DEPR: combine_first

[jbrockmendel]

I've never used it, and the implementations are slightly different between Series/DataFrame which I expect will surprise users.

[jreback]

see #22812

[phofl]

I 've used it quite often as a coalesce solution

[lukemanley]

I 've used it quite often as a coalesce solution

Same, I've used this quite a bit.

I do think the method name is not obvious to new users.

I also wonder if it would be better as a top level func so rather than doing:

ser1.combine_first(ser2).combine_first(ser3)

it could be something like:

pd.coalesce([ser1, ser2, ser3])

[Jython1415]

I'm new here but I think the comment above about switching to pd.coalesce([ser1, ser2, ser3]) connects to the discussion on #47862 that I happened to come across about what namespace these functions should fall under. @jreback's comment seems particularly relevant.

[Dr-Irv]

Just reviewed some code from my team, and they used combine_first().

So, if we deprecate, then we need a replacement.

[jbrockmendel]

For people who don't want to deprecate: do you care about both the Series and DataFrame methods, or just one of them? I suspect the answer is "just the DataFrame method". Deprecating Series.combine_first would satisfy my main issue of "the two methods behave differently"

[phofl]

I care only about the series method actually as a coalesce substitute