Six: Python 2 and 3 Compatibility Library

Six provides simple utilities for wrapping over differences between Python 2 and Python 3. It is intended to support codebases that work on both Python 2 and 3 without modification. six consists of only one Python file, so it is painless to copy into a project.

Six can be downloaded on PyPi. Its bug tracker and code hosting is on BitBucket.

The name, “six”, comes from the fact that 2*3 equals 6. Why not addition? Multiplication is more powerful, and, anyway, “five” has already been snatched away by the (admittedly now moribund) Zope Five project.

Indices and tables

Package contents


A boolean indicating if the code is running on Python 2.


A boolean indicating if the code is running on Python 3.


Six provides constants that may differ between Python versions. Ones ending _types are mostly useful as the second argument to isinstance or issubclass.


Possible class types. In Python 2, this encompasses old-style and new-style classes. In Python 3, this is just new-styles.


Possible integer types. In Python 2, this is long and int, and in Python 3, just int.


Possible types for text data. This is basestring() in Python 2 and str in Python 3.


Type for representing (Unicode) textual data. This is unicode() in Python 2 and str in Python 3.


Type for representing binary data. This is str in Python 2 and bytes in Python 3.


The maximum size of a container like list or dict. This is equivalent to sys.maxsize in Python 2.6 and later (including 3.x). Note, this is temptingly similar to, but not the same as sys.maxint in Python 2. There is no direct equivalent to sys.maxint in Python 3 because its integer type has no limits aside from memory.

Here’s example usage of the module:

import six

def dispatch_types(value):
    if isinstance(value, six.integer_types):
    elif isinstance(value, six.class_types):
    elif isinstance(value, six.string_types):

Object model compatibility

Python 3 renamed the attributes of several intepreter data structures. The following accessors are available. Note that the recommended way to inspect functions and methods is the stdlib inspect module.


Get the function out of unbound method meth. In Python 3, unbound methods don’t exist, so this function just returns meth unchanged. Example usage:

from six import get_unbound_function

class X(object):
    def method(self):
method_function = get_unbound_function(X.method)

Get the function out of method object meth.


Get the self of bound method meth.


Get the closure (list of cells) associated with func. This is equivalent to func.__closure__ on Python 2.6+ and func.func_closure on Python 2.5.


Get the code object associated with func. This is equivalent to func.__code__ on Python 2.6+ and func.func_code on Python 2.5.


Get the defaults tuple associated with func. This is equivalent to func.__defaults__ on Python 2.6+ and func.func_defaults on Python 2.5.


Get the globals of func. This is equivalent to func.__globals__ on Python 2.6+ and func.func_globals on Python 2.5.

Get the next item of iterator it. StopIteration is raised if the iterator is exhausted. This is a replacement for calling in Python 2 and next(it) in Python 3.


Check if obj can be called. Note callable has returned in Python 3.2, so using six’s version is only necessary when supporting Python 3.0 or 3.1.

six.iterkeys(dictionary, **kwargs)

Returns an iterator over dictionary‘s keys. This replaces dictionary.iterkeys() on Python 2 and dictionary.keys() on Python 3. kwargs are passed through to the underlying method.

six.itervalues(dictionary, **kwargs)

Returns an iterator over dictionary‘s values. This replaces dictionary.itervalues() on Python 2 and dictionary.values() on Python 3. kwargs are passed through to the underlying method.

six.iteritems(dictionary, **kwargs)

Returns an iterator over dictionary‘s items. This replaces dictionary.iteritems() on Python 2 and dictionary.items() on Python 3. kwargs are passed through to the underlying method.

six.iterlists(dictionary, **kwargs)

Calls dictionary.iterlists() on Python 2 and dictionary.lists() on Python 3. No builtin Python mapping type has such a method; this method is intended for use with multi-valued dictionaries like Werkzeug’s. kwargs are passed through to the underlying method.


Return a view over dictionary‘s keys. This replaces dict.viewkeys() on Python 2.7 and dict.keys() on Python 3.


Return a view over dictionary‘s values. This replaces dict.viewvalues() on Python 2.7 and dict.values() on Python 3.


