CLN: deduplicate __setitem__ and _reduce on masked arrays

[jorisvandenbossche]

For __setitem__, the only difference was that they both call to their own "coercing" function, so I made this available as a helper method, so it the rest of the implementation of __setitem__ can be shared.

For _reduce, everything was the same, BooleanArray just has an additional check for any and all.

[WillAyd]

lgtm

[WillAyd]

orthogranal to this PR but wondering if we even need this any more or could just use abc.abstractmethod decorator

[jorisvandenbossche]

We generally use this instead of the decorator in the EA base class, because the decorator requires it to actually be an abstract base class.

See

pandas/pandas/core/arrays/base.py

Lines 128 to 131 in 45fee32

	This class does not inherit from 'abc.ABCMeta' for performance reasons.
	Methods and properties required by the interface raise
	``pandas.errors.AbstractMethodError`` and no ``register`` method is
	provided for registering virtual subclasses.

[simonjayhawkins]

as a follow up, should this be self.dtype.na_value instead of hardcoded in the base class.

[jorisvandenbossche]

I personally like having it explicit as well (all subclasses use NA as missing value indicator, this method is not shared with ones where dtype.na_value is something different), but can go either way

[simonjayhawkins]

hmm,

we have

    @property
    def _na_value(self):
        return self.dtype.na_value

so this is not consistent.

[jorisvandenbossche]

We could also remove that property, from a quick search, that is only used once in the arrays code

[simonjayhawkins]

I agree, if not generally on an array class, I think self.dtype.na_value is generally clearer.

This is not the only case where libmissing.NA is used, so we should use one ore the other. I accept your argument about being explicit but also prefer the the case of allowing the class to be extensible. YAGNI favours your proposal.