pathwaycom/pathway · error · ValueError

`{column_name}` should be a column definition, found {type(c

Error message

`{column_name}` should be a column definition, found {type(column)}

What it means

When building a schema, each annotated column may only be assigned a column_definition(...) object. After dt.wrap of the annotation, the code pops the class-level field value and checks isinstance(column, ColumnDefinition); assigning anything else (a plain type, a tuple, a ColumnSchema, a default value directly) raises this ValueError.

Source

Thrown at python/pathway/internals/schema.py:208

    #  Update locals to handle recursive Schema definitions
    localns[schema.__name__] = schema
    annotations = get_type_hints(schema, localns=localns)
    fields = _cls_fields(schema)
    for base in bases:
        if not isinstance(base, SchemaMetaclass):
            continue
        for column_name, column_schema in base.__columns__.items():
            if column_name not in fields:
                fields[column_name] = column_schema.to_definition()

    columns = {}

    for column_name, annotation in annotations.items():
        col_dtype = dt.wrap(annotation)
        column = fields.pop(column_name, column_definition(dtype=col_dtype))

        if not isinstance(column, ColumnDefinition):
            raise ValueError(
                f"`{column_name}` should be a column definition, found {type(column)}"
            )

        dtype = column.dtype
        if dtype is None:
            dtype = col_dtype

        if col_dtype != dtype:
            raise TypeError(
                f"type annotation of column `{column_name}` does not match column definition"
            )

        column_name = column.name or column_name

        def _get_column_property(property_name: str, default: Any) -> Any:
            match (
                getattr(column, property_name),
                getattr(schema_properties, property_name),

View on GitHub (pinned to fa2f74a464)

Solutions

Wrap any per-column option in column_definition: age: int = pw.column_definition(default_value=42).
Do not assign raw types or values to annotated attributes; an annotation alone (age: int) is sufficient when no options are needed.
When deriving schemas programmatically, convert ColumnSchema back with .to_definition() before passing into schema_builder.

Example fix

# before
class MySchema(pw.Schema):
    age: int = 42

# after
class MySchema(pw.Schema):
    age: int = pw.column_definition(default_value=42)

Defensive patterns

Strategy: type-guard

Validate before calling

from pathway.internals.schema import column_definition, ColumnDefinition

def is_valid_column_value(v) -> bool:
    return v is None or isinstance(v, ColumnDefinition)

Type guard

from pathway.internals.schema import ColumnDefinition
from typing import TypeGuard

def is_column_definition(v) -> TypeGuard[ColumnDefinition]:
    return isinstance(v, ColumnDefinition)

Prevention

Only ever assign column_definition(...) (or nothing) to annotated schema attributes.
Use default_value= inside column_definition for defaults, never a bare value.

When it happens

Trigger: Writing class MySchema(pw.Schema): age: int = 42 or age: int = int (instead of column_definition(dtype=int, default_value=42)); also assigning a ColumnSchema object copied from another schema's columns() mapping.

Common situations: Trying to declare default values directly (pw.Schema requires column_definition(default_value=...)); migrating old ColumnSchema-based code; assigning dataclass-like sentinel values.

Related errors

AI-assisted analysis of pathwaycom/pathway@fa2f74a464 (2026-08-15). Data as JSON: /api/errors/8d777362ad79f221. Report an issue: GitHub.