Return a view over dictionary‘s items. This replaces dict.viewitems() on Python 2.7 and dict.items() on Python 3.

six.create_bound_method(func, obj)

Return a method object wrapping func and bound to obj. On both Python 2 and 3, this will return a types.MethodType object. The reason this wrapper exists is that on Python 2, the MethodType constructor requires the obj‘s class to be passed.

six.create_unbound_method(func, cls)

Return an unbound method object wrapping func. In Python 2, this will return a types.MethodType object. In Python 3, unbound methods do not exist and this wrapper will simply return func.

class six.Iterator

A class for making portable iterators. The intention is that it be subclassed and subclasses provide a __next__ method. In Python 2, Iterator has one method: next. It simply delegates to __next__. An alternate way to do this would be to simply alias next to __next__. However, this interacts badly with subclasses that override __next__. Iterator is empty on Python 3. (In fact, it is just aliased to object.)

@six.wraps(wrapped, assigned=functools.WRAPPER_ASSIGNMENTS, updated=functools.WRAPPER_UPDATES)

This is exactly the functools.wraps() decorator, but it sets the __wrapped__ attribute on what it decorates as functools.wraps() does on Python versions after 3.2.

Syntax compatibility

These functions smooth over operations which have different syntaxes between Python 2 and 3.

six.exec_(code, globals=None, locals=None)

Execute code in the scope of globals and locals. code can be a string or a code object. If globals or locals are not given, they will default to the scope of the caller. If just globals is given, it will also be used as locals.


Python 3’s exec() doesn’t take keyword arguments, so calling exec() with them should be avoided.

six.print_(*args, *, file=sys.stdout, end="\n", sep=" ", flush=False)

Print args into file. Each argument will be separated with sep and end will be written to the file after the last argument is printed. If flush is true, file.flush() will be called after all data is written.


In Python 2, this function imitates Python 3’s print() by not having softspace support. If you don’t know what that is, you’re probably ok. :)

six.raise_from(exc_value, exc_value_from)

Raise an exception from a context. On Python 3, this is equivalent to raise exc_value from exc_value_from. On Python 2, which does not support exception chaining, it is equivalent to raise exc_value.

six.reraise(exc_type, exc_value, exc_traceback=None)

Reraise an exception, possibly with a different traceback. In the simple case, reraise(*sys.exc_info()) with an active exception (in an except block) reraises the current exception with the last traceback. A different traceback can be specified with the exc_traceback parameter. Note that since the exception reraising is done within the reraise() function, Python will attach the call frame of reraise() to whatever traceback is raised.

six.with_metaclass(metaclass, *bases)

Create a new class with base classes bases and metaclass metaclass. This is designed to be used in class declarations like this:

from six import with_metaclass

class Meta(type):

class Base(object):

class MyClass(with_metaclass(Meta, Base)):

Another way to set a metaclass on a class is with the add_metaclass() decorator.


Class decorator that replaces a normally-constructed class with a metaclass-constructed one. Example usage:

class MyClass(object):

That code produces a class equivalent to

class MyClass(object, metaclass=Meta):

on Python 3 or

class MyClass(object):
    __metaclass__ = MyMeta

on Python 2.

Note that class decorators require Python 2.6. However, the effect of the decorator can be emulated on Python 2.5 like so:

class MyClass(object):
MyClass = add_metaclass(Meta)(MyClass)

Binary and text data

Python 3 enforces the distinction between byte strings and text strings far more rigoriously than Python 2 does; binary data cannot be automatically coerced to or from text data. six provides several functions to assist in classifying string data in all Python versions.


A “fake” bytes literal. data should always be a normal string literal. In Python 2, b() returns a 8-bit string. In Python 3, data is encoded with the latin-1 encoding to bytes.


Since all Python versions 2.6 and after support the b prefix, b(), code without 2.5 support doesn’t need b().


A “fake” unicode literal. text should always be a normal string literal. In Python 2, u() returns unicode, and in Python 3, a string. Also, in Python 2, the string is decoded with the unicode-escape codec, which allows unicode escapes to be used in it.


In Python 3.3, the u prefix has been reintroduced. Code that only supports Python 3 versions of 3.3 and higher thus does not need u().


On Python 2, u() doesn’t know what the encoding of the literal is. Each byte is converted directly to the unicode codepoint of the same value. Because of this, it’s only safe to use u() with strings of ASCII data.


Return the (Unicode) string representing the codepoint c. This is equivalent to unichr() on Python 2 and chr() on Python 3.


Converts i to a byte. i must be in range(0, 256). This is equivalent to chr() in Python 2 and bytes((i,)) in Python 3.


Converts the first byte of bs to an integer. This is equivalent to ord(bs[0]) on Python 2 and bs[0] on Python 3.

six.indexbytes(buf, i)

Return the byte at index i of buf as an integer. This is equivalent to indexing a bytes object in Python 3.


Return an iterator over bytes in buf as integers. This is equivalent to a bytes object iterator in Python 3.


This is an fake file object for textual data. It’s an alias for StringIO.StringIO in Python 2 and io.StringIO in Python 3.


This is a fake file object for binary data. In Python 2, it’s an alias for StringIO.StringIO, but in Python 3, it’s an alias for io.BytesIO.


A class decorator that takes a class defining a __str__ method. On Python 3, the decorator does nothing. On Python 2, it aliases the __str__ method to __unicode__ and creates a new __str__ method that returns the result of __unicode__() encoded with UTF-8.

unittest assertions

Six contains compatibility shims for unittest assertions that have been renamed. The parameters are the same as their aliases, but you must pass the test method as the first argument. For example:

import six
import unittest

class TestAssertCountEqual(unittest.TestCase):
    def test(self):
        six.assertCountEqual(self, (1, 2), [2, 1])

Note these functions are only available on Python 2.7 or later.


Alias for assertCountEqual() on Python 3 and assertItemsEqual() on Python 2.


Alias for assertRaisesRegex() on Python 3 and assertRaisesRegexp() on Python 2.


Alias for assertRegex() on Python 3 and assertRegexpMatches() on Python 2.

Renamed modules and attributes compatibility

Python 3 reorganized the standard library and moved several functions to different modules. Six provides a consistent interface to them through the fake six.moves module. For example, to load the module for parsing HTML on Python 2 or 3, write:

from six.moves import html_parser

Similarly, to get the function to reload modules, which was moved from the builtin module to the imp module, use:

from six.moves import reload_module

For the most part, six.moves aliases are the names of the modules in Python 3. When the new Python 3 name is a package, the components of the name are separated by underscores. For example, html.parser becomes html_parser. In some cases where several modules have been combined, the Python 2 name is retained. This is so the appropiate modules can be found when running on Python 2. For example, BaseHTTPServer which is in http.server in Python 3 is aliased as BaseHTTPServer.

Some modules which had two implementations have been merged in Python 3. For example, cPickle no longer exists in Python 3; it was merged with pickle. In these cases, fetching the fast version will load the fast one on Python 2 and the merged module in Python 3.

The urllib, urllib2, and urlparse modules have been combined in the urllib package in Python 3. The six.moves.urllib package is a version-independent location for this functionality; its structure mimics the structure of the Python 3 urllib package.


In order to make imports of the form:

from six.moves.cPickle import loads

work, six places special proxy objects in in sys.modules. These proxies lazily load the underlying module when an attribute is fetched. This will fail if the underlying module is not available in the Python interpreter. For example, sys.modules["six.moves.winreg"].LoadKey would fail on any non-Windows platform. Unfortunately, some applications try to load attributes on every module in sys.modules. six mitigates this problem for some applications by pretending attributes on unimportable modules don’t exist. This hack doesn’t work in every case, though. If you are encountering problems with the lazy modules and don’t use any from imports directly from six.moves modules, you can workaround the issue by removing the six proxy modules:

d = [name for name in sys.modules if name.startswith("six.moves.")]
for name in d:
    del sys.modules[name]

Supported renames:

Name Python 2 name Python 3 name
builtins __builtin__ builtins
configparser ConfigParser configparser
copyreg copy_reg copyreg
cPickle cPickle pickle
cStringIO cStringIO.StringIO() io.StringIO
dbm_gnu gdbm dbm.gnu
_dummy_thread dummy_thread _dummy_thread
email_mime_multipart email.MIMEMultipart email.mime.multipart
email_mime_nonmultipart email.MIMENonMultipart email.mime.nonmultipart
email_mime_text email.MIMEText email.mime.text
email_mime_base email.MIMEBase email.mime.base
filter itertools.ifilter() filter()
filterfalse itertools.ifilterfalse() itertools.filterfalse()
getcwd os.getcwdu() os.getcwd()
getcwdb os.getcwd() os.getcwdb()
http_cookiejar cookielib http.cookiejar
http_cookies Cookie http.cookies
html_entities htmlentitydefs html.entities
html_parser HTMLParser html.parser
http_client httplib http.client
BaseHTTPServer BaseHTTPServer http.server
CGIHTTPServer CGIHTTPServer http.server
SimpleHTTPServer SimpleHTTPServer http.server
input raw_input() input()
intern intern() sys.intern()
map itertools.imap() map()
queue Queue queue
range xrange() range
reduce reduce() functools.reduce()
reload_module reload() imp.reload(), importlib.reload() on Python 3.4+
reprlib repr reprlib
shlex_quote pipes.quote shlex.quote
socketserver SocketServer socketserver
_thread thread _thread
tkinter Tkinter tkinter
tkinter_dialog Dialog tkinter.dialog
tkinter_filedialog FileDialog tkinter.FileDialog
tkinter_scrolledtext ScrolledText tkinter.scrolledtext
tkinter_simpledialog SimpleDialog tkinter.simpledialog
tkinter_ttk ttk tkinter.ttk
tkinter_tix Tix tkinter.tix
tkinter_constants Tkconstants tkinter.constants
tkinter_dnd Tkdnd tkinter.dnd
tkinter_colorchooser tkColorChooser tkinter.colorchooser
tkinter_commondialog tkCommonDialog tkinter.commondialog
tkinter_tkfiledialog tkFileDialog tkinter.filedialog
tkinter_font tkFont tkinter.font
tkinter_messagebox tkMessageBox tkinter.messagebox
tkinter_tksimpledialog tkSimpleDialog tkinter.simpledialog
urllib.parse See six.moves.urllib.parse urllib.parse
urllib.error See six.moves.urllib.error urllib.error
urllib.request See six.moves.urllib.request urllib.request
urllib.response See six.moves.urllib.response urllib.response
urllib.robotparser robotparser urllib.robotparser
urllib_robotparser robotparser urllib.robotparser
UserDict UserDict.UserDict collections.UserDict
UserList UserList.UserList collections.UserList
UserString UserString.UserString collections.UserString
winreg _winreg winreg
xmlrpc_client xmlrpclib xmlrpc.client
xmlrpc_server SimpleXMLRPCServer xmlrpc.server
xrange xrange() range
zip itertools.izip() zip()
zip_longest itertools.izip_longest() itertools.zip_longest()

urllib parse

Contains functions from Python 3’s urllib.parse and Python 2’s:


and urllib:

urllib error

Contains exceptions from Python 3’s urllib.error and Python 2’s:


and urllib2:

urllib response

Contains classes from Python 3’s urllib.response and Python 2’s:


  • urllib.addbase
  • urllib.addclosehook
  • urllib.addinfo
  • urllib.addinfourl

Advanced - Customizing renames

It is possible to add additional names to the six.moves namespace.


Add item to the six.moves mapping. item should be a MovedAttribute or MovedModule instance.


Remove the six.moves mapping called name. name should be a string.

Instances of the following classes can be passed to add_move(). Neither have any public members.

class six.MovedModule(name, old_mod, new_mod)

Create a mapping for six.moves called name that references different modules in Python 2 and 3. old_mod is the name of the Python 2 module. new_mod is the name of the Python 3 module.

class six.MovedAttribute(name, old_mod, new_mod, old_attr=None, new_attr=None)

Create a mapping for six.moves called name that references different attributes in Python 2 and 3. old_mod is the name of the Python 2 module. new_mod is the name of the Python 3 module. If new_attr is not given, it defaults to old_attr. If neither is given, they both default to name.