Differences between revisions 31 and 55 (spanning 24 versions)
Revision 31 as of 2010-03-12 18:53:00
Size: 8268
Editor: drclaw
Comment: added a link to the python 2 and 3 metaclasses page
Revision 55 as of 2013-10-24 23:04:45
Size: 11701
Editor: Ed Schofield
Comment: More formatting and readability improvements
Deletions are marked like this. Additions are marked like this.
Line 4: Line 4:
Porting Python Code to 3.0 Porting Python Code to 3.x
Line 9: Line 9:
* drop support for Python older than 2.6
* maintain separate releases for Python 2 and Python 3
* support Python 2 and Python 3 simultaneously from one code base

Each approach has its strengths and weaknesses.


Dropping Support for Python Older than 2.6
==========================================

Make sure the code runs in Python 2.6 and use 2to3

2to3 is a Python program that reads Python 2.x source code and applies a series of fixers to transform it into valid Python 3.x code. The standard library contains a rich set of fixers that will handle almost all code. 2to3 supporting library lib2to3 is, however, a flexible and generic library, so it is possible to write your own fixers for 2to3. lib2to3 could also be adapted to custom applications in which Python code needs to be edited automatically. For more information about 2to3, see: http://doc.python.org/library/2to3.html

Python 2.6 introduces forward-compatibility for many new Python 3 features. The Python 3 builtins are accessible by doing an import from `future_builtins`. To use the new print function, use `from __future__ import print_function`, and to make string literals be interpreted as unicode, use `from __future__ import unicode_literals`. These forward-compatability features, and many others, are described in greater detail in `What's New in Python 2.6`_:

.. _What's New in Python 2.6: http://docs.python.org/whatsnew/2.6.html

Manual changes (not done by 2to3):

* os.path.walk => os.walk: see issue4601_.
* rfc822 => email

.. _issue4601: http://bugs.python.org/issue4601

Strings and Bytes
-----------------

Design decisions needs to be taken what exactly must be represented as bytes, and what as strings (Unicode data). In many cases, this is easy. However, data sent or received over pipes or sockets are usually in binary mode, so explicit conversions may be necessary. For example, the Postgres API requires SQL queries to be transmitted in the connection encoding. It is probably easiest to convert the queries to the connection encoding as early as possible.

The standard IO streams (sys.stdin, sys.stdout, sys.stderr), are in text mode by default in Python 3. Calling `sys.stdout.write(some_binary_data)` will fail. To write binary data to standard out, you must do the following::

    sys.stdout.flush()
    sys.stdout.buffer.write(some_binary_data)

The flush() clears out any text (unicode) data that is stored in the TextIOWrapper, and the buffer attribute provides access to the lower-level binary stream. If you are only writing binary data, you can remove the text layer with `sys.stdout = sys.stdout.detach()` in Python 3.1; in Python 3.0, do `sys.stdout = sys.stdout.buffer`.


Maintain Separate Releases for Python 2 and Python 3
====================================================

Convert a Python 2 tree to Python 3 with the 2to3 tool: http://doc.python.org/library/2to3.html


Support Python 2 and Python 3 Simultaneously
============================================

Supporting code that runs in both Python 2.6 and Python 3 is not much more difficult than porting to Python 3 (see above for more details). However, supporting older versions--like 2.3, 2.4, and 2.5--is much more difficult. This requires avoiding any syntax that only works in one version or the other, so it often involves resorting to hacks. These hacks make the code more awkward than it would be in Python 3 or in Python 2, but many people find that this is better than supporting only one version or the other. When support for Python 2 is eventually dropped, these hacks can be removed.

Additional resources:

* http://pydev.blogspot.com/2008/11/making-code-work-in-python-2-and-3.html
* http://mail.mems-exchange.org/durusmail/qp/441/
* http://mikewatkins.ca/2008/11/29/python-2-and-3-metaclasses/
1) make code run unmodified in both Python 2 and Python 3
2) maintain a Python 2 base and use 2to3_ to generate Python 3 code
3) maintain a Python 3 base and use 3to2_ to generate Python 2 code

Each approach has its strengths and weaknesses. Beyond this document, the approaches are also discussed in the official `HOWTO on porting Python 2 code to Python 3 <http://docs.python.org/py3k/howto/pyporting.html>`_.



