[Daniel Peng (original4422)]

Summary

Fixes #14642.

 native argument parsing
   revalidate Pydantic models
   recursively parse lists
+  preserve existing custom-class instances (including subclasses)
   construct other values

This avoids constructor errors for two-field dataclasses and silent nesting for one-field dataclasses. Built-in coercion, dictionary construction and Pydantic revalidation stay intact.

Evidence

• Before: 8 new public Kernel.invoke regression cases fail on main, covering ordinary/dataclass instances, subclasses, repeated calls, mixed lists and function-to-function values.
  After: python -m pytest tests/unit/functions tests/unit/kernel — 306 passed, 0 skipped on macOS/Python 3.12.
• Pydantic revalidate_instances="always", invalid-instance rejection and built-in integer coercion remain covered.
• pre-commit run --config python/.pre-commit-config.yaml --all-files — all 13 hooks pass, including Ruff check/format. The repository-locked mypy 1.19.1 reports no issues in all 556 package source files.
• Existing deprecation/OpenAI cleanup warnings remain in the unit run; no model or service integration tests were run.

Merge Danger

Door: two-way

Blast Radius: argument-parsing

Limited to preserving custom instances already accepted by their parameter type; no API signatures or generic/union handling change.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Daniel Peng (original4422)]

Jose Alvarez (@jpalvarezl), when convenient, could you review this Python fix for already-typed native function arguments? It preserves custom instances and subclasses while retaining Pydantic revalidation and built-in coercion. The local functions/kernel tests (306) and all 13 pre-commit hooks pass. Thank you!

[Daniel Peng (original4422)]

Eduard van Valkenburg (@eavanvalkenburg), when you have a chance, could you review this Python fix for preserving already-typed custom instances passed through KernelArguments to native functions? It also covers subclasses, lists and function-to-function values while retaining Pydantic revalidation and built-in coercion. Thanks for taking a look!