Keep floats in int and float unions when decoding typed JSON payloads

[LindseyZ1205]

value_to_type tries the members of a union in order, and the int conversion accepts a float and truncates it. So with a type hint such as int | float, a float is decoded as an int:

from dataclasses import dataclass

from temporalio.converter import JSONPlainPayloadConverter


@dataclass
class Price:
    amount: int | float


converter = JSONPlainPayloadConverter()
converter.from_payload(converter.to_payload(Price(9.99)), Price)
# Price(amount=9)

The same happens for int | float | None, list[int | float] and dict values. When the union has a float member and the value is a float, convert it as a float before trying the other members. A plain int hint still truncates, as ok(int, 5.5, 5) in test_json_type_hints expects, and an int value in an int | float union is still an int.

I found this by reading value_to_type, not in a production workflow.

Validation on Python 3.12.13:

• The new cases in test_json_type_hints fail before the fix (assert 5 == 5.5) and pass after it.
• tests/test_converter.py and tests/test_common.py: 49 tests pass.
• Ruff import/format checks, Pyright, BasedPyright and mypy report no issues for the changed files; pydocstyle passes for the changed source file.
• Tested with the current Python source and the unmodified Rust bridge from the temporalio 1.33.0 wheel. I did not run server-backed tests, the full repository lint, or other Python versions.

AI disclosure: I used Claude Code to find this bug and write the fix and the tests. I reviewed the change and ran the tests locally.

🤖 Generated with Claude Code

[tconley1428]

Generally we would recommend using the pydantic_data_converter if you hit the limitations of the built in one, which will never be fully robust. Given that this could potentially be backwards incompatible, and that there isn't a real use case being blocked, closing.