This section includes notes and tips for updating from Python 2 to Python 3, including summaries of and references for the changes in each module.
The notes in this section are based on the What’s New documents prepared by the Python development team and release manager for each release.
- What’s New In Python 3.0
- What’s New In Python 3.1
- What’s New In Python 3.2
- What’s New In Python 3.3
- What’s New In Python 3.4
- What’s New In Python 3.5
For more information about porting to Python 3, refer to
Python 3 includes a number of new modules, providing features not present in Python 2.
- Asynchronous I/O, event loop, and other concurrency tools
- Managing Pools of Concurrent Tasks
- Install the Python Package Installer, pip
- Defines enumeration type
- Classes for working with Internet Protocol (IP) addresses
- An object-oriented API for working with filesystem paths
- I/O Multiplexing Abstractions
- Statistical Calculations
- Create isolated installation and execution contexts
Many standard library modules were renamed between Python 2 and 3 as
part of PEP 3108. All of the new module names use consistent lower
case, and some have been moved into packages to better organize
related modules. Often, code using these modules can be updated to
work with Python 3 just by fixing the import statements. A complete
list of the renames can be found in the dictionary
lib2to3.fixes.fix_imports.MAPPING (the keys are the Python 2 name
and the values are the Python 3 name) and in the table below.
|Python 2 Name||Python 3 Name|
- The six package is useful for writing code that runs under both
Python 2 and 3. In particular, the
six.movesmodule allows your code to import renamed modules using a single import statement, automatically redirecting the import to the correct version of the name depending on the version of Python.
- PEP 3108 – Standard Library Reorganization
These modules are either no longer present at all, or have had their features merged into other existing modules.
dbm.bsd modules have been
removed. Bindings for Berkeley DB are now maintained outside of the
standard library as bsddb3.
commands module was deprecated in Python 2.6 and removed
in Python 3.0. See
compiler module has been removed. See
dircache module has been removed, without a replacement.
EasyDialogs module has been removed. See
exceptions module has been removed because all of the
exceptions defined there are available as built-in classes.
htmllib module has been removed. See
mimetools, MimeWriter, mimify, multifile, and rfc822¶
rfc822 modules have been removed. See
sets module was deprecated in Python 2.6 and removed in
Python 3.0. Use the built-in types
statvfs module was deprecated in Python 2.6 and removed
in Python 3.0. See
os.statvfs() in the
These modules are still present in the standard library, but are deprecated and should not be used in new Python 3 programs.
Summary of Changes to Modules¶
abstractstaticmethod() decorators are deprecated. Combining
abstractmethod() with the
staticmethod() decorators works as expected
(Python issue 11610).
version argument to
ArgumentParser has been removed
in favor of a special
action type (Python issue 13248).
The old form passed
version as an argument.
parser = argparse.ArgumentParser(version='1.0')
The new form requires adding an explicit argument definition.
parser = argparse.ArgumentParser() parser.add_argument('--version', action='version', version='%(prog)s 1.0')
The option name and version format string can be modified to suit the needs of the application.
In Python 3.4, the version action was changed to print the version string to stdout instead of stderr (Python issue 18920).
'c' type used for character bytes in early version of Python 2
has been removed. Use
'B' for bytes instead.
'u' type for characters from Unicode strings has been
deprecated and will be removed in Python 4.0.
fromstring() have been renamed
frombytes() to remove ambiguity
(Python issue 8990).
atexit was updated to include a C implementation
(Python issue 1680961), a regression was introduced in the error
handling logic that caused only the summary of the exception to be
shown, without the traceback. This regression was fixed in Python 3.3
(Python issue 18776).
decodestring() have been renamed
decodebytes() respectively. The old
names still work as aliases, but are deprecated (Python issue 3613).
Two new encodings using 85-character alphabets have been
b85encode() implements an encoding used in Mercurial and
a85encode() implements the Ascii85 format used by PDF
files (Python issue 17618).
BZ2File instances now support the context manager protocol,
and do not need to be wrapped with
getstatusoutput() have been
commands has been deleted.
ConfigParser module has been renamed to
ConfigParser class was removed in favor of
SafeConfigParser which has in turn been renamed to
ConfigParser. The deprecated interpolation behavior is
read() method now supports an
encoding argument, so it
is no longer necessary to use
codecs to read configuration
files with Unicode values in them.
