Hypothesis fails with ForwardRefs and recursive pydantic models

[apirogov]

I would like to use hypothesis to generate instances of pydantic models, but it looks like it cannot handle recursive models with ForwardRefs.

(Quite) minimal example triggering the problem:

from pydantic import BaseModel
from typing import Optional, Union
from hypothesis import given, strategies as st

class Dummy(BaseModel):
    x: Optional[Union[int, Dummy]]

Dummy.update_forward_refs()


@given(st.builds(Dummy))
def test_dummy(obj):
    print(obj.json(indent=2))

When running this with pytest, it fails with:

E           hypothesis.errors.ResolutionFailed: Could not resolve ForwardRef('Optional[Union[int, Dummy]]') to a strategy; consider using register_type_strategy

.../site-packages/hypothesis/strategies/_internal/types.py:442: ResolutionFailed

Extra info:

• Python 3.8
• pydantic 1.10.2
• hypothesis 6.58
• pytest 7.0.1

[Zac-HD]

Thanks for the report - I cut this down further to:

import typing
from pydantic import BaseModel
from hypothesis.internal import compat

class Dummy(BaseModel):
    x: "typing.Union[int, Dummy]"

Dummy.update_forward_refs()
print(typing.get_type_hints(Dummy))
print(compat.get_type_hints(Dummy))

{'__slots__': typing.Tuple[str, ...], 'x': typing.Union[int, __main__.Dummy]}
{'__slots__': typing.Tuple[str, ...], 'x': ForwardRef('typing.Union[int, Dummy]'), 'data': typing.Any, 'return': <class 'NoneType'>}

digging even further, I think this is because Hypothesis prioritizes the annotations in a .__signature__ attribute over the results of typing.get_type_hints(). That still seems correct in general, but arguably we shouldn't do so when the new value is a ForwardRef and there's an existing non-forward-ref annotation.

[rsokl]

With #3521

from pydantic import BaseModel
from typing import Optional, Union
from hypothesis import given, strategies as st

class Dummy(BaseModel):
    x: Union[int, "Dummy"]

Dummy.update_forward_refs()


@given(st.builds(Dummy))
def test_dummy(obj):
    print(obj)

test_dummy()

x=0
x=22658
x=Dummy(x=0)
x=Dummy(x=16792)
x=Dummy(x=0)
x=0
x=0
x=2271479883078318998
x=Dummy(x=0)
x=Dummy(x=Dummy(x=0))
...

[apirogov]

Thanks for the quick fix! :)