path: root/Doc/whatsnew/2.7.rst

****************************
  What's New in Python 2.7
****************************

:Author: A.M. Kuchling (amk at amk.ca)
:Release: |release|
:Date: |today|

.. Fix accents on Kristjan Valur Jonsson, Fuerstenau

.. Big jobs: ElementTree 1.3, pep 391, sysconfig
..  unittest test discovery
..  hyperlink all the methods & functions.

.. $Id$
   Rules for maintenance:

   * Anyone can add text to this document.  Do not spend very much time
   on the wording of your changes, because your text will probably
   get rewritten to some degree.

   * The maintainer will go through Misc/NEWS periodically and add
   changes; it's therefore more important to add your changes to
   Misc/NEWS than to this file.

   * This is not a complete list of every single change; completeness
   is the purpose of Misc/NEWS.  Some changes I consider too small
   or esoteric to include.  If such a change is added to the text,
   I'll just remove it.  (This is another reason you shouldn't spend
   too much time on writing your addition.)

   * If you want to draw your new text to the attention of the
   maintainer, add 'XXX' to the beginning of the paragraph or
   section.

   * It's OK to just add a fragmentary note about a change.  For
   example: "XXX Describe the transmogrify() function added to the
   socket module."  The maintainer will research the change and
   write the necessary text.

   * You can comment out your additions if you like, but it's not
   necessary (especially when a final release is some months away).

   * Credit the author of a patch or bugfix.  Just the name is
   sufficient; the e-mail address isn't necessary.

   * It's helpful to add the bug/patch number in a parenthetical comment.

   XXX Describe the transmogrify() function added to the socket
   module.
   (Contributed by P.Y. Developer; :issue:`12345`.)

   This saves the maintainer some effort going through the SVN logs
   when researching a change.

This article explains the new features in Python 2.7.  The final
release of 2.7 is currently scheduled for June 2010; the detailed
schedule is described in :pep:`373`.

Python 2.7 is planned to be the last major release in the 2.x series.
Though more major releases have not been absolutely ruled out, the
Python maintainers are planning to focus more on Python 3.x.  Despite
that, it's likely that the 2.7 release will have a longer period of
maintenance compared to earlier 2.x versions.

.. Compare with previous release in 2 - 3 sentences here.
   add hyperlink when the documentation becomes available online.

.. _whatsnew27-python31:

Python 3.1 Features
=======================

Much as Python 2.6 incorporated features from Python 3.0,
version 2.7 incorporates some of the new features
in Python 3.1.  The 2.x series continues to provide tools
for migrating to the 3.x series.

A partial list of 3.1 features that were backported to 2.7:

* A version of the :mod:`io` library, rewritten in C for performance.
* The ordered-dictionary type described in :ref:`pep-0372`.
* The new format specifier described in :ref:`pep-0378`.
* The :class:`memoryview` object.
* A small subset of the :mod:`importlib` module `described below <#importlib-section>`__.
* Float-to-string and string-to-float conversions now round their
  results more correctly.  And :func:`repr` of a floating-point
  number *x* returns a result that's guaranteed to round back to the
  same number when converted back to a string.
* The :cfunc:`PyLong_AsLongAndOverflow` C API function.

One porting change: the :option:`-3` switch now automatically
enables the :option:`-Qwarn` switch that causes warnings
about using classic division with integers and long integers.

Other new Python3-mode warnings include:

* :func:`operator.isCallable` and :func:`operator.sequenceIncludes`,
  which are not supported in 3.x.

.. ========================================================================
.. Large, PEP-level features and changes should be described here.
.. ========================================================================

.. _pep-0372:

PEP 372: Adding an ordered dictionary to collections
====================================================

Regular Python dictionaries iterate over key/value pairs in arbitrary order.
Over the years, a number of authors have written alternative implementations
that remember the order that the keys were originally inserted.  Based on
the experiences from those implementations, a new
:class:`~collections.OrderedDict` class has been introduced in the
:mod:`collections` module.

