Handle a mergedBy NaN value

[jasongrout]

This avoids a TypeError when retrieving PRs that have not been merged, which was causing a critical error. From where mergedBy is set, it can be a pandas na or it can be a username.

Fixes #164

To reproduce:

github-activity jupyterlab/jupyterlab --since v4.6.0a1 --kind pr

Error:

Traceback (most recent call last):
  File "/[snip]/github-activity", line 7, in <module>
    sys.exit(main())
             ~~~~^^
  File "/[snip]/github-activity/github_activity/cli.py", line 234, in main
    md = generate_activity_md(
        args.target,
    ...<3 lines>...
        **common_kwargs,
    )
  File "/[snip]/github-activity/github_activity/github_activity.py", line 594, in generate_activity_md
    data.at[ix, "contributors"] = list(item_contributors)
                                  ~~~~^^^^^^^^^^^^^^^^^^^
  File "/[snip]/github-activity/github_activity/github_activity.py", line 386, in __iter__
    for item in sorted(self.other - {self.author}):
                ~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: '<' not supported between instances of 'float' and 'str'

[jasongrout]

What is weird to me is that this seems to only have recently started failing. Why did things pass before?

[bsipocz]

What is weird to me is that this seems to only have recently started failing. Why did things pass before?

Some upstream version differences?

[krassowski]

Some upstream version differences?

It would be pandas v3.0.0 I bet. The timing agrees with when it started failing, and the changeset here looks relevant.

[choldgraf]

hah I just ran into this in myst-theme - let's make a release so we get this fix out

[jasongrout]

It would be pandas v3.0.0 I bet. The timing agrees with when it started failing, and the changeset here looks relevant.

Yep, comparing the logs between a failing and succeeding run, the difference is pandas 3.0 is in the failing run, pandas 2.3.3 in the run that succeeded.

[jasongrout]

I bet it's that there is a new str type, where the missing value is always NaN. I bet before it was None, which is falsey, but NaN is truthy, so that boolean check used to work, but now failed.

https://pandas.pydata.org/docs/user_guide/migration-3-strings.html#the-missing-value-sentinel-is-now-always-nan

According to the migration guide, the correct fix is to use pd.isna (or notna for the opposite check), so I think this PR does the correct thing.

[jasongrout]

hah I just ran into this in myst-theme - let's make a release so we get this fix out

+1 - this is blocking us making a jupyterlab release

[choldgraf]

release is already out in the wild!

[jasongrout]

And our CI tests are now passing with 1.1.5. Thanks!