],
            )
            if i in refs and d.reference
            else d
            for i, d in enumerate(parsed_schemas)
        ]

#2399 is weakly related, but #2400 does not fix this issue