The :class:`~collections.OrderedDict` API is substantially the same as regular
dictionaries but will iterate over keys and values in a guaranteed order
depending on when a key was first inserted::

    >>> from collections import OrderedDict
    >>> d = OrderedDict([('first', 1), ('second', 2),
    ...                  ('third', 3)])
    >>> d.items()
    [('first', 1), ('second', 2), ('third', 3)]

If a new entry overwrites an existing entry, the original insertion
position is left unchanged::

    >>> d['second'] = 4
    >>> d.items()
    [('first', 1), ('second', 4), ('third', 3)]

Deleting an entry and reinserting it will move it to the end::

    >>> del d['second']
    >>> d['second'] = 5
    >>> d.items()
    [('first', 1), ('third', 3), ('second', 5)]

The :meth:`~collections.OrderedDict.popitem` method has an optional *last*
argument that defaults to True.  If *last* is True, the most recently
added key is returned and removed; if it's False, the
oldest key is selected::

    >>> od = OrderedDict([(x,0) for x in range(20)])
    >>> od.popitem()
    (19, 0)
    >>> od.popitem()
    (18, 0)
    >>> od.popitem(last=False)
    (0, 0)
    >>> od.popitem(last=False)
    (1, 0)

Comparing two ordered dictionaries checks both the keys and values,
and requires that the insertion order was the same::

    >>> od1 = OrderedDict([('first', 1), ('second', 2),
    ...                    ('third', 3)])
    >>> od2 = OrderedDict([('third', 3), ('first', 1),
    ...                    ('second', 2)])
    >>> od1 == od2
    False
    >>> # Move 'third' key to the end
    >>> del od2['third']; od2['third'] = 3
    >>> od1 == od2
    True

Comparing an :class:`~collections.OrderedDict` with a regular dictionary
ignores the insertion order and just compares the keys and values.

How does the :class:`~collections.OrderedDict` work?  It maintains a
doubly-linked list of keys, appending new keys to the list as they're inserted.
A secondary dictionary maps keys to their corresponding list node, so
deletion doesn't have to traverse the entire linked list and therefore
remains O(1).

.. XXX check O(1)-ness with Raymond
..     Also check if the 'somenamedtuple' in the collection module should
..     be replaced/removed in order to use
..     :meth:`~collections.namedtuple._asdict()` (see below)

The standard library now supports use of ordered dictionaries in several
modules.

* The :mod:`ConfigParser` module uses them by default, letting
  configuration files be read, modified, and then written back in their original
  order.

* The :meth:`~collections.somenamedtuple._asdict()` method for
  :func:`collections.namedtuple` now returns an ordered dictionary with the
  values appearing in the same order as the underlying tuple indices.

* The :mod:`json` module's :class:`~json.JSONDecoder` class
  constructor was extended with an *object_pairs_hook* parameter to
  allow :class:`OrderedDict` instances to be built by the decoder.
  Support was also added for third-party tools like
  `PyYAML <http://pyyaml.org/>`_.

.. seealso::

   :pep:`372` - Adding an ordered dictionary to collections
     PEP written by Armin Ronacher and Raymond Hettinger;
     implemented by Raymond Hettinger.

.. _pep-0378:

PEP 378: Format Specifier for Thousands Separator
=================================================

To make program output more readable, it can be useful to add
separators to large numbers and render them as
18,446,744,073,709,551,616 instead of 18446744073709551616.

The fully general solution for doing this is the :mod:`locale` module,
which can use different separators ("," in North America, "." in
Europe) and different grouping sizes, but :mod:`locale` is complicated
to use and unsuitable for multi-threaded applications where different
threads are producing output for different locales.

Therefore, a simple comma-grouping mechanism has been added to the
mini-language used by the :meth:`str.format` method.  When
formatting a floating-point number, simply include a comma between the
width and the precision::

   >>> '{:20,.2f}'.format(18446744073709551616.0)
   '18,446,744,073,709,551,616.00'

When formatting an integer, include the comma after the width:

   >>> '{:20,d}'.format(18446744073709551616)
   '18,446,744,073,709,551,616'

This mechanism is not adaptable at all; commas are always used as the
separator and the grouping is always into three-digit groups.  The
comma-formatting mechanism isn't as general as the :mod:`locale`
module, but it's easier to use.

.. XXX "Format String Syntax" in string.rst could use many more examples.

.. seealso::

   :pep:`378` - Format Specifier for Thousands Separator
     PEP written by Raymond Hettinger; implemented by Eric Smith.

PEP 389: The argparse Module for Parsing Command Lines
======================================================

The :mod:`argparse` module for parsing command-line arguments was
added, intended as a more powerful replacement for the
:mod:`optparse` module.