Using the old
RawConfigParser is discouraged. New projects
ConfigParser(interpolation=None) instead to achieve the
contextlib.nested() has been removed. Pass multiple context
managers to the same
with statement instead.
Instead of using the
next() method of a reader directly, use the
next() function to invoke the iterator properly.
Starting with Python 3.3, equality comparisons between naive and
datetime instances return
False instead of
TypeError (Python issue 15006).
Prior to Python 3.5, a
datetime.time object representing
midnight evaluated to
False when converted to a Boolean. This
behavior has been removed in Python 3.5 (Python issue 13936).
Python 3.3 incorporated a C implementation of
decimal based on
libmpdec. This change improved performance, but also includes some
API changes and behavior differences from the pure-Python
implementation. See the Python 3.3 release notes for
from_decimal() class methods are no
longer needed. Floating point and
Decimal values can be
passed directly to the
DEBUG_INSTANCE have been
removed. They are no longer needed to differentiate between new and
All of the translation functions in
gettext assume Unicode
input and output, and the Unicode variants such as
have been removed.
The new function
escape() implements a work-around for searching
for files with meta-characters in the name (Python issue 8402).
Under Python 3,
imaplib returns byte-strings encoded as
UTF-8. There is support for accepting Unicode strings and encoding
them automatically as outgoing commands are sent or as the username
and password for logging in to the server.
formatargvalues() have been
deprecated in favor of
signature() (Python issue 20438).
been replaced with versions of the built-in functions that return
iterables instead of
list objects (
ifilterfalse() has been renamed
json API was updated to only support
str and not
bytes because the JSON specification is defined using
logging module now includes a
lastResort logger that is
used if no other logging configuration is performed by an
application. This eliminates the need to configure logging in an
application solely to avoid having a user see error messages in case a
library imported by an application uses logging but the application
itself does not.
mailbox reads and writes mailbox files in binary mode, relying on the email package to parse messages. StringIO and text file input is deprecated (Python issue 9124).
Values returned from read APIs are byte strings, and need to be decoded before being treated as text.
div() function has been removed. Use either
truediv(), depending on the desired semantics.
repeat() function is removed. Use
are removed. Use
with slice indexes instead.
isCallable() has been removed. Use the abstract
The type checking functions
isNumberType() have been
removed. Use the relevant abstract base classes from
isinstance(obj, collections.Mapping) isinstance(obj, collections.Sequence) isinstance(obj, numbers.Number)
sequenceIncludes() function has been removed. Use
popen() is still present but deprecated and emits
warnings if used. Code using these functions should be rewritten to
subprocess instead to be more portable across operating
os.tmpfile() have been removed. Use the
os.stat_float_times() is deprecated
(Python issue 14711).
os.unsetenv() no longer ignores errors (Python issue 13415).
os.path.walk() has been removed. Use
print() function (Python issue 18764). The
The C implementation of the pickle module from Python 2 has been moved
to a new module that is automatically used to replace the Python
implementation when possible. The old import idiom of looking for
pickle is no longer needed.
try: import cPickle as pickle except: import pickle
With the automatic import of the C implementation, it is only
necessary to import the
pickle module directly.
Interoperability between Python 2.x and 3.x has been improved for pickled data using the level 2 protocol or lower to resolve an issue introduced when a large number of standard library modules were renamed during the transition to Python 3. Because pickled data includes references to class and type names, and those names changed, it was difficult to exchange pickled data between Python 2 and 3 programs. Now for data pickled using protocol level 2 or older, the old names of the classes are automatically used when writing to and reading from a pickle stream.
This behavior is available by default, and can be turned off using the
fix_imports option. This change improves the situation, but does
not eliminate incompatibilities entirely. In particular, it is
possible that data pickled under Python 3.1 can’t be read under Python
3.0. To ensure maximum portability between Python 3 applications, use
protocol level 3, which does not include this compatibility feature.
The default protocol version has changed from
human-readable version, to
3, the binary format with the best
interoperability when shared between Python 3 applications.
Byte string data written to a pickle by a Python 2.x application is
decoded when it is read back to create a Unicode string object. The
encoding for the transformation defaults to ASCII, and can be changed
by passing values to the
platform.popen() has been deprecated. Use
instead (Python issue 11377).
platform.uname() now returns a
Because Linux distributions do not have a consistent way to describe
themselves, the functions for getting the descriptions
deprecated and scheduled to be removed in Python 3.7
(Python issue 1322).
jumpahead() was removed in Python 3.0.
