Preserve positional-only parameters in curry signatures

[FanWu-ai]

Problem

Keywords whose names coincide with positional-only parameters belong to **kwargs. curry.__signature__ currently treats those keywords as bindings for the positional-only parameters instead. This changes their defaults and kinds, converts later parameters to keyword-only, and can remove *args entirely.

For example:

import inspect
from toolz import curry

def f(a=1, /, *args, **kwargs):
    return a, args, kwargs

wrapped = curry(f, a=10)
wrapped(2, 3)  # (2, (3,), {'a': 10})
inspect.signature(wrapped).bind(2, 3)  # TypeError: too many positional arguments

The advertised signature is (*, a=10, **kwargs), while a remains positional-only with its original default and *args remains available.

Change

Exclude positional-only parameters from keyword-binding transformations in curry.__signature__. Add regression coverage for same-named keyword bindings, multiple positional-only parameters, preservation of variadic arguments, and a simultaneous real positional-or-keyword binding.

Validation

Python 3.12.14. Four new regression cases fail before the source fix and pass afterward. Final local checks:

• coverage run -m pytest --doctest-modules toolz/: 272 passed, 2 expected skips (a skipped doctest and Python 3.14-only annotation test).

• pytest bench/: 19 passed.

• Repository-wide pycodestyle --ignore="E731,W503,W504,E402" --exclude=conf.py,tests,examples,bench -r --show-source .: passed.

• git diff --check: passed.

• pre-commit run --all-files: passed, including PySentry, actionlint, zizmor, and all remaining applicable hooks.

Additional validation completed after opening this draft:

• python -m build and python -m twine check --strict dist/*: passed for baseline and final code, including wheel and sdist.
• Sphinx HTML builds with warnings treated as errors: passed for baseline and final code.
• Re-ran the full suite after installing build tools: 272 passed, 2 expected skips.

AI assistance disclosure: this fix, regression tests, and description were prepared with an OpenAI coding assistant.