Datetime underflow causes validation to fail

[hamilton-earthscope]

Hi Pydantic team, thanks for the great library!!

I'm reporting a bug that looks to me like an underflow issue when validating datetimes are greater than a given datetime.

Here's a reproducible example. I'm using Pydantic v2.6.4. Which is the latest at the time of writing.

import datetime as dt

from annotated_types import Ge, Lt
from pydantic import AwareDatetime, BaseModel, ValidationError

MyDatetime = Annotated[
    AwareDatetime,
    Ge(dt.datetime(1900, 1, 1, tzinfo=dt.timezone.utc)),
]

class TestModel(BaseModel):
    t: MyDatetime


@pytest.mark.parametrize(
    "t, exp_throws",
    [
        ("1500-01-01T00:00:00+00:00", True), # doesn't throw!!!
        ("1600-01-01T00:00:00+00:00", True),
        ("1899-12-31T23:59:59.999999+00:00", True),
        ("1900-01-01T00:00:00+00:00", False),
        ("2024-12-31T00:00:00+00:00", False),
    ],
)
def test_time_constraints(t: str, exp_throws: bool):
    if exp_throws:
        with pytest.raises(ValidationError) as e:
            TestModel(t=t)
        assert "greater than or equal" in str(e)
    else:
        m = TestModel(t=t)
        assert m.t.isoformat() == t

The reason I think it's an underflow issue is that if I change my model to

MyDatetime = Annotated[
    AwareDatetime,
    Ge(dt.datetime(1900, 1, 1, tzinfo=dt.timezone.utc)),
    Lt(dt.datetime(2030, 1, 1, tzinfo=dt.timezone.utc)),
]

Then I get a validation error for 1500-01-01T00:00:00Z that it should be less than 2030-01-01T00:00:00Z.

I reproduced the same thing by constructing the schema directly.

class MyDatetime:
    """A datetime that requires timezone info and a sane year."""

    @classmethod
    def __get_pydantic_core_schema__(
        cls, source: type[Any], handler: GetCoreSchemaHandler
    ) -> core_schema.CoreSchema:
        schema_args = {
            "tz_constraint": "aware",
            "ge": dt.datetime(1900, 1, 1, tzinfo=dt.timezone.utc),
            "lt": dt.datetime(2030, 1, 1, tzinfo=dt.timezone.utc),
        }
        if cls is source:
            return core_schema.datetime_schema(**schema_args)
        else:
            schema = handler(source)
            schema.update(schema_args)
            return schema

    def __repr__(self) -> str:
        return "MyDatetime"

I can work around this by using an AfterValidator and performing the Ge/Lt validation in python-land.

Cheers.

[davidhewitt]

Agreed, I think a duplicate of pydantic/pydantic#8570, which has a proposed fix by erroring in pydantic/speedate#55 but that got stuck after I suggested instead fixing the underflow. I see you also agree with my assessment that fixing the underflow is the right solution 😂