Lower severity message than redundant-keyword-arg for calling distinct positional-only arg and variadic keyword arg with same name

[jacobtylerwalls]

Current problem

When writing a function, you can defeat the point of a PEP 570 positional-only argument (to the left of /) if:

• it has a default
• later, there is a variadic positional-or-keyword argument e.g. **kwargs

GOOD example (adapted from PEP 570, "Semantic Corner Case"):

def foo(name, /, **kwds):
    return name

See that name is actually positional-only:

>>> foo(name="Jacob")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: foo() missing 1 required positional argument: 'name'

BAD example:

def foo(name="Sarah", /, **kwds):
    return name

See that name is completely ignored:

>>> foo(name="Jacob")
'Sarah'

Desired solution

I suggest a new warning e.g. defeated-positional-only-argument

Additional context

Inspired by #8555

PEP 570

[mbyrnepr2]

The idea for this checker also crossed my mind while pondering #8555 but I dismissed the idea as noise if it is expected Python behaviour. But now that you are also having the same thought then I think there must be some value to it!

[Pierre-Sassoulas]

I'm on mobile so I can't link to the proper documentation but it seems close to what 'keyword arg before vararg' (w1113) does.

[jacobtylerwalls]

Interestingly, if you change the **kwargs to *args, Python complains. I'm starting to think this is a bug in the PEP 570 implementation. @mbyrnepr2 do you want first dibs writing the issue? :D

def foo(name="Sarah", /, *args):
  print(name)

foo(name="Jacob")

TypeError: foo() got some positional-only arguments passed as keyword arguments: 'name'

---

I'm on mobile so I can't link to the proper documentation but it seems close to what 'keyword arg before vararg' (w1113) does.

Yes, we could consider raising that very message, except that the argument we're talking about is supposedly not a keyword arg.

[jacobtylerwalls]

Even more interestingly, even inspect.signature doesn't like it when you do the bad stuff:

>>> import inspect
>>> def foo(name="Sarah", /, **kwds):
...     return name
... 
>>> inspect.signature(foo).bind(name="Jacob")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Library/Frameworks/Python.framework/Versions/3.11/lib/python3.11/inspect.py", line 3211, in bind
    return self._bind(args, kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Library/Frameworks/Python.framework/Versions/3.11/lib/python3.11/inspect.py", line 3107, in _bind
    raise TypeError(msg) from None
TypeError: 'name' parameter is positional only, but was passed as a keyword
>>> foo(name="Jacob")
'Sarah'

[mbyrnepr2]

Thanks for looking into it @jacobtylerwalls; please feel free to go ahead with opening an issue since you had the insight and idea to do so. Also I'm tired as heck today :D

[jacobtylerwalls]

I'm changing my mind now; I see there's no issue in CPython. Mark's comments on #8556 were helpful, as was python/cpython#87106.

I want to suggest instead warning on the problematic call, not on the function definition itself. We should probably just craft a pylint warning for the TypeError mentioned at the top of #8558 (comment). I'll update the title of this issue when I have more time.

[jacobtylerwalls]

Okay, finally figured out what to suggest here. #8556 makes this situation better; we should merge it. Then, we still have two independent problems:

     1  # pylint: disable=missing-function-docstring, missing-module-docstring
     2  def required_positional_with_var_kwargs(a, /, **_kwargs):
     3      return a
     4
     5  def optional_positional_with_var_kwargs(a=1, /, **_kwargs):
     6      return a
     7
     8
     9  # 3 of these 4 succeed, but it's a bit smelly
    10  required_positional_with_var_kwargs(42, a=43)  # returns 42
    11  required_positional_with_var_kwargs(a=43)  # TypeError: should raise no-value-for-parameter
    12
    13  optional_positional_with_var_kwargs(42, a=43)  # returns 42
    14  optional_positional_with_var_kwargs(a=43)  # returns 1

Currently only gives:

************* Module a
a.py:10:0: E1124: Argument 'a' passed by position and keyword in function call (redundant-keyword-arg)
a.py:13:0: E1124: Argument 'a' passed by position and keyword in function call (redundant-keyword-arg)

------------------------------------------------------------------
Your code has been rated at 0.00/10 (previous run: 0.00/10, +0.00)

1
The redundant-keyword-arg is misleading: it should be some new message, since they're not redundant, they're actually two different arguments, and it should also be emitted on line 14. Something like same-named-positional-and-keyword-args?

2

We should raise no-value-for-parameter on line 13. I'll open a new issue for that, since it's more straightforward. EDIT: done here: #8559.

This card can be for the first suggestion. Retitling.

[jacobtylerwalls]

Enh, or we could just not emit any message at all for 1. One way or another, it's a bug, not an enhancement.