array.array()¶

The first argument to array.array(typecode[, initializer]) must be a native platform string: unicode string on Python 3, byte string on Python 2.

Python 2::

>>> array.array(b'b')
array.array(b'b')

>>> array.array(u'u')
TypeError: must be char, not unicode

Python 3::

>>> array.array(b'b')
TypeError: must be a unicode character, not bytes

>>> array.array(u'b')
array('b')

This means that the typecode cannot be specified portably across Python 3 and Python 2 with a single string literal when from __future__ import unicode_literals is in effect.

You can use the following code on both Python 3 and Python 2:

from __future__ import unicode_literals
from future.utils import bytes_to_native_str
import array

# ...

a = array.array(bytes_to_native_str(b'b'))

This was fixed in Python 2.7.11. Since then, array.array() now also accepts unicode format typecode.

array.array.read()¶

This method has been removed in Py3. This crops up in e.g. porting http.client.

base64.decodebytes() and base64.encodebytes()¶

The base64 module on Py2 has no decodebytes or encodebytes functions.

re.ASCII¶

Python 3 code using regular expressions sometimes looks like this (from urllib.request):

re.compile(r":\d+$", re.ASCII)

This enables ‘ASCII mode’ for regular expressions (see the docs here). Python 2’s re module has no equivalent mode.

struct.pack()¶

Before Python version 2.7.7, the struct.pack() function required a native string as its format argument. For example:

>>> from __future__ import unicode_literals
>>> from struct import pack
>>> pack('<4H2I', version, rec_type, build, year, file_hist_flags, ver_can_read)

raised TypeError: Struct() argument 1 must be string, not unicode.

This was fixed in Python 2.7.7. Since then, struct.pack() now also accepts unicode format strings.