BUG: BooleanArray.any with all False values and skipna=False is buggy

[jorisvandenbossche]

From a report on my blog, this seems buggy:

In [10]: a = pd.array([False, False], dtype="boolean")  

In [11]: a  
Out[11]: 
<BooleanArray>
[False, False]
Length: 2, dtype: boolean

In [12]: a.any() 
Out[12]: False

In [13]: a.any(skipna=False)   
Out[13]: <NA>

I don't think there is a reason to return NA if there are no NAs in the array? (even with skipna=False)

We probably need to do a check for the presence of any NA here in the else clause:

pandas/pandas/core/arrays/boolean.py

Lines 520 to 526 in 37dc5dc

	if skipna:
	return result
	else:
	if result or len(self) == 0:
	return result
	else:
	return self.dtype.na_value

[jorisvandenbossche]

And the same for all() with an all True array:

In [14]: a = pd.array([True, True], dtype="boolean") 

In [15]: a.all() 
Out[15]: True

In [16]: a.all(skipna=False)    
Out[16]: <NA>

[jorisvandenbossche]

@TomAugspurger I think it is quite obvious that the result should be same here for skipna=True/False if there are no NAs, but can you do a quick sanity check? (we didn't include thoses cases in the overview table at #29686)

[TomAugspurger]

Agreed with your proposed changes to the return values.

[linxiaow]

It seems that checking if there is any NA in both .any() method and .all() method right after the second else clause may solve this issue based on the table #29686. I am new to open source and just looking around. I was wondering whether I can have a try if you believe this issue is suitable for a beginner?

[jorisvandenbossche]

@linxiaow Yes, this issue should be suitable. A pull request is very welcome!

[linxiaow]

I took a look at the code

pandas/pandas/core/arrays/boolean.py

Lines 517 to 526 in 37dc5dc

	values = self._data.copy()
	np.putmask(values, self._mask, False)
	result = values.any()
	if skipna:
	return result
	else:
	if result or len(self) == 0:
	return result
	else:
	return self.dtype.na_value

I quickly went through previous code and find a docstring for mask variable

pandas/pandas/core/arrays/boolean.py

Lines 239 to 241 in 37dc5dc

	mask : numpy.ndarray
	A 1-d boolean-dtype array indicating missing values (True
	indicates missing).

I would assume self._mask in line 518 is also 1d ndarray, and I believe a .any() method for self._mask in line 523 would be enough to know whether there is a missing value. Same for .all(). If I am right, I can make a PR ASAP

[jorisvandenbossche]

@linxiaow that sounds about right! So PR welcome, that will also be easier to give feedback (and I would also advice to first write tests, so you can actually check if your fix is working)

[linxiaow]

@jorisvandenbossche I made a PR but it seems to fail the CI PR tests. A lot of PR fails, too. I am not sure what I missed. Any suggestions?