Repository navigation
implicit ascii encoding for dtypes of kind S #6315
Description
Greetings,
Investigating PyTables/PyTables#499 I came across the following numpy's behaviour that I find unexpected:
In [122]: z = np.zeros(1, dtype='S10') # 'S' is bytes, not a string
In [123]: z
Out[123]:
array([b''],
dtype='|S10')
In [124]: z[0] = 'ciao' # enters a string
In [125]: z[0] # exit encoded bytes
Out[125]: b'ciao'
In [126]: z[0] = '❤' # and the encoding is just ascii
---------------------------------------------------------------------------
UnicodeEncodeError Traceback (most recent call last)
<ipython-input-126-387e70d94037> in <module>()
----> 1 z[0] = '❤'
UnicodeEncodeError: 'ascii' codec can't encode character '\u2764' in position 0: ordinal not in range(128)So it seems that the string 'ciao' automagically becomes 'ciao'.encode('ascii') on its way to z[0] which is surprising and kind of wrong anyway. Said otherwise, I won't expect any implicit conversion between bytes and strings.
Activity
mattip commentedon Aug 28, 2019
Closing. The original reporter should have used
z = np.zeros(1, dtype='U10')to hold encoded strings (U10for unicode)