Django JSONField with Pydantic schema support

json_schemed_field.py

from functools import partial

import pydantic
import logging

from django.contrib.postgres.fields import JSONField
from typing import Type, Union, Tuple

from django.core.serializers.json import DjangoJSONEncoder


logger = logging.getLogger(__name__)


def default_error_handler(obj, errors):
    logger.warning(
        'Can not parse stored object with schema obj=%s, errors=%s',
        obj, errors
    )
    return obj


class FieldToPythonSetter:
    """
    Forces Django to call to_python on fields when setting them.
    This is useful when you want to add some custom field data postprocessing.

    Should be added to field like a so:

    ```
    def contribute_to_class(self, cls, name, *args, **kwargs):
        super(JSONField, self).contribute_to_class(cls, name,  *args, **kwargs)
        setattr(cls, name, FieldToPythonSetter(self))
    ```
    """

    def __init__(self, field):
        self.field = field

    def __get__(self, obj, cls=None):
        return obj.__dict__[self.field.name]

    def __set__(self, obj, value):
        obj.__dict__[self.field.name] = self.field.to_python(value)


class JSONSchemedEncoder(DjangoJSONEncoder):

    def __init__(
        self,
        *args,
        schema: Union[Tuple[Type[pydantic.BaseModel]], Type[pydantic.BaseModel]],
        **kwargs
    ):
        if not isinstance(schema, tuple):
            self.schemas = (schema, )
        else:
            self.schemas = schema
        super().__init__(*args, **kwargs)

    def encode(self, obj):
        if not isinstance(obj, pydantic.BaseModel):
            # this flow used for expressions like .filter(data__contains={})
            # we don't want that {} to be parsed as schema
            return super().encode(obj)

        return obj.json()


class JSONSchemedDecoder:

    def __init__(
        self,
        schema: Union[Tuple[Type[pydantic.BaseModel]], Type[pydantic.BaseModel]],
        error_handler=default_error_handler,
    ):
        if not isinstance(schema, tuple):
            self.schemas = (schema, )
        else:
            self.schemas = schema
        self.error_handler = error_handler

    def decode(self, obj):

        if isinstance(obj, self.schemas):
            return obj

        errors = []
        for schema in self.schemas:
            try:
                return schema.parse_obj(obj)
            except pydantic.ValidationError as exc:
                errors.append((schema, exc.errors()))
            except TypeError as exc:
                errors.append((schema, str(exc)))
        
        return self.error_handler(obj, errors)


class JSONSchemedField(JSONField):

    def __init__(self, *args, schema=None, error_handler=default_error_handler, **kwargs):
        super().__init__(*args, **kwargs)

        self._schemas = self._populate_schemas(schema)
        self.decoder = JSONSchemedDecoder(schema=self._schemas, error_handler=error_handler)
        self.encoder = partial(JSONSchemedEncoder, schema=self._schemas)

    def deconstruct(self):
        name, path, args, kwargs = super().deconstruct()
        kwargs['schema'] = self._schemas
        return name, path, args, kwargs

    @staticmethod
    def _populate_schemas(schema) -> Tuple[Type[pydantic.BaseModel]]:
        assert schema is not None, 'Schema can not be None'

        if isinstance(schema, tuple):
            return schema

        if isinstance(schema, type) and issubclass(schema, pydantic.BaseModel):
            return schema,

        origin = getattr(schema, '__origin__', None)
        if origin is Union:
            for s in schema.__args__:
                assert issubclass(s, pydantic.BaseModel)

            return schema.__args__

        # only pydantic.BaseModel and typing.Union are supported
        raise AssertionError('Unsupported schema type: {0}'.format(type(schema)))

    def to_python(self, value):
        if value is None:
            return None

        return self.decoder.decode(value)

    def contribute_to_class(self, cls, name, *args, **kwargs):
        super(JSONField, self).contribute_to_class(cls, name,  *args, **kwargs)
        setattr(cls, name, FieldToPythonSetter(self))

Comments from original:

• about FieldToPythonSetter

as described in the documentation this hack is needed to convert from python to pydantic object when Django creates an instance of a model. When you call item = Item.objects.get(id=1) Django creates item instance and FieldToPythonSetter forces it to parse json content into pydantic schema, so you are able to use item.field as a pydantic schema immediately. Otherwise you have to do schema.parse_obj(item.field) by yourself (manually).

• another one

From what I know, querysets would always call either to_python or from_db_value so simply implementing these two methods (you can even make from_db_value just call to_python in most cases) would be enough.
I haven't tried that on your snippet yet, but that's what I did when I wrote a custom field that parses strings to enum instances.

BTW in order to support pydantic dataclasses you just need the following code in _populate_schemas:

if hasattr(schema, '__pydantic_model__'):
    return schema.__pydantic_model__,