Approach 1: Make code run unmodified in both Python 2 and Python 3
==================================================================

Supporting code that runs in both Python 2.6 and Python 3 is now not much more difficult than porting to Python 3 and leaving behind Python 2.6.

Python 2.6 and 2.7 provide forward-compatibility for some of the new syntax in Python 3 through ``__future__`` import statements::

    from __future__ import (absolute_import, division,
                            print_function, unicode_literals)

Beyond ``__future__`` imports, two 3rd-party packages that greatly ease the task of supporting both Python 2 and Python 3 in a single codebase are:

* `future <http://python-future.org>`_. `future` makes it possible to support both Python versions from one clean, hack-free codebase.
* `six <http://pythonhosted.org/six/>`_. `six` is a thinner layer that also supports older versions--like 2.3, 2.4, and 2.5--which is very awkward without `six`.

For more information about the many forward-compatibility features that were added in Python 2.6 and Python 2.7, see these pages: `What's New in Python 2.6 <http://docs.python.org/whatsnew/2.6.html>`_ and `What's New in Python 2.7 <http://docs.python.org/whatsnew/2.7.html>`_.
Line 68: Line 39:
As of Python 2.6 the 2.x line includes a "bytes = str" alias in the builtins. Along with the bytes literal syntax, this allows binary data stored in a str instance to be clearly flagged so that it will use the correct type when the code is run in 3.x (either directly or via the 2to3 conversion tool). The reason it is done this way rather than backporting the 3.x bytes type is that most 2.x APIs that expect immutable binary data expect it as an 8-bit str instance - trying to pass in a backported 3.x bytes type wouldn't have the desired effect. To support versions earlier than 2.6, it is possible to define the alias at the top of the module:: As of Python 2.6 the 2.x line includes a "bytes = str" alias in the builtins. Along with the bytes literal syntax, this allows binary data stored in a ``str`` instance to be clearly flagged so that it will use the correct type when the code is run in 3.x (either directly or via the 2to3_ conversion tool). The reason it is done this way rather than backporting the 3.x ``bytes`` type is that most 2.x APIs that expect immutable binary data expect it as an 8-bit ``str`` instance.

`future <http://python
-future.org>`_ solves this problem by providing `backports <http://python-future.org/what_else.html#bytes>`_ of the ``bytes`` and ``str`` objects from Python 3 that inherit from Python 2's ``str`` and ``unicode`` objects, respectively.

To support versions earlier than 2.6, it is possible to define the alias at the top of the module::
Line 83: Line 58:
Supporting older Python versions
--------------------------------
Line 84: Line 62:
------------------------

Since `print` is a statement in Python 2 and a function in Python 3, it cannot be used in code that runs in both Python 2.5 and Python 3. Instead, use the `write` method. Replace this::
~~~~~~~~~~~~~~~~~~~~~~~~

Since ``print`` is a statement in Python 2 and a function in Python 3, the ``print`` statement cannot be used in code that runs in both Python 2.5 and Python 3.

The solution for supporting both Python 2.6/2.7 and Python 3 is to use ``print`` as a function together with the ``__future__`` import::

    from __future__ import print_function
    print('hello world')

To support Python versions older than 2.6, you can instead use the `print_` function provided by `six`, or you can use the ``sys.stdout.write`` method as follows, replacing this::
Line 95: Line 80:
or if you really like to be tricky, with this::

    print('hello world')

This uses the print function in Python 3 but still works with the print statement in Python 2. Just don't use anything fancy, like writing to a file with `>>`.
Line 103: Line 82:
----------

Python before 2.6 does not have the `as` keyword, so Python 2 and Python 3 have incompatible syntax for accessing the value of an exception. Compatible code should use the following idiom to save the value of an exception::
~~~~~~~~~~

Python 2.6 introduced the ``as`` keyword for exceptions which is used in Python 3::

    try:
        1 / 0
    except ZeroDivisionError as e:
        pass

Python versions <2.6, which do not support the ``as`` keyword, have incompatible syntax with Python 3 for accessing the value of an exception. To support older Python versions, you can use the following idiom to save the value of an exception::
Line 113: Line 99:

Line 116: Line 104:
Python 3 makes a distinction between relative and absolute imports. In Python 2.5, use `from __future__ import absolute_import` to get the same behavior as Python 3. To support older versions as well, only use absolute imports. Replace a relative import:: Python 3 makes a distinction between relative and absolute imports, dropping support for implicit relative imports. In Python 2.5+, use ``from __future__ import absolute_import`` to get the same behavior as Python 3. To support older versions as well, only use absolute imports. Replace a relative import::
Line 131: Line 119:
Approach 2: Maintain a Python 2 base and use 2to3 to generate Python 3 code
============================================================================


Make sure the code runs in Python 2.6 and use 2to3_

This approach does *not* require that support for versions before 2.6 needs to be dropped.
Even the need to test with 2.6 is not strict - one could just as well use this approach on
code that has never been tested with 2.6, and is only known to work on 2.5.

2to3_ is a Python program that reads Python 2.x source code and applies a series of fixers to transform it into valid Python 3.x code. The standard library contains a rich set of fixers that will handle almost all code. 2to3_ supporting library *lib2to3* is, however, a flexible and generic library, so it is possible to write your own fixers for 2to3_. *lib2to3* could also be adapted to custom applications in which Python code needs to be edited automatically. For more information about 2to3_, see: http://doc.python.org/library/2to3.html

Usage of 2to3_ can be integrated into the installation process: 2to3_ can be run as a build step in distutils. With distutils, the following fragments are needed in a setup.py::

  try:
      from distutils.command.build_py import build_py_2to3 as build_py
  except ImportError:
      # 2.x
      from distutils.command.build_py import build_py
  ...
  setup(...
    cmdclass = {'build_py':build_py}
  )

This will leave intact all source files, and convert them to Python 3 in the build area, if setup.py is run on Python 3. 'setup.py install' will then copy the 3.x version of the code into the target directory. If run on 2.x, nothing will change at all (as 2.x doesn't provide the build_py_2to3 class).

For distribute, this approach can be expressed as::

  setup(...
    use_2to3=True
  )

Manual changes (not done by 2to3_):

* os.path.walk => os.walk: see issue4601_.
* rfc822 => email

.. _issue4601: http://bugs.python.org/issue4601


Strings and Bytes
-----------------

Design decisions needs to be taken what exactly must be represented as bytes, and what as strings (Unicode data). In many cases, this is easy. However, data sent or received over pipes or sockets are usually in binary mode, so explicit conversions may be necessary. For example, the Postgres API requires SQL queries to be transmitted in the connection encoding. It is probably easiest to convert the queries to the connection encoding as early as possible.

The standard IO streams (sys.stdin, sys.stdout, sys.stderr), are in text mode by default in Python 3. Calling `sys.stdout.write(some_binary_data)` will fail. To write binary data to standard out, you must do the following::

    sys.stdout.flush()
    sys.stdout.buffer.write(some_binary_data)

The flush() clears out any text (unicode) data that is stored in the TextIOWrapper, and the buffer attribute provides access to the lower-level binary stream. If you are only writing binary data, you can remove the text layer with `sys.stdout = sys.stdout.detach()` in Python 3.1; in Python 3.0, do `sys.stdout = sys.stdout.buffer`.


Approach 3: Maintain a Python 3 base and use `3to2` or `futurize` to generate Python 2 code
============================================================================================

`3to2` is a tool to help with automatically converting Python 3-only code into Python 2-only code. See 3to2_.

An alternative is to use the ``futurize`` script from the `future <http://python-future.org>`_ package with the `--from3 <http://python-future.org/automatic_conversion.html#backwards-3-to-both>`_ option. This extends `3to2` by adding imports from the `future` package to provide Python 2 compatibility from the same codebase.


