[damccorm]

Description

Optimizes bytecode return-type inference and schema type translation during pipeline graph construction:

1. Cache opcode dispatch tables in trivial_inference.infer_return_type_func (sdks/python/apache_beam/typehints/trivial_inference.py):
   • Previously, infer_return_type_func imported opcodes and intrinsic_one_ops and rebuilt the simple_ops dictionary ({k.upper(): v for k, v in opcodes.__dict__.items()}) on every call.
   • Hoists _SIMPLE_OPS and _INT_ONE_OPS to module-level cached tables initialized once on first use.
2. Fast-path primitive atomic types in SchemaTranslation.typing_to_runner_api (sdks/python/apache_beam/typehints/schemas.py):
   • Checks PRIMITIVE_TO_ATOMIC_TYPE at the top of typing_to_runner_api before inspecting generic origin/dataclass/NamedTuple metadata (1.26x faster schema field translation).

Testing

• Verified with pytest sdks/python/apache_beam/typehints/trivial_inference_test.py sdks/python/apache_beam/typehints/schemas_test.py.

[damccorm]

/gemini review

[github-actions]

📋 Review Summary

This PR introduces two clean, well-targeted performance optimizations for pipeline construction: caching the opcode dispatch tables (_SIMPLE_OPS and _INT_ONE_OPS) at the module level in trivial_inference.py, and fast-pathing primitive atomic types in SchemaTranslation.typing_to_runner_api. Both changes preserve existing semantics while avoiding redundant dictionary reconstructions and costly type introspection.

🔍 General Feedback

• In trivial_inference.py, hoisting _SIMPLE_OPS and _INT_ONE_OPS out of infer_return_type_func eliminates repeated module imports and opcodes.__dict__ iterations across recursive inference calls.
• In schemas.py, guarding the early PRIMITIVE_TO_ATOMIC_TYPE check with try ... except TypeError safely handles unhashable type representations (such as schema_pb2.Schema instances) while returning a fresh schema_pb2.FieldType proto for each call.

---

💡 Tip: Comment /gemini review on this pull request to trigger a new Gemini code review.