What causes "definitions of columns {names} lack type annotation" in pathwaycom/pathway?

Add a type annotation to every column attribute: name: str = pw.column_definition(...).

pathwaycom/pathway · error · ValueError

definitions of columns {names} lack type annotation

Error message

definitions of columns {names} lack type annotation

What it means

After schema building consumes all annotated columns, any leftover entries in the fields dict are class attributes that were assigned a column_definition (or inherited one) but received no type annotation. Pathway requires every column definition to be paired with an annotation so the dtype is unambiguous, so leftovers raise this ValueError listing the offending names.

Source

Thrown at python/pathway/internals/schema.py:256

                            + f" value {schema_property!r} but column"
                            + f" `{column_name}` got {column_property!r}"
                        )
                    return column_property

        columns[column_name] = ColumnSchema(
            primary_key=column.primary_key,
            default_value=column.default_value,
            dtype=dt.wrap(dtype),
            name=column_name,
            append_only=_get_column_property("append_only", False),
            description=column.description,
            example=column.example,
            source_component=column.source_component,
        )

    if fields:
        names = ", ".join(fields.keys())
        raise ValueError(f"definitions of columns {names} lack type annotation")

    return columns


def _universe_properties(
    columns: list[ColumnSchema],
    schema_properties: SchemaProperties,
    dtype: dt.DType,
    append_only: bool | None = None,
) -> ColumnProperties:
    if append_only is None:
        append_only = False
        if len(columns) > 0:
            append_only = any(c.append_only for c in columns)
        elif schema_properties.append_only is not None:
            append_only = schema_properties.append_only
    return ColumnProperties(dtype=dtype, append_only=append_only)

View on GitHub (pinned to fa2f74a464)

Solutions

Add a type annotation to every column attribute: name: str = pw.column_definition(...).
If the attribute is not meant to be a column, rename it or remove the column_definition assignment.
When subclassing an existing schema, annotate overridden columns again.

Example fix

# before
class S(pw.Schema):
    name = pw.column_definition(dtype=str)

# after
class S(pw.Schema):
    name: str = pw.column_definition(dtype=str)

Defensive patterns

Strategy: type-guard

Validate before calling

import inspect
from pathway.internals.schema import ColumnDefinition

def all_definitions_annotated(cls) -> bool:
    ann = inspect.getannotations(cls)
    return all(
        name in ann
        for name, v in vars(cls).items()
        if isinstance(v, ColumnDefinition)
    )

Prevention

Always pair 'name: <type> = column_definition(...)' — annotation is mandatory.
Enable mypy with disallow_untyped_defs on schema modules to catch bare assignments.

When it happens

Trigger: class S(pw.Schema): name = pw.column_definition(dtype=str) — missing the 'name: str' annotation. Also inheriting __columns__ from a base schema but reassigning a definition without re-annotating.

Common situations: Writing attribute-only style (as in dataclasses without annotations); refactoring a schema and dropping annotations while keeping definitions.

Related errors

AI-assisted analysis of pathwaycom/pathway@fa2f74a464 (2026-08-15). Data as JSON: /api/errors/351a548082ba1b7b. Report an issue: GitHub.