Additional resources
====================

* http://python-future.org
* http://diveintopython3.org/porting-code-to-python-3-with-2to3.html
* http://packages.python.org/distribute/python3.html
* http://lucumr.pocoo.org/2010/2/11/porting-to-python-3-a-guide
* http://lucumr.pocoo.org/2011/1/22/forwards-compatible-python/
* http://techspot.zzzeek.org/2011/01/24/zzzeek-s-guide-to-python-3-porting
* http://dabeaz.blogspot.com/2011/01/porting-py65-and-my-superboard-to.html
* http://python3porting.com/strategies.html
* http://lwn.net/Articles/426906/


Line 132: Line 196:
============================ ----------------------------
Line 147: Line 211:
  not auto-converted with 2to3 support. I explicitly added a 2to3   not auto-converted with 2to3_ support. I explicitly added a 2to3_
Line 152: Line 216:
Links
=====
Differences in specific libraries
---------------------------------

These are porting notes on differences in third party libraries when run on Python 2 and Python 3.

* `PyQt4 </PyQt4>`_


More links
----------
Line 162: Line 234:
* A blog post about porting experience: `psycopg2 porting to Python 3: a report <http://initd.org/psycopg/articles/2011/01/24/psycopg2-porting-python-3-report/>`_
* Reference Card: Moving from Python 2 to Python 3: http://ptgmedia.pearsoncmg.com/imprint_downloads/informit/promotions/python/python2python3.pdf
* `Lennart Regebro's overview of Porting strategies <http://python3porting.com/strategies.html>`_

Porting Python Code to 3.x

There are three ways to support Python 3:

  1. make code run unmodified in both Python 2 and Python 3
  2. maintain a Python 2 base and use 2to3 to generate Python 3 code
  3. maintain a Python 3 base and use 3to2 to generate Python 2 code

Each approach has its strengths and weaknesses. Beyond this document, the approaches are also discussed in the official HOWTO on porting Python 2 code to Python 3.

Approach 1: Make code run unmodified in both Python 2 and Python 3

Supporting code that runs in both Python 2.6 and Python 3 is now not much more difficult than porting to Python 3 and leaving behind Python 2.6.

Python 2.6 and 2.7 provide forward-compatibility for some of the new syntax in Python 3 through __future__ import statements:

from __future__ import (absolute_import, division,
                        print_function, unicode_literals)

Beyond __future__ imports, two 3rd-party packages that greatly ease the task of supporting both Python 2 and Python 3 in a single codebase are:

  • future. future makes it possible to support both Python versions from one clean, hack-free codebase.
  • six. six is a thinner layer that also supports older versions--like 2.3, 2.4, and 2.5--which is very awkward without six.

For more information about the many forward-compatibility features that were added in Python 2.6 and Python 2.7, see these pages: What's New in Python 2.6 and What's New in Python 2.7.

Strings and Unicode

As of Python 2.6 the 2.x line includes a "bytes = str" alias in the builtins. Along with the bytes literal syntax, this allows binary data stored in a str instance to be clearly flagged so that it will use the correct type when the code is run in 3.x (either directly or via the 2to3 conversion tool). The reason it is done this way rather than backporting the 3.x bytes type is that most 2.x APIs that expect immutable binary data expect it as an 8-bit str instance.

future solves this problem by providing backports of the bytes and str objects from Python 3 that inherit from Python 2's str and unicode objects, respectively.

To support versions earlier than 2.6, it is possible to define the alias at the top of the module:

try:
  bytes # Forward compatibility with Py3k
except NameError:
  bytes = str

When using "from __future__ import unicode_literals" in a module, it may also be useful to insert "str = unicode" near the top of the module. This will ensure that "isinstance('', str)" remains true in that module:

try:
  str = unicode
except NameError:
  pass # Forward compatibility with Py3k

Supporting older Python versions

Exceptions

Python 2.6 introduced the as keyword for exceptions which is used in Python 3:

try:
    1 / 0
except ZeroDivisionError as e:
    pass

Python versions <2.6, which do not support the as keyword, have incompatible syntax with Python 3 for accessing the value of an exception. To support older Python versions, you can use the following idiom to save the value of an exception:

import sys
try:
    open('/path/to/some/file')
except IOError:
    _, e, _ = sys.exc_info()

Relative Imports

Python 3 makes a distinction between relative and absolute imports, dropping support for implicit relative imports. In Python 2.5+, use from __future__ import absolute_import to get the same behavior as Python 3. To support older versions as well, only use absolute imports. Replace a relative import:

from xyz import abc

with an absolute import:

from mypackage.xyz import abc

Integer Division

Make sure to use from __future__ import division (introduced in Python 2.2) to get the non-truncating behavior, which is default in Python 3.

