TST: DataFrame operations with ExtensionArrays

[jorisvandenbossche]

@jbrockmendel the use case I was trying to explain in words is this (which currently works):

In [4]: df1 = pd.DataFrame({'a': DecimalArray(make_data()[:5]), 'b': DecimalArray(make_data()[:5])}) 

In [5]: df1 
Out[5]: 
                                                   a                                                  b
0  Decimal: 0.12461374221703280795736645814031362...  Decimal: 0.52335889201915997137604108502273447...
1  Decimal: 0.31309497234009509014640570967458188...  Decimal: 0.00133572791466307627672449598321691...
2  Decimal: 0.84547604466523051947035582998069003...  Decimal: 0.73944523257987115893996588056324981...
3  Decimal: 0.78411203581007593577112402272177860...  Decimal: 0.60960149820278708432397252181544899...
4  Decimal: 0.73864636517230020107405152884894050...  Decimal: 0.53212511863152778257557429242297075...

In [6]: df1.dtypes 
Out[6]: 
a    decimal
b    decimal
dtype: object

In [7]: df2 = pd.DataFrame({'a': np.arange(5), 'b': np.arange(5)})

In [8]: df1 + df2 
Out[8]: 
                                         a                                        b
0  Decimal: 0.1246137422170328079573664581  Decimal: 0.5233588920191599713760410850
1   Decimal: 1.313094972340095090146405710   Decimal: 1.001335727914663076276724496
2   Decimal: 2.845476044665230519470355830   Decimal: 2.739445232579871158939965881
3   Decimal: 3.784112035810075935771124023   Decimal: 3.609601498202787084323972522
4   Decimal: 4.738646365172300201074051529   Decimal: 4.532125118631527782575574292

In [9]: (df1 + df2).dtypes 
Out[9]: 
a    decimal
b    decimal
dtype: object

As you mentioned, in those case the ops code would keep doing that column by column (or 1D block by 1D block) and the right integer 2D block will be splitted.
But I thought you mentioned something about such a case currently not being covered by the tests? (in which case it would be good to add some)

[jbrockmendel]

But I thought you mentioned something about such a case currently not being covered by the tests?

Not sure if this is the case I was referring to, but df2.add(df1["a"], axis=0) currently raises