pathwaycom/pathway · error · TypeError

Pathway doesn't support casting {source_type} to {target_typ

Error message

Pathway doesn't support casting {source_type} to {target_type}.

What it means

The cast expression evaluator first tries cheap paths (no-op casts, equivalent dtypes, NONE into Optional, anything into Any), then consults the registered cast-operator mapping. If no cast handler exists between the source and target dtype, it raises this TypeError: Pathway simply does not support that conversion via .cast().

Source

Thrown at python/pathway/internals/graph_runner/expression_evaluator.py:575

        target_type = expression._return_type
        source_type = dt.normalize_pointers(source_type)
        target_type = dt.normalize_pointers(target_type)

        if (
            dt.dtype_equivalence(target_type, source_type)
            or dt.dtype_equivalence(dt.Optional(source_type), target_type)
            or (source_type == dt.NONE and isinstance(target_type, dt.Optional))
            or target_type == dt.ANY
        ):
            return arg  # then cast is noop
        if (
            result_expression := get_cast_operators_mapping(
                arg, source_type, target_type
            )
        ) is not None:
            return result_expression

        raise TypeError(
            f"Pathway doesn't support casting {source_type} to {target_type}."
        )

    def eval_convert(
        self,
        expression: expr.ConvertExpression,
        eval_state: RowwiseEvalState | None = None,
    ):
        expr = self.eval_expression(expression._expr, eval_state=eval_state)
        default = self.eval_expression(expression._default, eval_state=eval_state)
        unwrap = expression._unwrap
        source_type = expression._expr._dtype
        target_type = expression._return_type

        if (
            dt.dtype_equivalence(target_type, source_type)
            or dt.dtype_equivalence(dt.Optional(source_type), target_type)
            or (source_type == dt.NONE and isinstance(target_type, dt.Optional))

View on GitHub (pinned to fa2f74a464)

Solutions

For Json/Any sources use Json accessors or conversion expressions (pw.Json.as_int(), dt parsing helpers) instead of .cast()
Convert via an intermediate supported step: first cast to a compatible dtype that is registered (e.g. int->float), or format datetimes with dt methods rather than casting
Fix the schema at ingestion so the expensive cast is unnecessary; use Optional only where the engine supports the inner cast

Example fix

// before
value = pw.this.payload.cast(int)  # payload is pw.Json
// after
value = pw.this.payload.as_int()

Defensive patterns

Strategy: type-guard

Validate before calling

KNOWN_CASTS = {(int, float), (float, int), (str, str), (int, str), (float, str)}

def cast_supported(src, dst) -> bool:
    return (src, dst) in KNOWN_CASTS

Type guard

def is_directly_castable(src_dtype, dst_dtype) -> bool:
    # conservative check: only primitives of the same family or identical dtypes
    return src_dtype == dst_dtype or {str(src_dtype), str(dst_dtype)} <= {'int', 'float'}

Try / catch

try:
    pw.this.col.cast(target)
except TypeError as e:
    if "support casting" in str(e):
        # fall back to a UDF conversion or Json accessor
        ...

Prevention

Do not treat .cast() as python's int()/str(); use it only for registered pairs
Convert Json fields with accessors (as_int/as_float) instead of casting
Fix types at the schema/connector level so casts are rarely needed

When it happens

Trigger: pw.this.col.cast(str) where col is a Pointer or Json; casting Any/Json directly to a concrete dtype (unsupported — Json needs its own accessors); casting between unrelated types like bool->str, bytes->int; casting DateTimeNaive<->DateTimeUtc; casting to Optional-of-unsupported inner type.

Common situations: Assuming .cast() behaves like str()/int() in python and works between anything; coercing Json fields by casting instead of using Json accessors; cleaning up badly-typed connector schemas with blanket casts.

Related errors

AI-assisted analysis of pathwaycom/pathway@fa2f74a464 (2026-08-15). Data as JSON: /api/errors/86f3a72df2709aca. Report an issue: GitHub.