Skip to content

Instantly share code, notes, and snippets.

@Bahus
Last active September 19, 2024 23:53
Show Gist options
  • Save Bahus/98a9848b1f8e2dcd986bf9f05dbf9c65 to your computer and use it in GitHub Desktop.
Save Bahus/98a9848b1f8e2dcd986bf9f05dbf9c65 to your computer and use it in GitHub Desktop.
Django JSONField with Pydantic schema support
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))
@wwarne
Copy link

wwarne commented Dec 3, 2020

@Bahus 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.

I've just tried this and realised that we need FieldToPythonSetter to work with instances because from_db_value is called only when the data is loaded from the database.

Imagine we have the next models

class PydanticModel(BaseModel):
    url: str
    title: str


class Item(models.Model):
    ...
    data: PydanticModel = JSONSchemedField(schema=PydanticModel)
my_item = Item.objects.first()  #  from_db_value will be called if we override it
>>> print(type(my_item.data))
<class 'PydanticModel'>
my_item.data = {'url': 'http://example.com', 'title': 'test'}

# if we don't set FieldToPythonSetter descriptor
>>> print(type(my_item.data ))
<class 'dict'>  # without FieldToPythonSetter

>>> print(type(my_item.data ))
<class 'PydanticModel'> # with FieldToPythonSetter

So FieldToPythonSetter is a very nice touch for a consistency of usage.

@Bahus - thank a lot for this code, very interesting stuff.

@hassanselim0
Copy link

@wwarne Thank you for the explanation, I haven't actually considered the case of assigning a dict instead of a pydantic model instance (your type-checker wouldn't like it), but you could design it as a valid assignment for simplicity.
Incidentally I've just finished making my own version of this custom field, I toyed around with Generics and made it possible to define an ORM Model Field like this: data = PydanticField[SomeData]('Data') (it also supports List[SomeData] as the generic type), and implements __get__ only if TYPE_CHECKING so you get nice auto-complete for it.
I'll still see how this all plays out in django admin, and if all works well I'll probably post my own gist for it 😁

@kuldeepk
Copy link

kuldeepk commented Feb 11, 2021

Thanks for this snippet @Bahus! Super helpful.

Everything works perfectly but in Django Admin i get error as Object of type PydanticModel is not JSON serializable. Any way to solve this? Am I doing anything wrong?

@hassanselim0
Copy link

hassanselim0 commented Feb 22, 2021

@kuldeepk I'm not sure what exactly is causing your error, but when I made my implementation I noticed that I need a custom django forms field to make django admin work properly, and in the orm field I've overridden the formfield function to return my custom forms field class.

Here is the custom forms field:

from django.contrib.postgres.forms import JSONField as JSONFormField

class PydanticFormField(JSONFormField):
    def _serialize(self, value):
        if isinstance(value, list):
            value = [(self._serialize(v) if isinstance(v, BaseModel) else v) for v in value]
        elif isinstance(value, BaseModel):
            value = value.dict(exclude_unset=True)
            if '__root__' in value:
                value = value['__root__']
            for k, v in value.items():
                if isinstance(v, Decimal):
                    value[k] = str(v)
        return value

    def prepare_value(self, value):
        value = self._serialize(value)
        return json.dumps(value, indent=2)

    def has_changed(self, initial, data):
        initial = self._serialize(initial)
        return super().has_changed(initial, data)

Note that this snippet also consider lists of pydantic models and have special handling for Decimal type, you can remove these bits if you want.

@srivastavaharsh9888
Copy link

@Bahus
I tried to implement the same, but while saving the model I was facing error like this -:
list_of_stages=[ABCModel(uuid=UUID('1102f63e-b7db-11eb-a149-1f45af5c49e5'), name='asdas', type='JO')] must be of type dict/list
Can anyone help me out, that how to handle .save() method while using this.

@abriemme
Copy link

I had issue with this gist (pydantic v2), so i found a simpler approach by using getter & setter:

class PydanticModel(BaseModel):
    url: str
    title: str


class Item(models.Model):
    _data: PydanticModel = JSONField(default=dict)

   @property
   def data(self) -> PydanticModel:
        return PydanticModel(**self._data)

   @data.setter
   def data(self, value: PydanticModel):
       self._data = value.model_dump(mode="json")

@hbd
Copy link

hbd commented May 30, 2024

@abriemme great solution. Big thanks to everyone here for the ideas

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment