This documentation is for an old version of IPython. You can find docs for newer versions here.

Writing code for Python 2 and 3


Boolean indicating whether we’re currently in Python 3.


Many built in functions and methods in Python 2 come in pairs, one returning a list, and one returning an iterator (e.g. range() and xrange()). In Python 3, there is usually only the iterator form, but it has the name which gives a list in Python 2 (e.g. range()).

The way to write compatible code depends on what you need:

  • A list, e.g. for serialisation, or to test if something is in it.
  • Iteration, but it will never be used for very many items, so efficiency isn’t especially important.
  • Iteration over many items, where efficiency is important.
list iteration (small) iteration(large)
list(range(n)) range(n) py3compat.xrange(n)
list(map(f, it)) map(f, it)
list(zip(a, b)) zip(a, b)
list(d.items()) d.items() py3compat.iteritems(d)
list(d.values()) d.values() py3compat.itervalues(d)

Iterating over a dictionary yields its keys, so there is rarely a need to use dict.keys() or dict.iterkeys().

Avoid using map() to cause function side effects. This is more clearly written with a simple for loop.


A reference to range on Python 3, and xrange() on Python 2.


Iterate over (key, value) pairs of a dictionary, or just over values. iterkeys is not defined: iterating over the dictionary yields its keys.

Changed standard library locations

Several parts of the standard library have been renamed and moved. This is a short list of things that we’re using. A couple of them have names in IPython.utils.py3compat, so you don’t need both imports in each module that uses them.

Python 2 Python 3 py3compat
raw_input() input input
__builtin__ builtins builtin_mod
StringIO io  
Queue queue  
cPickle pickle  
thread _thread  
copy_reg copyreg  
urlparse urllib.parse  
repr reprlib  
Tkinter tkinter  
Cookie http.cookie  
_winreg winreg  

Be careful with StringIO: io.StringIO is available in Python 2.7, but it behaves differently from StringIO.StringIO, and much of our code assumes the use of the latter on Python 2. So a try/except on the import may cause problems.


Behaves like raw_input() on Python 2.


A reference to the module containing builtins, and its name as a string.


Always be explicit about what is text (unicode) and what is bytes. Encoding goes from unicode to bytes, and decoding goes from bytes to unicode.

To open files for reading or writing text, use io.open(), which is the Python 3 builtin open function, available on Python 2 as well. We almost always need to specify the encoding parameter, because the default is platform dependent.

We have several helper functions for converting between string types. They all use the encoding from IPython.utils.encoding.getdefaultencoding() by default, and the errors='replace' option to do best-effort conversions for the user’s system.

IPython.utils.py3compat.unicode_to_str(u, encoding=None)
IPython.utils.py3compat.str_to_unicode(s, encoding=None)

Convert between unicode and the native str type. No-ops on Python 3.

IPython.utils.py3compat.str_to_bytes(s, encoding=None)
IPython.utils.py3compat.bytes_to_str(u, encoding=None)

Convert between bytes and the native str type. No-ops on Python 2.

IPython.utils.py3compat.cast_unicode(s, encoding=None)
IPython.utils.py3compat.cast_bytes(s, encoding=None)

Convert strings to unicode/bytes when they may be of either type.

IPython.utils.py3compat.cast_unicode_py2(s, encoding=None)
IPython.utils.py3compat.cast_bytes_py2(s, encoding=None)

Convert strings to unicode/bytes when they may be of either type on Python 2, but return them unaltered on Python 3 (where string types are more predictable).


A reference to str on Python 3, and to unicode on Python 2.


A tuple for isinstance checks: (str,) on Python 3, (str, unicode) on Python 2.

Relative imports

# This makes Python 2 behave like Python 3:
from __future__ import absolute_import

import io  # Imports the standard library io module
from . import io  # Import the io module from the package
                  # containing the current module
from .io import foo  # foo from the io module next to this module
from IPython.utils import io  # This still works


The syntax for declaring a class with a metaclass is different in Python 2 and 3. A helper function works for most cases:


Create a base class with a metaclass. Copied from the six library.

Used like this:

class FormatterABC(with_metaclass(abc.ABCMeta, object)):

Combining inheritance between Qt and the traitlets system, however, does not work with this. Instead, we do this:

class QtKernelClientMixin(MetaQObjectHasTraits('NewBase', (HasTraits, SuperQObject), {})):

This gives the new class a metaclass of MetaQObjectHasTraits, and the parent classes HasTraits and SuperQObject.



Refactors print statements in doctests in Python 3 only. Accepts a string or a function, so it can be used as a decorator.


Handle doctests written with {u}'abcþ', replacing the {u} with u for Python 2, and removing it for Python 3.

Accepts a string or a function, so it can be used as a decorator.


IPython.utils.py3compat.execfile(fname, glob, loc=None)

Equivalent to the Python 2 execfile() builtin. We redefine it in Python 2 to better handle non-ascii filenames.



unicode(e) with various fallbacks. Used for exceptions, which may not be safe to call unicode() on.

IPython.utils.py3compat.isidentifier(s, dotted=False)

Checks whether the string s is a valid identifier in this version of Python. In Python 3, non-ascii characters are allowed. If dotted is True, it allows dots (i.e. attribute access) in the string.


Return the current working directory as unicode, like os.getcwdu() on Python 2.


Constructor for types.MethodType that takes two arguments, like the real constructor on Python 3.