UNICODE flag represents the default behavior. To restore
the ASCII-specific behavior from Python 2, use the
The default output format for
shelve may create a file with a
.db extension added to the name given to
PEP 475 means that system calls interrupted and returning with
EINTR are retried. This changes the behavior of signal handlers
and other system calls, since now after the signal handler returns the
interrupted call will be retried, unless the signal handler raises an
exception. Refer to the PEP documentation for complete details.
Under Python 2 typically
str objects could be sent directly
over a socket. Because
Python 3 the values must be encoded before being sent. The examples in
socket section use byte strings, which are already encoded.
All functions from the
string module that are also methods of
str objects have been removed.
uppercase have been removed. The new constants with similar
names are limited to the ASCII character set.
maketrans() function has been replaced by methods on
bytearray to clarify which
input types are supported by each translation table.
struct.pack() now only supports byte strings when using the
s string pack code, and no longer implicitly encodes string
objects to UTF-8 (Python issue 10783).
The default value for the
close_fds argument to
subprocess.Popen has changed from always being
always defaults to
True under Unix. It defaults to
Windows if the standard I/O stream arguments are set to
otherwise it defaults to
sys.exitfunc is no longer checked for a clean-up
action to be run when a program exits. Use
sys.subversion is no longer defined.
sys.flags.unicode are no
sys.maxint is no longer defined, use
sys.maxsize instead. See PEP 237 (Unifying Long Integers
The global exception tracking variables
sys.exc_traceback have been
removed. The function
sys.exc_clear() has also been removed.
sys.version_info is now a :py``namedtuple``
instance with attributes
serial (Python issue 4285).
The check interval feature, controlling the number of opcodes to
execute before allowing a thread context switch has been replaced with
an absolute time value instead, managed with
sys.setswitchinterval(). The old functions for managing the
sys.setcheckinterval(), are deprecated.
sys.path_hooks variables now
expose all of the path finders and entry hooks for importing
modules. In earlier versions, only finders and hooks explicitly added
to the path were exposed, and the C import used values in its
implementation that could not be modified from the outside.
For Linux systems,
sys.platform no longer includes the version
number. The value is now just
linux and not
thread module is deprecated in favor of the API in
Older implementations of
threading used factory functions for
some of the classes because they were implemented in C as extension
types and could not be subclassed. That limitation of the language has
been removed, and so many of the old factory functions have been
converted to standard classes, which allow subclassing
(Python issue 10968).
time.ctime() have been reimplemented to
not use the system functions of the same time to allow larger years to
time.ctime() now supports years from 1900 through
maxint, although for values higher than
9999 the output
string is longer than the standard 24 characters to allow for the
extra year digits (Python issue 8013).
TestCase methods starting with “fail” (
failUnless(), etc.) have been deprecated. Use the alternate form
of the assert methods instead.
Several older method aliases have been deprecated and replaced with preferred names. Using the deprecated names produces a warning (Python issue 9424). See the table below for a mapping between the old and new names.
|Deprecated Name||Preferred Name|
UserDict, UserList, and UserString¶
The UserDict, UserList, and UserString classes have been moved out of
their own modules into the
str can be subclassed directly, but the
collections may make implementing the subclass
simpler because the content of the container is available directly
through an instance attribute. The abstract classes in
collections.abc are also useful for creating custom containers
that follow the APIs of the built-in types.
uuid.getnode() now uses the
PATH environment variable to
find programs that can report the MAC address of the host under Unix
(Python issue 19855). It falls back to looking in
/usr/sbin if no program is found on the search path. This search
behavior may give different results than with earlier versions of
Python if alternate versions of programs like
arp are present and produce different
XMLTreeBuilder has been renamed
TreeBuilder, and the
API has undergone several changes.
ElementTree.getchildren() has been deprecated. Use
list(elem) to build a list of the children.
ElementTree.getiterator() has been deprecated. Use
to create an iterator using the normal iterator protocol
When parsing fails, rather than raising
XMLParser now raises
The data returned from
get_data() is a byte string, and needs to
be decoded before being used as a unicode string.