This means Python now supports three different modules for parsing
command-line arguments: :mod:`getopt`, :mod:`optparse`, and
:mod:`argparse`.  The :mod:`getopt` module closely resembles the C
:cfunc:`getopt` function, so it remains useful if you're writing a
Python prototype that will eventually be rewritten in C.
:mod:`optparse` becomes redundant, but there are no plans to remove it
because there are many scripts still using it, and there's no
automated way to update these scripts.  (Making the :mod:`argparse`
API consistent with :mod:`optparse`'s interface was discussed but
rejected as too messy and difficult.)

In short, if you're writing a new script and don't need to worry
about compatibility with earlier versions of Python, use
:mod:`argparse` instead of :mod:`optparse`.

Here's an example::

    import argparse

    parser = argparse.ArgumentParser(description='Command-line example.')

    # Add optional switches
    parser.add_argument('-v', action='store_true', dest='is_verbose',
                        help='produce verbose output')
    parser.add_argument('-o', action='store', dest='output',
                        metavar='FILE',
                        help='direct output to FILE instead of stdout')
    parser.add_argument('-C', action='store', type=int, dest='context',
                        metavar='NUM', default=0,
                        help='display NUM lines of added context')

    # Allow any number of additional arguments.
    parser.add_argument(nargs='*', action='store', dest='inputs',
                        help='input filenames (default is stdin)')

    args = parser.parse_args()
    print args.__dict__

Unless you override it, :option:`-h` and :option:`--help` switches
are automatically added, and produce neatly formatted output::

    -> ./python.exe argparse-example.py --help
    usage: argparse-example.py [-h] [-v] [-o FILE] [-C NUM] [inputs [inputs ...]]

    Command-line example.

    positional arguments:
      inputs      input filenames (default is stdin)

    optional arguments:
      -h, --help  show this help message and exit
      -v          produce verbose output
      -o FILE     direct output to FILE instead of stdout
      -C NUM      display NUM lines of added context

Similarly to :mod:`optparse`, the command-line switches and arguments
are returned as an object with attributes named by the *dest* parameters::

    -> ./python.exe argparse-example.py -v
    {'output': None, 'is_verbose': True, 'context': 0, 'inputs': []}

    -> ./python.exe argparse-example.py -v -o /tmp/output -C 4 file1 file2
    {'output': '/tmp/output', 'is_verbose': True, 'context': 4,
     'inputs': ['file1', 'file2']}

:mod:`argparse` has much fancier validation than :mod:`optparse`; you
can specify an exact number of arguments as an integer, 0 or more
arguments by passing ``'*'``, 1 or more by passing ``'+'``, or an
optional argument with ``'?'``.  A top-level parser can contain
sub-parsers, so you can define subcommands that have different sets of
switches, as in ``svn commit``, ``svn checkout``, etc.  You can
specify an argument type as :class:`~argparse.FileType`, which will
automatically open files for you and understands that ``'-'`` means
standard input or output.

.. seealso::

   `argparse module documentation <http://docs.python.org/dev/library/argparse.html>`__

   `Upgrading optparse code to use argparse <http://docs.python.org/dev/library/argparse.html#upgrading-optparse-code>`__

   :pep:`389` - argparse - New Command Line Parsing Module
     PEP written and implemented by Steven Bethard.

PEP 391: Dictionary-Based Configuration For Logging
====================================================

.. not documented in library reference yet.

The :mod:`logging` module is very flexible; an application can define
a tree of logging subsystems, and each logger in this tree can filter
out certain messages, format them differently, and direct messages to
a varying number of handlers.

All this flexibility can require a lot of configuration.  You can
write Python statements to create objects and set their properties,
but a complex set-up would require verbose but boring code.
:mod:`logging` also supports a :func:`~logging.config.fileConfig`
function that parses a file, but the file format doesn't support
configuring filters, and it's messier to generate programmatically.

Python 2.7 adds a :func:`~logging.config.dictConfig` function that
uses a dictionary, and there are many ways to produce a dictionary
from different sources.  You can construct one with code, of course.
Python's standard library now includes a JSON parser, so you could
parse a file containing JSON, or you could use a YAML parsing library
if one is installed.

XXX describe an example.

Two smaller enhancements to the logging module are:

.. rev79293

* :class:`Logger` instances gained a :meth:`getChild` that retrieves a
  descendant logger using a relative path.  For example,
  once you retrieve a logger by doing ``log = getLogger('app')``,
  calling ``log.getChild('network.listen')`` is equivalent to
  ``getLogger('app.network.listen')``.

