Error when using from_type with optional integers with numeric constraints about hypothesis

[jobh]

...or even better, put this error message in Pydantic 🙂

[Zac-HD]

Seems fine to me: None can't be less than 42, so your type annotation is invalid!

Instead, try using Optional[Annotated[int, Le(42)]] - I'd guess this is how Field(default=None, le=42) comes out.

[jcklie]

Thank you for the quick reply. Both the pydantic and annotated types fail. I am not sure whether I understood your response. How should I write the pydantic model so that it works with hypothesis?

[Zac-HD]

How should I write the pydantic model so that it works with hypothesis?

Ah, apparently I was too optimistic about Pydantic semantics making sense matching the rest of Python.

Maybe using my suggested type annotation would do the trick? But I don't actually know, sorry, and pydantic/pydantic#4682 is ongoing but obviously doesn't solve this yet.

[jobh]

@jcklie The annotated type fails because it's written as Annotated[Optional] instead of Optional[Annotated], and of course the assert must also cover None. @Zac-HD's suggestion seems to work fine. But only without a field validator — if the field validator is added, it is applied to both types i.e. one substrategy is none().filter(functools.partial(ge, 42)), which doesn't make sense.

class Foo(BaseModel):
    x: Optional[Annotated[int, Le(42)]] = Field(default=None)


@given(st.from_type(Foo))
def test_foo_1(foo):
    assert foo.x is None or foo.x <= 42


def test_foo_2():
    Foo(x=None)
    Foo(x=42)
    with pytest.raises(ValidationError):
        Foo(x=43)

[jcklie]

All the examples I find on the internet using Annotated and Optional use the syntax of Annotated[Optional[x]], e.g. 1, 2, so I think at least both orders should be valid. The first link is even from cpython. Thus I still think this is a bug and both the pydantic and annotated types examples I gave should not throw. Could you maybe open this issue again at least to track this?

[jobh]

All the examples I find on the internet using Annotated and Optional use the syntax of Annotated[Optional[x]]

It depends on what should be annotated. I.e.: what should be <=42? It's the int. In example 1, what should be documented? It's the top level type i.e. the Optional[int].

But if pydantic accepts the swapped order, that's unfortunate and IMO too lenient... so yeah, possibly it would be nice to recognize this and either accept or have a customized error message.

[Zac-HD]

All the examples I find on the internet using Annotated and Optional use the syntax of Annotated[Optional[x]], e.g. 1, 2, so I think at least both orders should be valid.

The CPython example is adding metadata but not a type-specific constraint. The Pydantic example also doesn't have any constraints.

But even if there was precedent in other libraries, I would reject this idea for Hypothesis - special cases are the bane of user understanding.

(I'd be happy to accept a PR with a specific error message, but don't think that's worth a tracking issue)