bug: Validators on an inherited field cause a bug

[awgymer]

Description of the bug

I believe if you create a validator method on a field called id that it will cause an error:

.../lib/python3.10/site-packages/griffe_pydantic/common.py", line 72, in <listcomp>
    targets = [cls.members[field] for field in fields]
KeyError: 'id'

Docs were building fine until I added the following validator to my model:

@field_validator("id", mode="before")
@classmethod
def convert_null_id_to_empty(cls, v: Any) -> Any:
    return v if v is not None else ""

Since the cls.members is coming from a griffe.Class object I'm not sure if this is a bug there rather than in this extension, but I suspect that it's much more likely to crop up here as typically you would avoid using id as an object attribute.

UPDATE: I loaded in my package using griffe in my interpreter and I found the real cause of the bug. It seems that the field doesn't exist on the Class object because I guess griffe does not include the inherited attributes from the base class (where id is defined)?

So I think MRE:

from typing import Any
from pydantic import BaseModel, field_validator

class MyBase(BaseModel):
    id: str


class AllowNones(MyBase):

    @field_validator("id", mode="before")
    @classmethod
    def convert_null_id_to_empty(cls, v: Any) -> Any:
        return v if v is not None else ""

I would guess a fix might be to use cls.all_members?

[pawamoy]

Thanks for the report @awgymer 🙂

I would guess a fix might be to use cls.all_members?

Yep, seems like the solution!

However! During dynamic analysis, common.process_function will run while the package is not entirely loaded. Using all_members would be bad: it computes inherited members, and if Griffe didn't yet see the base classes, some inherited members will be missing, and we'll likely get a key error again. If we want to use all_members safely, then we have to refactor the extension to use on_package_loaded for dynamic analysis as well. Which means the current hooks (on_class_instance, on_attribute_instance, on_function_instance, on_class_members) should only record the objects somewhere (in a dict, self._objects[obj.path] = obj). Then we would iterate on those in on_package_loaded. Something like that 😅

mkdocstrings/griffe#347 could alleviate the need to wait for the whole package to be loaded, but this is a consequent effort.

[pawamoy]

@awgymer I've pushed a fix on main branch, could you try it? I refactor the extension so I want to make sure I didn't break anything.

[awgymer]

Unfortunately I have moved on from the role where I was working on this and as such I don't have access to the codebase to test this on now.

The best I could do is create a small MRE and test it locally?

[pawamoy]

A MRE would be absolutely great @awgymer 🙇

[pawamoy]

I published v1.1.3, should be fixed 🙂