Python: fix: preserve null tool arguments

[he-yufeng]

Fixes #5934.

This keeps explicitly provided null tool arguments when the inferred Pydantic input model validates the payload. The old model_dump(exclude_none=True) path removed those keys before schema validation and invocation, so a required-but-nullable argument looked missing.

The helper still omits default None fields, and it only restores fields known to the validated model so unknown null keys are not reintroduced after validation.

To verify

• uv run --no-sync pytest python\packages\core\tests\core\test_tools.py::test_tool_invoke_preserves_required_nullable_argument python\packages\core\tests\core\test_tools.py::test_tool_invoke_does_not_reintroduce_unknown_null_arguments python\packages\core\tests\core\test_function_invocation_logic.py::test_base_client_with_function_calling_preserves_null_arguments -q
• uv run --no-sync ruff check agent_framework\_tools.py tests\core\test_tools.py tests\core\test_function_invocation_logic.py
• uv run --no-sync mypy agent_framework\_tools.py
• python -m py_compile python\packages\core\agent_framework\_tools.py python\packages\core\tests\core\test_tools.py python\packages\core\tests\core\test_function_invocation_logic.py
• git diff --check

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

This PR fixes tool argument serialization so explicitly provided null/None values for required-but-nullable parameters are preserved, while unknown null arguments are not reintroduced—covering both direct FunctionTool.invoke calls and the auto function-calling flow.

Changes:

• Added _dump_tool_arguments to preserve explicitly provided None fields when dumping validated Pydantic models.
• Updated tool invocation and auto-invocation flows to use _dump_tool_arguments instead of model_dump(exclude_none=True).
• Added tests to validate preservation of nullable required args and dropping of unknown null args, including function-calling scenarios.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
python/packages/core/agent_framework/_tools.py	Adds helper to dump tool args while preserving explicit None, and wires it into invoke/auto-invoke paths.
python/packages/core/tests/core/test_tools.py	Adds tests for direct tool invocation behavior with nullable required args and unknown null args.
python/packages/core/tests/core/test_function_invocation_logic.py	Adds test ensuring base client function-calling preserves null arguments through execution.

[moonbox3]

What happens with an aliased field on the mapping path? include_none_from there is the raw caller keys, but field_names / getattr use Python field names. With a custom input_model using Field(alias=...) + populate_by_name, an explicit null sent under the alias key fails name in field_names, so the None is dropped, re-introducing the exact bug this fixes (aliased models only; auto-generated @tool models are unaffected). Could we resolve alias -> field name first? One option, which also folds in the non-str-key guard:

Suggested change

	field_names = type(model).model_fields
	field_names = type(model).model_fields
	alias_to_field = {f.alias: n for n, f in field_names.items() if isinstance(f.alias, str)}
	for name in include_none_from if include_none_from is not None else model.model_fields_set:
	if not isinstance(name, str):
	continue
	resolved = alias_to_field.get(name, name)
	if resolved in field_names and getattr(model, resolved, None) is None:
	arguments[resolved] = None

[moonbox3]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/core/agent_framework
_tools.py	1021	79	92%	219–220, 404, 406, 419, 444–446, 454, 472, 486, 493, 500, 523, 525, 532, 540, 660, 695–697, 700–702, 704, 710, 762–764, 789, 815, 819, 857–859, 863, 885, 1028–1029, 1033, 1069, 1081, 1083, 1088–1091, 1112, 1116, 1120, 1134–1136, 1483, 1569, 1597, 1619, 1623, 1753, 1757, 1803, 1864–1865, 1968, 2021, 2041, 2043, 2099, 2162, 2334–2335, 2355, 2411–2412, 2550–2551, 2618, 2623, 2630
TOTAL	36843	4339	88%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
7357	34 💤	0 ❌	0 🔥	1m 51s ⏱️

[moonbox3]

Please make sure you're not duplicating work. #5944 was the first PR introduced with the fix.