GeoJSON object no longer receives a default type

[Jeffrey-Vervoort-KNMI]

After upgrading from 0.4.3 to 0.6.2 the validation raises an exception for type in FeatureCollection, Feature and Point, in our code but also for the Usage example in the readme.md.

At version 0.4.3 GeoJSON objects received a default see below:

class FeatureCollection(GenericModel, Generic[Geom, Props], GeoInterfaceMixin):
    """FeatureCollection Model"""
    type: str = Field("FeatureCollection", const=True)

The type field changed in 0.6.0 to a Literal and the default is removed:

class FeatureCollection(GenericModel, Generic[Geom, Props], GeoInterfaceMixin):
    """FeatureCollection Model"""
    type: Literal["FeatureCollection"]

When I read the spec: A GeoJSON object has a member with the name "type". The value of the member MUST be one of the GeoJSON types., I think that it should receive a default. If that would be the case then it could be changed to:

class FeatureCollection(GenericModel, Generic[Geom, Props], GeoInterfaceMixin):
    """FeatureCollection Model"""
    type: Literal["FeatureCollection"] = "FeatureCollection"

If this would be an issue, then it would be necessary to also change it for the other types: Feature and Point as well.

[vincentsarago]

Hi @Jeffrey-Vervoort-KNMI
Sorry for the frustration, we changed this knowing it will be quite a breaking change but we also wanted to make sure the library was following the specification to the letter. (https://github.com/developmentseed/geojson-pydantic/blob/main/CHANGELOG.md#060a0---2023-04-04)

ref: #92 (comment)

a Feature, FeatureCollection and Geometry dictionary should have a type. If not we cannot discriminate between them. We do think that this make the library more robust but we also understand that this was a big breaking change

also already discussed in #124

Thanks for pointing out the error in the Readme, It will be fixed

[alexismanin]

@vincentsarago I am not really convinced about that justification.

I understand that an input json that we want to decode must provide the type attribute. But on the Python class, that's a whole different story.

"type" is required in JSON, because the only way to define a "named type" (i.e. a class) in this format is by putting a property holding type name (in geojson, its the role of "type" attribute) in the object itself.

In Python, this information is held by the class name itself.

Alas, when instantiating a Python object, you give the type through its class name. Therefore, there is not point to repeat it.
Moreover, with a default value, type attribute is still present and mandatory, as Pydantic instantiate the field with the default value, so it is not missing.

Also, validation is expressed through Literal typing, that checks input value is the expected literal. So, even with a default value, type remains mandatory and constant: if you do Feature.model_validate({"type": "FeatureCollection"}), you will get a validation error for type object that has the wrong value, even if you define a default value. And when you do : Feature.model_validate({}), you also get a validation error because the overral structure does not match the Feature model.

In fact, the type attribute in geojson is here for polymorphism purpose, which is not applicable anyway with the current pydantic mapping, as each model defines independently a constant "type" attribute.

So, I think type should have a default/freezed value. It would:

• avoid redundancy, which, in the current case, is really very annoying
• Remain valid with the spec: any GeoJSON object will always have a type value set with the right value.
• It does not impact JSON (de)serialization in any way.