Deserialization using the default deserializer

[pchanial]

After adding a deserializer to a class, the default deserializer is not accessible anymore. By default deserializer, I mean the one that can be used before adding a deserializer:

import pytest
from apischema import ValidationError, deserialize, deserializer, schema
from apischema.conversions import Conversion, catch_value_error

@schema(min=1, max=2)
class Foo(int):
    pass

deserializer(Conversion(catch_value_error(Foo), source=str, target=Foo))
with pytest.raises(ValidationError) as exc:
    deserialize(Foo, 1)
assert exc.value.messages == ['expected type string, found integer']

When Foo derives from int, I can make it work by adding the following deserializer:

deserializer(Conversion(catch_value_error(Foo), source=int, target=Foo))

and the str deserializer will check the constraints.

But what to do for dataclasses and still have the constraints checked by apischema ? The following fails since the default deserializer with dict is no more available:

from dataclasses import dataclass
import pytest
from apischema import ValidationError, deserialize, deserializer, schema, validator
from apischema.conversions import Conversion, catch_value_error

@dataclass
class Foo:
    value: int

    @validator
    def check_value(self):
        if self.value != 42:
            raise ValidationError('Invalid value')

@deserializer
def from_int(value: int) -> Foo:
    return deserialize(Foo, {'value': value})

deserialize(Foo, 42)

So, shouldn't the default deserializer be kept ?

[wyfo]

I'm not sure to understand your question.

Default deserializer is not kept when a conversion is registered, because there must be a way to remove it, so I've chosen this design.
However, it's possible to ignore registered conversion and fall back on default deserializer, as explained in the documentation. Also, identity conversion can be registered itself in order to add the default deserializer back.

Does it solve your issue?