[ARROW-17542] [Python][Dev] Document the process to run numpydoc checks

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: Archery, Developer Tools, Python
• External issue URL: https://github.com/apache/arrow/issues/32796
• Assignee: Unassigned
• Reporter: Bryce Mecum
• Created: 27/Aug/22 02:37
• Updated: 11/Jan/23 11:51

Description

In ARROW-15006 we are adding additional validation checks which are run during CI but the Python development documentation doesn't contain any instructions for running those checks locally. Adding instructions would likely save everyone time.

However, we actually have two separate ways to run the numpydoc checks locally:

• archery lint --numpydoc (runs all checks)
• archery numpydoc (runs all checks except except four we've disabled by default)

And these two methods differ from how CI runs numpydoc which only runs the three checks (PR01, PR03, and PR10) we know currently pass.

My thoughts so far are:

• Having two similar but different ways of running numpydoc might be confusing
• Ideally, new work should pass all numpydoc checks, not just those we run in CI. Otherwise we're just making future work for ourselves.
• Giving developers an easy way to run just the checks we have enabled in CI would be good

 

Issue Links

• relates to: ARROW-15006 [Python][Doc] Iteratively enable more numpydoc checks