[Feature Request]: Allow non-identifier column names in Python Row #40698

What would you like to happen?

What happened?

In the Beam Portability API (schema.proto) and the Java SDK (org.apache.beam.sdk.schemas.Schema / org.apache.beam.sdk.values.Row), schema field names can be arbitrary strings. For example, Calcite SQL (SqlTransform) routinely produces output schemas with column names such as:

• Unaliased expression names: EXPR$0, EXPR$1
• Quoted SQL column aliases containing spaces, punctuation, or leading digits: >120 days, 31-60 days, 91-120 days (e.g., TPC-DS Q62, Q99)
• Python reserved keywords: from, class, return

However, when a schema is decoded in the Python SDK (for instance, when receiving the output of cross-language SqlTransform or any portable RowCoder), SchemaTranslation.named_tuple_from_schema (sdks/python/apache_beam/typehints/schemas.py) attempts to construct a typing.NamedTuple (or dataclasses.make_dataclass when a field starts with _, thanks to a recent patch in #40542):

if any(name.startswith('_') for name, _ in subfields):
  user_type = dataclasses.make_dataclass(
      type_name,
      subfields,
      namespace={'_asdict': _dataclass_asdict},
      frozen=True)
else:
  user_type = NamedTuple(type_name, subfields)

Both typing.NamedTuple and dataclasses.make_dataclass enforce strict Python identifier rules (name.isidentifier() and not keyword.iskeyword(name)), causing RowCoder initialization to fail with:

ValueError: Type names and field names must be valid identifiers: 'EXPR$0'

or:

TypeError: Field names must be valid identifiers: '31-60 days'

Why is this a feature gap?

1. Cross-language incompatibility: A valid PCollection<Row> produced by a Java transform (such as SqlTransform or xlang I/O connectors) cannot be consumed in Python if any column name is not a strict Python identifier.
2. Incremental workarounds in named_tuple_from_schema: Previously, dataclasses.make_dataclass was added as a fallback specifically for leading-underscore column names like _change_type (CDC metadata), because NamedTuple rejected them. However, make_dataclass still rejects non-identifier characters ($, spaces, -, >), leading digits, dunder names, and Python keywords.
3. Python runtime already supports arbitrary attribute names: RowCoderImpl (sdks/python/apache_beam/coders/coder_impl.py) encodes and decodes fields using getattr(value, name) and positional construction self.constructor(*values). In Python, getattr and class namespaces (type(name, (tuple,), namespace) with property(operator.itemgetter(i))) work with arbitrary string keys, not just valid Python identifiers.

Steps to Reproduce

from apache_beam.coders import RowCoder
from apache_beam.portability.api import schema_pb2

schema = schema_pb2.Schema(
    id="sql_expr_schema",
    fields=[
        schema_pb2.Field(
            name="EXPR$0",
            type=schema_pb2.FieldType(atomic_type=schema_pb2.INT64)),
        schema_pb2.Field(
            name="31-60 days",
            type=schema_pb2.FieldType(atomic_type=schema_pb2.INT64)),
    ])
coder = RowCoder(schema)  # Raises ValueError: Type names and field names must be valid identifiers: 'EXPR$0'

Proposed Solution

In SchemaTranslation.named_tuple_from_schema (sdks/python/apache_beam/typehints/schemas.py), when any field name in subfields is not a valid Python identifier or is a Python keyword (i.e., rejected by both NamedTuple and make_dataclass), construct a dynamic tuple-backed row class via type(type_name, (tuple,), namespace) that:

• Sets _fields = tuple(name for name, _ in subfields) and __annotations__ = dict(subfields) so match_is_named_tuple() and RowTypeConstraint.from_user_type() continue to work seamlessly.
• Binds property(operator.itemgetter(i)) in the class dictionary for each field name (which Python allows for arbitrary string keys, so getattr(row, "EXPR$0") and getattr(row, "31-60 days") work in O(1) time).
• Implements __new__(cls, *args, **kwargs), _asdict(), and __repr__() matching NamedTuple semantics while preserving the exact portable schema field names.

Issue Priority

Priority: 2 (default / most feature requests should be filed as P2)

Issue Components

• Component: Python SDK
• Component: Java SDK
• Component: Go SDK
• Component: Typescript SDK
• Component: IO connector
• Component: Beam YAML
• Component: Beam examples
• Component: Beam playground
• Component: Beam katas
• Component: Website
• Component: Infrastructure
• Component: Spark Runner
• Component: Flink Runner
• Component: Prism Runner
• Component: Twister2 Runner
• Component: Hazelcast Jet Runner
• Component: Google Cloud Dataflow Runner