Hey, ran into a weird error when using sensors, and optional inputs at the top level of a graph. The error linked below gets raised even tho an empty input should be valid. If the input to the graph is provided, the sensor works.

from typing import Optional
from dagster import Nothing, PythonObjectDagsterType, RunRequest, graph, In, op, sensor
from pathlib import Path

DagsterConfigurablePathType = PythonObjectDagsterType(python_type=(Path, Nothing))


@op()
def no_op(foo: Optional[Path]) -> None:
    print(foo)


@graph(ins={"foo": In(DagsterConfigurablePathType)})
def bar(foo: Optional[Path]) -> None:
    no_op(foo)


job = bar.to_job(name="job")


@sensor(job_name="job")
def foobar_sensor():
    # Following fails due to an error thrown here.  This is on 1.2.7 so maybe it was fixed since then.
    # <https://github.com/dagster-io/dagster/blob/1.2.7/python_modules/dagster/dagster/_core/execution/plan/plan.py#L437-L445>
    # DagsterConfigurablePathType.is_nothing evaluates to False
    # this is called downstream which causes the error to be raised.
    yield RunRequest(run_key="1", run_config={"inputs": {}})

Hi Jordan. I think the error message can definitely be confusing here--jobs cannot have unconnected inputs, so when a graph is converted to a job, input values have to be passed in:

job = bar.to_job(name="job", input_values={"foo": None})

Another option would be to specify these input values through config. Since you have a custom input type, you may have to define a default value on the input, and override those defaults via config.