refactor: remodel DoapTarget

[jnussbaum]

No description provided.

[jnussbaum]

@BalduinLandolt This is my try-out of your idea how the DoapTarget could be remodeled. The result is pleasing, but I have a problem with the validation of nested pydantic models which makes the tests fail.

The problem is that inside the pydantic model Doap, I have a field target of type DoapTarget. Now, this type DoapTarget should be a protocol or ABC, because I want that the code uses one of the subtypes GroupDoapTarget or EntityDoapTarget.

The problem appears when reading serialized JSON data: The model_validator doesn't know what to do with the JSON data, because DoapTarget is nothing that pydantic can call.

I tried several solutions, but nothing worked. If you don't have an idea, I'll have to close this PR.

[BalduinLandolt]

The problem is that inside the pydantic model Doap, I have a field target of type DoapTarget. Now, this type DoapTarget should be a protocol or ABC, because I want that the code uses one of the subtypes GroupDoapTarget or EntityDoapTarget.

The problem appears when reading serialized JSON data: The model_validator doesn't know what to do with the JSON data, because DoapTarget is nothing that pydantic can call.

I tried several solutions, but nothing worked. If you don't have an idea, I'll have to close this PR.

Not sure if this would work and make things nice... but what about

class DoapTarget(BaseModel):
    project_iri: str
    target: GroupDoapTarget | EntityDoapTarget


class GroupDoapTarget(BaseModel):
    group: Group


class EntityDoapTarget(BaseModel):
    resource_class: str | None = None
    property: str | None = None

Or if that nesting turns out to be undesireable:

type DoapTarget = GroupDoapTarget | EntityDoapTarget


class GroupDoapTarget(BaseModel):
    project_iri: str
    group: Group


class EntityDoapTarget(BaseModel):
    project_iri: str
    resource_class: str | None = None
    property: str | None = None

But maybe it's just not possible to do this in a nice way, in python... in a more functional language, instead of a union type, an enum type could be used, which would make it much nicer. But Python's enums simply&sadly just don't work as types.

Edit: Another option may be extracting the deserialization to a factory method. But that would make the use of pydantic a bit pointless.

[jnussbaum]

The problem is that inside the pydantic model Doap, I have a field target of type DoapTarget. Now, this type DoapTarget should be a protocol or ABC, because I want that the code uses one of the subtypes GroupDoapTarget or EntityDoapTarget.
The problem appears when reading serialized JSON data: The model_validator doesn't know what to do with the JSON data, because DoapTarget is nothing that pydantic can call.
I tried several solutions, but nothing worked. If you don't have an idea, I'll have to close this PR.

Not sure if this would work and make things nice... but what about

class DoapTarget(BaseModel):
    project_iri: str
    target: GroupDoapTarget | EntityDoapTarget


class GroupDoapTarget(BaseModel):
    group: Group


class EntityDoapTarget(BaseModel):
    resource_class: str | None = None
    property: str | None = None

Or if that nesting turns out to be undesireable:

type DoapTarget = GroupDoapTarget | EntityDoapTarget


class GroupDoapTarget(BaseModel):
    project_iri: str
    group: Group


class EntityDoapTarget(BaseModel):
    project_iri: str
    resource_class: str | None = None
    property: str | None = None

But maybe it's just not possible to do this in a nice way, in python... in a more functional language, instead of a union type, an enum type could be used, which would make it much nicer. But Python's enums simply&sadly just don't work as types.

Edit: Another option may be extracting the deserialization to a factory method. But that would make the use of pydantic a bit pointless.

Thanks for your input. But the problem are not python's union types, I believe. Rather, when pydantic deserializes a JSON, and then encounters a field that is of type GroupDoapTarget | EntityDoapTarget, the poor pydantic is confused and doesn't know what to do: shall it deserialize this field as a GroupDoapTarget or as an EntityDoapTarget?

Perhaps it's possible to override the built-in model_validate or something like that, but I don't know pydantic well enough.

[BalduinLandolt]

The problem is that inside the pydantic model Doap, I have a field target of type DoapTarget. Now, this type DoapTarget should be a protocol or ABC, because I want that the code uses one of the subtypes GroupDoapTarget or EntityDoapTarget.
The problem appears when reading serialized JSON data: The model_validator doesn't know what to do with the JSON data, because DoapTarget is nothing that pydantic can call.
I tried several solutions, but nothing worked. If you don't have an idea, I'll have to close this PR.

Not sure if this would work and make things nice... but what about

class DoapTarget(BaseModel):
    project_iri: str
    target: GroupDoapTarget | EntityDoapTarget


class GroupDoapTarget(BaseModel):
    group: Group


class EntityDoapTarget(BaseModel):
    resource_class: str | None = None
    property: str | None = None

Or if that nesting turns out to be undesireable:

type DoapTarget = GroupDoapTarget | EntityDoapTarget


class GroupDoapTarget(BaseModel):
    project_iri: str
    group: Group


class EntityDoapTarget(BaseModel):
    project_iri: str
    resource_class: str | None = None
    property: str | None = None

But maybe it's just not possible to do this in a nice way, in python... in a more functional language, instead of a union type, an enum type could be used, which would make it much nicer. But Python's enums simply&sadly just don't work as types.
Edit: Another option may be extracting the deserialization to a factory method. But that would make the use of pydantic a bit pointless.

Thanks for your input. But the problem are not python's union types, I believe. Rather, when pydantic deserializes a JSON, and then encounters a field that is of type GroupDoapTarget | EntityDoapTarget, the poor pydantic is confused and doesn't know what to do: shall it deserialize this field as a GroupDoapTarget or as an EntityDoapTarget?

Perhaps it's possible to override the built-in model_validate or something like that, but I don't know pydantic well enough.

I rest my case, if proper sum types were a thing in Python, then Pydantic would know how to handle them... but that's splitting hairs.
In any case, I don't think it's worth fighting the type system over this kind of cosmetic modelling improvements.
By the way: Pydantic provides mechanisms to model sum types: https://blog.yossarian.net/2024/08/12/Approximating-sum-types-in-Python-with-Pydantic & https://docs.pydantic.dev/latest/concepts/unions/ but I don't think these provide great mechanisms compared to languages that provide fist class support for sum types.

[jnussbaum]

I rest my case, if proper sum types were a thing in Python, then Pydantic would know how to handle them... but that's splitting hairs. In any case, I don't think it's worth fighting the type system over this kind of cosmetic modelling improvements. By the way: Pydantic provides mechanisms to model sum types: https://blog.yossarian.net/2024/08/12/Approximating-sum-types-in-Python-with-Pydantic & https://docs.pydantic.dev/latest/concepts/unions/ but I don't think these provide great mechanisms compared to languages that provide fist class support for sum types.

Thanks for your links, thanks to them I could finally solve it. Now it works :-)

[BalduinLandolt]

See my comments, otherwise I think it looks pretty clean, right?