DEPR: pd.Grouper key -> on

[jreback]

xref #15326 (comment)

• when pd.Grouper was introduced, we used key to indicate the column which to group on
• .resample added the same idea as on
• we use on in pd.merge (existing for a long time)
• .groupby uses by (though to be fair this can take lots of things and is not strictly a string column name).

So the proposal is to deprecate pd.Grouper(key=...) in favor of on=.

[jreback]

@chrisaycock gives +1
@jorisvandenbossche -0
@jreback +1

[chrisaycock]

pd.merge_asof() has both on and by parameters.

[jreback]

any thoughts about whether we should do this (at some point)?

@TomAugspurger @jorisvandenbossche

[TomAugspurger]

No strong opinion.

[jbrockmendel]

Any clarity on this?

[jorisvandenbossche]

Since it's only about a possible future deprecation (not a deprecation that can be enforced in 1.0), there is not need to have this in 1.0 (we can deprecate things in any 1.x afterwards) and so I moved this off the 1.0 milestone.

[mroeschke]

Since we decided to keep Grouper (#41297) and there is no strong leaning on this, probably best to keep it as necessary to the Grouper API. Closing but can reopen if we redecide.