-
Notifications
You must be signed in to change notification settings - Fork 449
Description
Constructing expression predicates like EqualTo("col", value) produces mypy errors because the Pydantic model field declarations use UnboundTerm as the type for term, and without the pydantic mypy plugin active, mypy generates __init__ signatures from the field types rather than from the explicit __init__ overrides.
Example:
from pyiceberg import expressions as ice
expr = ice.EqualTo("my_column", 42)mypy output (v1.19.1, pyiceberg 0.11.0, no pydantic mypy plugin):
error: Argument 1 to "EqualTo" has incompatible type "str"; expected "UnboundTerm" [arg-type]
Root cause:
The base classes define explicit __init__ methods that accept str | UnboundTerm:
UnboundPredicate.__init__:def __init__(self, term: str | UnboundTerm, **kwargs: Any) -> NoneLiteralPredicate.__init__:def __init__(self, term: str | UnboundTerm, literal: Any | None = None, **kwargs: Any) -> NoneUnaryPredicate.__init__:def __init__(self, term: str | UnboundTerm, **_: Any) -> None
But since these classes extend IcebergBaseModel (Pydantic BaseModel), mypy ignores the explicit __init__ and instead generates the constructor signature from the field declaration term: UnboundTerm, which does not include str.
This affects all predicate constructors: EqualTo, NotEqualTo, GreaterThan, GreaterThanOrEqual, LessThan, LessThanOrEqual, IsNull, NotNull, IsNaN, NotNaN, In, NotIn, StartsWith, NotStartsWith.
Workaround:
Downstream projects can either enable the pydantic mypy plugin or suppress type checking for the affected call sites.
Possible fix:
Widen the field type annotation on UnboundPredicate from term: UnboundTerm to term: str | UnboundTerm with a validator that coerces str to UnboundTerm. This way the type seen by mypy (from the field declaration) matches what the constructor actually accepts at runtime, regardless of whether the pydantic mypy plugin is enabled.