pathwaycom/pathway · error · TypeError

Pathway does not support using reducer {self.name} on column

Error message

Pathway does not support using reducer {self.name} on column of type {arg_type}.

What it means

Tuple-producing reducers such as pathway.reducers.tuple (and ndarray variants) build a dt.List of the argument type, which only makes sense when the column is float-like, array-like, or tuple-like. In TupleConvertibleToNDArrayWrappingReducer.return_type, any other dtype (str, bool, Pointer, json...) fails the dtype_issubclass check against FLOAT/ANY_ARRAY/ANY_TUPLE and raises this TypeError.

Source

Thrown at python/pathway/internals/reducers.py:234

        if context.sort_by is not None:
            return (context.sort_by.to_column_expression(),)
        else:
            return ()


class TupleConvertibleToNDArrayWrappingReducer(TupleWrappingReducer):
    def return_type(
        self, arg_types: builtins.list[dt.DType], id_type: dt.DType
    ) -> dt.DType:
        arg_type = arg_types[0]
        if self._skip_nones:
            arg_type = dt.unoptionalize(arg_type)
        if builtins.any(
            dt.dtype_issubclass(arg_type, dtype)
            for dtype in [dt.FLOAT, dt.ANY_ARRAY, dt.ANY_TUPLE]
        ):
            return dt.List(arg_type)
        raise TypeError(
            f"Pathway does not support using reducer {self.name}"
            + f" on column of type {arg_type}.\n"
        )


class StatefulManyReducer(Reducer):
    name = "stateful_many"
    combine_many: api.CombineMany

    def __init__(self, combine_many: api.CombineMany):
        self.combine_many = combine_many

    def return_type(self, arg_types: list[dt.DType], id_type: dt.DType) -> dt.DType:
        return dt.ANY

    def engine_reducer(self, arg_types: list[dt.DType]) -> api.Reducer:
        return api.Reducer.stateful_many(self.combine_many)

View on GitHub (pinned to fa2f74a464)

Solutions

1. Verify the column dtype is float/int/array/tuple before using these reducers; cast if numeric-like data arrived as str.
2. For collecting arbitrary values (including strings), write a custom reducer with pw.custom_reducers.red_state_many or use a different aggregation approach.
3. If the column is Optional[numeric] and _skip_nones semantics matter, ensure the underlying type after unoptionalize is numeric/array/tuple.

Example fix

# before
agg = t.groupby(t.key).reduce(names=pw.reducers.tuple(t.name))  # name: str

# after
collect_names = pw.custom_reducers.red_state_many(
    lambda state, name: (state or []) + [name]
)
agg = t.groupby(t.key).reduce(names=collect_names(t.name))

from pathway.internals import dtype as dt

def column_accepts_tuple_reducer(dtype) -> bool:
    return any(
        dt.dtype_issubclass(dtype, d) for d in (dt.FLOAT, dt.ANY_ARRAY, dt.ANY_TUPLE)
    )

When it happens

Trigger: table.reduce(vals=pw.reducers.tuple(pw.this.name)) where name is a str column; or the numpy-oriented reducers (np.values etc.) applied to non-numeric columns. _skip_nones transparently unoptionalizes Optional[...] first, so Optional[str] also fails.

Common situations: Collecting string values per group with pw.reducers.tuple instead of a dedicated string aggregation; feeding json or pointer columns into tuple/np-style reducers.

Related errors

• Pathway does not support using reducer {self} on column of t
• Pathway does not support using binary operator {expression._
• Incompatible types in for a binary operator. The types are:
• {role} {col._name!r} must be of type str, got {col._column.d
• Cannot flatten column of type {dtype}.

AI-assisted analysis of pathwaycom/pathway@fa2f74a464 (2026-08-15). Data as JSON: /api/errors/24ae6b056e54bd15. Report an issue: GitHub.