fix: handle unions properly

[guacs]

Pull Request Checklist

• New code has 100% test coverage
• (If applicable) The prose documentation has been updated to reflect the changes introduced by this PR
• (If applicable) The reference documentation has been updated to reflect the changes introduced by this PR
• Pre-Commit Checks were ran and passed
• Tests were ran and passed

Description

This PR fixes the handling of unions. The use of unwrap_args would result in the incorrect creation of children for the field meta. It would just return a random instance of the union and it fails completely in the case of unions of collection types like list[int] | list[str] where unwrap_args might return (int,) for type_args when in fact we want (list[int], list[str]).

This also fixes the parsing of the model type for batch_factory_type i.e. types like list[SomeModel]. Before, the model was taken from the value of type_args but that gives the incorrect value in the cases of unions like list[SomeModel] | int due to the change in the behavior of type_args. Instead we should just directly parse the model type from the unwrapped_type since the is_batch_factory_type is checked on the unwrapped_annotation anyway.

Close Issue(s)

• Fixes Bug: Broken union type generation since "polyfactory<2.6" (pydantic_factory) #366

[JacobCoffee]

Is this ready now?

[guacs]

Is this ready now?

Unfortunately, no.

[guacs]

I'm a bit stuck on this due to the way optional collection types are handled in pydantic v1.

from typing import Optional

from pydantic import BaseModel


OptionalList = Optional[list[str]]
OptionalStr = Optional[str]


class Foo(BaseModel):
    optional_list: OptionalList


optional_list_model_field = Foo.__fields__["optional_list"]
sub_field = optional_list_model_field.sub_fields[0]

print(sub_field.type)  # gives 'str' but we want 'list[str]'

Due to the way the handling of the unions is being changed in this PR, we want to get list[str] instead of str so that the children of the FieldMeta of the optional_list field will have a FieldMeta with the annotation list[str]. Since we are getting back str, the child FieldMeta ends up having the annotation str and the generation of the model instance fails.

[Leobouloc]

Hi. Is this still work in progress ? Would it be possible to fix this for Pydantic v2 as an intermediate measure ?

[guacs]

Hi. Is this still work in progress ? Would it be possible to fix this for Pydantic v2 as an intermediate measure ?

Yeah, this is still in progress due to pydantic v1 tests failing. I think it might be possible to do this only for v2, though ideally, I'd like it to be done in such a way that we don't have to maintain a separate get_field_value for just v1.

[guacs]

I'm closing this for now due to #366 being fixed in #468. However, these changes should still be made but it may have to wait until we drop support for pydantic v1.