Refactor DataFrame.combine for datetimelike · Issue #23079 · pandas-dev/pandas (original) (raw)

In DataFrame.combine, we handle datetimelike data special.

We attempt to pass a third argument to the combining func. This breaks things like

In [16]: def combiner(x, y): return x

In [17]: a = pd.DataFrame({"A": range(4)})

In [18]: b = pd.DataFrame({"A": pd.date_range('2017', periods=4)})

In [19]: a.combine(a, combiner) Out[19]: A 0 0 1 1 2 2 3 3

In [20]: b.combine(b, combiner)

TypeError Traceback (most recent call last) in ----> 1 b.combine(b, combiner)

~/sandbox/pandas/pandas/core/frame.py in combine(self, other, func, fill_value, overwrite) 5116 needs_i8_conversion_i = needs_i8_conversion(new_dtype) 5117 if needs_i8_conversion_i: -> 5118 arr = func(series, otherSeries, True) 5119 else: 5120 arr = func(series, otherSeries)

TypeError: combiner() takes 2 positional arguments but 3 were given

We could document that, but it still feels hacky. I'd rather that the combiner func infer whether an i8 conversion is needed.

xref #3595