[BEAM-12136] Add deterministic coders for common Python types

• Type: Improvement
• Status: Open
• Priority: P3
• Resolution: Unresolved
• Component/s: sdk-py-core
• Labels: TFX+Beam types
• Assignee: Unassigned
• Reporter: Udi Meiri
• Created: 08/Apr/21 21:36
• Updated: 04/Jun/22 20:08

Description

FastPrimitivesCodeImpl has "fallback" deterministic encoding of certain types in encode_special_deterministic. This encoding is not as efficient as a type-specific coder (https://github.com/apache/beam/pull/14455).

This bug is to implement type-specific coders for the types supported in encode_special_deterministic. This method will still handle encoding for cases in which type information is not available, and the new coders will be used where type hints are available.

Types to implement coders for: enum.Enum, dataclasses.dataclass, collections.namedtuple / typing.NamedTuple (verify there's no conflict/duplication with RowCoder)

Issue Links

• links to: GitHub Pull Request #14455
• links to: GitHub Pull Request #14532