CLN: remove unused row_bool_subset

[jbrockmendel]

@jreback can you confirm my hunch that the result.ndim == 2 block in groupby.ops was for Panel?

Tracing back the blame, row_bool_subset was introduced here in 2012 to close #152 and it started being called only for result.ndim == 2 here

[WillAyd]

Haven't looked deeply but seems like this is more for ohlc than Panel right?

[jbrockmendel]

I speculate that it’s for Panel because if it is an agg op that should be a reduction, so we shouldn’t get a 2D result unless we started with 3D. But “speculate” is definitely the important word in that sentence

[WillAyd]

so we shouldn’t get a 2D result unless we started with 3D

Yea that's not quite true with ohlc though. To illustrate:

In [20]: ser = pd.Series([1], index=[pd.Timestamp('today')])

In [21]: ser.groupby([1]).ohlc()
Out[21]:
   open  high  low  close
1     1     1    1      1

Not sure what would specifically hit this branch but I think ohlc test coverage is weak so just need to be careful

[jbrockmendel]

Dangit Will, you're gonna make me go and do actual critical thinking instead of mechanically reading the coverage results...

[jbrockmendel]

Hmm that example doesn't break on this branch, so must not go through this code path. Still, the point that test coverage is spotty is well-taken.

[jreback]

thanks @jbrockmendel

I don't really remember where this was used, does seem like its related to Panel though.