Naming Conventions: "_data", "data", "values", "_values"

[jbrockmendel]

A bunch of different classes have one or more of the attributes data, _data, values, _values, plus an assortment of external_values, internal_values, formatting_values, get_values. These mean different things in different places.

Maintenance would be easier if the naming conventions were more uniform. Index has all four of these attributes and I'm not sure there exists a nice backwards-compatible way to reconcile them with the naming in Series/DataFrame. Any thoughts? Does anything else think this matters?

(Motivating example: "Where are all the places in the code that touch a BlockManager. Let's just grep for \.data...")

The lowest-hanging fruit for cleanup here is in the Accessor classes. StringAccessor, SeriesPlotMethods, and FramePlotMethods all define _data to point back to their parent Series/Index, Series, and Frame, respectively. I suggest that _data be replaced with just _parent. The other two existing accessors CategoricalAccessor and CombinedDatetimelikeProperties use categories and values for these, respectively. Ideally these would get standardized to _parent in the process.

Another option would be to change NDFrame._data to something like NDFrame._mgr so it there is little risk of name-overlap. I expect this would meet more resistance than the accessor cleanup idea.

[jreback]

well .values is an external property. I don't think there should be any usage of .data

[jbrockmendel]

well .values is an external property. I don't think there should be any usage of .data

Yah, .values is pretty set-in-stone, though there may be some non-external cases like DatetimeProperties.values that could be avoided.

.data looks like it usually is the IndexOpsMixin.data property which points to self.values.data. There are a bunch of other places that set self.data = (just based on grep) that could be given more informative names if they are not public-facing.

[chris-b1]

Index.data and Series.data are likely leftovers from ndarray subclassing / compat - I'd think they could be deprecated and removed.

[jbrockmendel]

Subsumed by #19658