make_array with booleans will be converted to integers

[pdtxie]

Hi, current Data 8 TA here!

During a TA meeting we've realised that when creating boolean arrays with make_array(), the elements will be converted to integers and will not follow numpy behaviour due to the extra check for Windows machines (lines 42-45 in util.py).

if elements and all(isinstance(item, (int, np.integer)) for item in elements):
        # Specifically added for Windows machines where the default 
        # integer is int32 - see GH issue #339.
        return np.array(elements, dtype="int64")

You can see this behaviour as follows:

In [1]: Table().with_columns("bool", make_array(True, False))
Out[1]: 
bool
1
0

In [2]: Table().with_columns("bool", np.array([True, False]))
Out[2]: 
bool
True
False

This is because Python behaviour will cause booleans to also return True for isinstance(b, int) and hence they will be converted to int64 by the extra check above.

>>> isinstance(True, int)
True

Unless this is intentional, we could fix this by either explicitly checking for bools so this behaviour does not occur, or changing the isinstance check.

I've created a pull request which adds an extra check for bools, but feel free to suggest another fix!

[davidwagner]

Fixed by #629, I believe