Fix FunctionInputOutputTypeChecker crash on list[int] | None annotations

[breken-ai]

With lifecycle.FunctionInputOutputTypeChecker enabled, any node whose input or return annotation is a PEP 604 union with a parameterized member, such as list[int] | None, crashes the run. This happens even when the value is correct.

def evens(n: int) -> list[int] | None:
    return [i * 2 for i in range(n)] if n else None

dr = (
    driver.Builder()
    .with_modules(my_module)
    .with_adapters(lifecycle.FunctionInputOutputTypeChecker())
    .build()
)
dr.execute(["evens"], inputs={"n": 3})
# on main (Python 3.10-3.13):
#   File "hamilton/htypes.py", in check_instance
#     return isinstance(obj, type_)
# TypeError: isinstance() argument 2 cannot be a parameterized generic

htypes.check_instance detects unions through type_.__origin__ is Union. A types.UnionType (the X | Y form) has no __origin__ on Python 3.10-3.13, so it skips the union branch and reaches isinstance(obj, type_). isinstance raises when a union member is a parameterized generic. Optional[list[int]] works because it is a typing.Union, so only the | spelling crashes. On 3.14, X | Y and Union[X, Y] are the same type, so it no longer crashes there.

Changes

• hamilton/htypes.py: in check_instance, check each member of a types.UnionType recursively, the same way typing.Union is already handled.
• tests/test_type_utils.py: test_check_instance_with_pep604_union_of_generics. It covers matching and non-matching values for list[int] | None and dict[str, int] | list[int].
• tests/lifecycle/test_default.py: two driver tests with FunctionInputOutputTypeChecker. The first uses list[int] | None as a node's return type and as a downstream input, for both a list and None. The second checks that a wrong result (["not", "ints"]) is still rejected with the adapter's own TypeError.

How I tested this

• On unmodified main (4f4c48fe, Python 3.12): all 3 new tests fail with TypeError: isinstance() argument 2 cannot be a parameterized generic.
• With the fix, all 3 pass. They also pass on Python 3.10 and 3.14.
• pytest tests/test_type_utils.py tests/test_hamilton_driver.py tests/test_end_to_end.py tests/test_async_driver.py tests/test_graph.py tests/function_modifiers tests/test_function_modifiers.py tests/lifecycle tests/test_base.py tests/execution tests/io tests/caching tests/test_default_data_quality.py tests/test_node.py tests/test_ad_hoc_utils.py tests/test_parallel_graceful.py on Python 3.12: 1185 passed, 42 failed. The same 42 fail on main without this change (1182 passed): the test_graph.py display tests need graphviz, and the test_fingerprinting.py hash tests fail in my local env. None of them touch this code.
• ruff check and ruff format --check (0.16.9, the pre-commit pin) are clean. git diff --check is clean.

Notes

A companion PR, #1735, fixes htypes.check_input_type rejecting dict inputs for Mapping[...] parameters. The two PRs touch different functions and do not conflict.

Checklist

• PR has an informative and human-readable title (this will be pulled into the release notes)
• Changes are limited to a single goal (no scope creep)
• Code passed the pre-commit check & code is left cleaner/nicer than when first encountered.
• Any change in functionality is tested
• New functions are documented (with a description, list of inputs, and expected output) — no new functions
• Placeholder code is flagged / future TODOs are captured in comments — none added
• Project documentation has been updated if adding/changing functionality — no documented behavior changes

---

AI disclosure (per the ASF Generative Tooling guidance): this change was written with an AI coding tool (Claude Code, Claude Opus 5.5) working through the breken-ai account. The tool found the bug, wrote the fix and the tests, and wrote this description. The commit carries a Generated-by: trailer. The diff is a small original change to existing Hamilton code and includes no third-party material. The red/green runs above are real and can be re-run from the diff. If you would rather not take AI-assisted contributions here, say so and I will close this.

🤖 Generated with Claude Code