Schema generation failure when using future annotations import

[gfletcherMerc]

Whilst experimenting with some Pydantic forward annotations I came across a case which I believe causes py-avro-schema to generate an invalid Avro schema when using base & derived Pydantic models.

Consider a project with 2 files within a Python module folder.

base.py

from __future__ import annotations # inclusion of this line causes a failure
from pydantic import BaseModel
import py_avro_schema as pas

class MyBaseClass(BaseModel):
    name: str = ""
    @classmethod
    def get_avro_schema_json(cls) -> bytes:
        return pas.generate(cls)

derived.py

from .base import MyBaseClass
class MyDerivedClass(MyBaseClass):
    pass

If the JSON representation of the Avro schema is then retrieved using MyDerivedClass.get_avro_schema_json() I see the results:
{"type":"record","name":"MyDerivedClass","fields":[{"name":"name","type":"str","default":""}],"namespace":"Common","doc":"Usage docs: https://docs.pydantic.dev/2.9/concepts/models/"}

If avro.schema.parse() is then used on this, it fails due to the type of the name field being str. If the from __future__ import annotations is removed/commented from base.py, the schema generation proceeds as expected.

(My environment is Python 3.12.0 on Windows, avro==1.12.0, py-avro-schema==3.8.2)

[faph]

@gfletcherMerc thanks for raising! Are you able to reproduce this with a unit test at all?

[gfletcherMerc]

@faph just to confirm, you'd like me to write another test case (which I expect would currently fail) to illustrate the problem? Within test_pydantic.py I presume? If so then yeah, I'll get on it sometime soon.

(BTW - I couldn't decide if this problem is "caused" by py-avro-schema or whether there was some Pydantic behaviour/issue which leads to the problem I observed. Hopefully that will become clearer as this progresses).

[faph]

@gfletcherMerc yes with a minimal failing unit test it’s easiest to discover where the issue lies.