DOC: include positional-only and keyword-only parameters in standard

This PR adds keyword-only parameters to the numpydoc style guide, which were added in PEP 3102. This change would specify that keyword-only gets added next to the type information, like with optional. This was suggested in #330. Here are two examples:

def f(*, x, y=None):
    """
    Parameters
    ----------
    x : int, keyword-only
        ...
    y : int, keyword-only, optional
        ...
    """
    ...

Here are some alternatives that I considered:

• Specify that a parameter is keyword-only in the description of the parameter. Some examples.

  def f(*, x=None, y=None):
      """
      Parameters
      ----------
      x : int, optional
          Some number.  Keyword-only.
      y : int, optional
          Or it could be in parentheses (keyword-only).  
      """
  

  I didn't choose this approach since the keyword-only might be missed by someone who is quickly scanning through the parameters, and because the information about if something is keyword-only is closely related to the information about if a parameter is optional.
• Include a separate section for keyword-only parameters. I didn't do this because if a parameter is keyword-only and required, someone might miss it if it is in a different section.
• Do not include keyword-only parameters in the standard. I didn't choose this approach because right now there are different practices for specifying keyword-only parameters. If someone is expecting keyword-only to show up next to the type, then they would be more likely to miss it if it is in the parameter description. Also, if someone wanted to create an automated tool to automatically link to a definition of keyword-only, having a standard to follow would make it easier to implement.

Closes #330. I'm open to alternatives and edits, in case anyone has suggestions. Many thanks!

Sounds reasonable to me. I recently started adding keyword-only via ..., *, ... in some projects. It would be great if this were automagically documented!

Cool! I just edited this to include positional-only parameters also, which I had forgotten about because the main project I'm working on is currently Python 3.7+, and this capability was added in 3.8.

def f(x, /):
    """
    Parameters
    ----------
    x : int, positional-only
    """

One thing I'm not sure about is if the Python code needs to be updated to reflect these changes, but if so, I'd be happy to try to make those changes too.

One thing I'm not sure about is if the Python code needs to be updated to reflect these changes, but if so, I'd be happy to try to make those changes too.

It probably does at least in terms of not hitting a problem with automatic xref resolution. It shouldn't try to resolve these new terms

In principle it'd be nice to add related checks to the docstring validation as well, to help codify the updates to the standard.

I was thinking that "keyword-only" and "positional-only" should be treated the same as "optional", since they're used in the same place for similar reasons.

I took a look at the code and if I'm understanding things right, numpydoc_xref_ignore defaults to an empty set, so at present it is necessary to include "optional" in numpydoc_xref_ignore in order for it to be ignored.

I put "keyword-only" and "positional-only" in the example for numpydoc_xref_ignore in the configuration section, and tentatively did the same thing in doc/conf.py too.

In principle it'd be nice to add related checks to the docstring validation as well, to help codify the updates to the standard.

That's an interesting thought! I'd really like a tool (maybe a pre-commit hook?) that would look for positional-only, keyword-only, or optional parameters in function signature, and then make sure that the appropriate term shows up in the type description.

Just to follow up on this since it's been a few months, is there a timeline for deciding on this? Thanks!

@jarrodmillman marked it for the 1.5.0 milestone, so hopefully we'll discuss before then

Just to follow up on this since it's been a few months, is there a timeline for deciding on this? Thanks!

I'm generally in favor, though I do think that having an associated validation check for this is important. @namurphy I'm happy to take a stab at this, though I can't make any promises re: timeline!

@rossbar — please feel free to! Since you're not sure about the timeline, would it make sense to have the validation be in a separate pull request?

"Keyword-only" and "optional" behave really similarly here. I did a search for "optional" in this repo, and it didn't seem like there was a validation for "optional" either (though I'm not particularly familiar with the source code). I'm wondering if it would be helpful to include a validation for "optional" at the same time.

Thank you!

it didn't seem like there was a validation for "optional" either (though I'm not particularly familiar with the source code). I'm wondering if it would be helpful to include a validation for "optional" at the same time.

I think this is just historical - "optional" has been around since the beginning but the validation module is much newer. +1 for adding additional validation for "optional" as well, and I agree it will be very similar to the validation checks proposed here so it may make sense to add it here as well.

The reason I'm strongly advocating for adding validation checks for new additions to the standard is that we can then run the checks against some of the core ecosystem libraries (numpy, scipy, matplotlib, the scikits, etc.) to check that any new additions to the standard aren't particularly disruptive. It's a bit more work, but it will provide strong evidence in favor of adopting new standards.