ENH/ERR: More consistent error reporting with invalid win_type in rolling

[mroeschke]

• closes ERR: Most consistent error handling when passing win_type='freq' in rolling #15969
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

Also in-lining methods that were only used once and moving a BaseIndexer test that was using win_type

[jreback]

thanks @mroeschke

[jorisvandenbossche]

@mroeschke this seems to have broken dask (this PR, or one of the other recent rolling-related PRs, but from a quick look at the diff, I suspect this one)

AFAIU, the issue is that the Rolling object (returned from eg df.rolling("1s"), which is a public object) its win_type attribute changed in case of time-based windows. Before, this was set to r.win_type == "freq", but that's no longer the case.

[mroeschke]

Gotcha, do you have a reference issue how dask was using win_type attribute?

[jorisvandenbossche]

Not yet opened an issue about it, but so this is the code where it is failing: https://github.com/dask/dask/blob/4a7a2438219c4ee493434042e50f4cdb67b6ec9f/dask/dataframe/rolling.py#L317-L326. So basically they used win_type == "freq" (self._win_type in dask is pandas' Rolling.win_type) to check if they have time-based rolling, and because that changed they now end up in a different if/else branch

The failure it is causing is eg https://github.com/dask/dask/runs/1644840945#step:5:4998