Bug in Series constructor returning wrong missing values

[phofl]

• closes BUG: Default value for Series(dtype=int) is not pd.NA #43017
• xref BUG: Inconsistent behavior of setting default value for type bool #43018 (if the reindex case is ok, then this is done too)
• tests added / passed
• Ensure all linting tests pass, see here for how to run them
• whatsnew entry

This ensures consistency with the DataFrame constructor

[jreback]

this is True?

[jbrockmendel]

bool(np.nan) returns True

[phofl]

Yes exactly.

[jreback]

this is True?

[jreback]

thanks @phofl

[simonjayhawkins]

@phofl we have had reports regarding both these cases. Is the new behavior now consistent?

import pandas as pd

print(pd.__version__)
s = pd.Series(dtype="int", index=[0])
print(s)
s2 = pd.Series(dtype="bool", index=[0])
print(s2)

1.3.5
0    0
dtype: int64
0    False
dtype: bool

1.4.1
0   NaN
dtype: float64
0    True
dtype: bool

It appears to me (and maybe others from the issues) that the changes are confusing.

We have changed the int case and and instead create a float array saying that the missing value cannot be held in a integer array and yet for the bool case we continue to keep the bool dtype even though we cannot represent a missing value in a boolean array?

I appreciate that int(np.nan) raises ValueError: cannot convert float NaN to integer and bool(np.nan) is True but the users are not specifying np.nan, they are not supplying data. I wonder whether we ought to revert this until we change the constructors to use nullable dtypes?

[jbrockmendel]

I wonder whether we ought to revert this until we change the constructors to use nullable dtypes?

no opinion on the reversion, but i advise against the implicit assumption that constructors are going to default to nullable dtypes. support has improved, but its still a mess.