Approach 2: Maintain a Python 2 base and use 2to3 to generate Python 3 code

Make sure the code runs in Python 2.6 and use 2to3

This approach does not require that support for versions before 2.6 needs to be dropped. Even the need to test with 2.6 is not strict - one could just as well use this approach on code that has never been tested with 2.6, and is only known to work on 2.5.

2to3 is a Python program that reads Python 2.x source code and applies a series of fixers to transform it into valid Python 3.x code. The standard library contains a rich set of fixers that will handle almost all code. 2to3 supporting library lib2to3 is, however, a flexible and generic library, so it is possible to write your own fixers for 2to3. lib2to3 could also be adapted to custom applications in which Python code needs to be edited automatically. For more information about 2to3, see: http://doc.python.org/library/2to3.html

Usage of 2to3 can be integrated into the installation process: 2to3 can be run as a build step in distutils. With distutils, the following fragments are needed in a setup.py:

try:
    from distutils.command.build_py import build_py_2to3 as build_py
except ImportError:
    # 2.x
    from distutils.command.build_py import build_py
...
setup(...
  cmdclass = {'build_py':build_py}
)

This will leave intact all source files, and convert them to Python 3 in the build area, if setup.py is run on Python 3. 'setup.py install' will then copy the 3.x version of the code into the target directory. If run on 2.x, nothing will change at all (as 2.x doesn't provide the build_py_2to3 class).

For distribute, this approach can be expressed as:

setup(...
  use_2to3=True
)

Manual changes (not done by 2to3):

  • os.path.walk => os.walk: see issue4601.
  • rfc822 => email

Strings and Bytes

Design decisions needs to be taken what exactly must be represented as bytes, and what as strings (Unicode data). In many cases, this is easy. However, data sent or received over pipes or sockets are usually in binary mode, so explicit conversions may be necessary. For example, the Postgres API requires SQL queries to be transmitted in the connection encoding. It is probably easiest to convert the queries to the connection encoding as early as possible.

The standard IO streams (sys.stdin, sys.stdout, sys.stderr), are in text mode by default in Python 3. Calling sys.stdout.write(some_binary_data) will fail. To write binary data to standard out, you must do the following:

sys.stdout.flush()
sys.stdout.buffer.write(some_binary_data)

The flush() clears out any text (unicode) data that is stored in the TextIOWrapper, and the buffer attribute provides access to the lower-level binary stream. If you are only writing binary data, you can remove the text layer with sys.stdout = sys.stdout.detach() in Python 3.1; in Python 3.0, do sys.stdout = sys.stdout.buffer.

Approach 3: Maintain a Python 3 base and use 3to2 or futurize to generate Python 2 code

3to2 is a tool to help with automatically converting Python 3-only code into Python 2-only code. See 3to2.

An alternative is to use the futurize script from the future package with the --from3 option. This extends 3to2 by adding imports from the future package to provide Python 2 compatibility from the same codebase.

Additional resources

Martin's notes from psycopg2

  • the buffer object is gone; I use memoryview in 3.x.
  • various tests where in the code of the form if version_major == and version_minor > 4 (say, or 5) This will break for 3.x; you have to write if (version_major == 2 and version_minor > 4) or version_major > 2
  • Python code 2: setup.py needs to run in both versions. I had to replace popen2 with subprocess if available. Also, map() now returns an iterator, which I explicitly convert into list, and so on.
  • Python code 3: the test suite doesn't get installed, and hence not auto-converted with 2to3 support. I explicitly added a 2to3 conversion into the test runner, which copies the py3 version of the test into a separate directory.

Differences in specific libraries

These are porting notes on differences in third party libraries when run on Python 2 and Python 3.

PortingPythonToPy3k (last edited 2019-10-26 22:24:27 by FrancesHocutt)

Unable to edit the page? See the FrontPage for instructions.