* The :class:`LoggerAdapter` class gained a :meth:`isEnabledFor` method
  that takes a *level* and returns whether the underlying logger would
  process a message of that level of importance.

.. seealso::

   :pep:`391` - Dictionary-Based Configuration For Logging
     PEP written and implemented by Vinay Sajip.

PEP 3106: Dictionary Views
====================================================

The dictionary methods :meth:`keys`, :meth:`values`, and :meth:`items`
are different in Python 3.x.  They return an object called a :dfn:`view`
instead of a fully materialized list.

.. Views can be iterated over, but they also behave like sets.  XXX not working.

It's not possible to change the return values of :meth:`keys`,
:meth:`values`, and :meth:`items` in Python 2.7 because too much code
would break.  Instead the 3.x versions were added under the new names
of :meth:`viewkeys`, :meth:`viewvalues`, and :meth:`viewitems`.

::

    >>> d = dict((i*10, chr(65+i)) for i in range(26))
    >>> d
    {0: 'A', 130: 'N', 10: 'B', 140: 'O', 20: ..., 250: 'Z'}
    >>> d.viewkeys()
    dict_keys([0, 130, 10, 140, 20, 150, 30, ..., 250])

The view keeps track of the dictionary and its contents change as the
dictionary is modified::

    >>> vk = d.viewkeys()
    >>> vk
    dict_keys([0, 130, 10, ..., 250])
    >>> d[260] = '&'
    >>> vk
    dict_keys([0, 130, 260, 10, ..., 250])

However, note that you can't add or remove keys while you're iterating
over the view::

    >>> for k in vk:
    ...     d[k*2] = k
    ...
    Traceback (most recent call last):
      File "<stdin>", line 1, in <module>
    RuntimeError: dictionary changed size during iteration

You can use the view methods in Python 2.x code, and the 2to3
converter will change them to the standard :meth:`keys`,
:meth:`values`, and :meth:`items` methods.

.. seealso::

   :pep:`3106` - Revamping dict.keys(), .values() and .items()
     PEP written by Guido van Rossum.
     Backported to 2.7 by Alexandre Vassalotti; :issue:`1967`.


Other Language Changes
======================

Some smaller changes made to the core Python language are:

* The syntax for set literals has been backported from Python 3.x.
  Curly brackets are used to surround the contents of the resulting
  mutable set; set literals are
  distinguished from dictionaries by not containing colons and values.
  ``{}`` continues to represent an empty dictionary; use
  ``set()`` for an empty set.

    >>> {1,2,3,4,5}
    set([1, 2, 3, 4, 5])
    >>> set() # empty set
    set([])
    >>> {}    # empty dict
    {}

  Backported by Alexandre Vassalotti; :issue:`2335`.

* Dictionary and set comprehensions are another feature backported from
  3.x, generalizing list/generator comprehensions to use
  the literal syntax for sets and dictionaries.

    >>> {x:x*x for x in range(6)}
    {0: 0, 1: 1, 2: 4, 3: 9, 4: 16, 5: 25}
    >>> {'a'*x for x in range(6)}
    set(['', 'a', 'aa', 'aaa', 'aaaa', 'aaaaa'])

  Backported by Alexandre Vassalotti; :issue:`2333`.

* The :keyword:`with` statement can now use multiple context managers
  in one statement.  Context managers are processed from left to right
  and each one is treated as beginning a new :keyword:`with` statement.
  This means that::

   with A() as a, B() as b:
       ... suite of statements ...

  is equivalent to::

   with A() as a:
       with B() as b:
           ... suite of statements ...

  The :func:`contextlib.nested` function provides a very similar
  function, so it's no longer necessary and has been deprecated.

  (Proposed in http://codereview.appspot.com/53094; implemented by
  Georg Brandl.)

* Conversions between floating-point numbers and strings are
  now correctly rounded on most platforms.  These conversions occur
  in many different places: :func:`str` on
  floats and complex numbers; the :class:`float` and :class:`complex`
  constructors;
  numeric formatting; serialization and
  deserialization of floats and complex numbers using the
  :mod:`marshal`, :mod:`pickle`
  and :mod:`json` modules;
  parsing of float and imaginary literals in Python code;
  and :class:`~decimal.Decimal`-to-float conversion.

  Related to this, the :func:`repr` of a floating-point number *x*
  now returns a result based on the shortest decimal string that's
  guaranteed to round back to *x* under correct rounding (with
  round-half-to-even rounding mode).  Previously it gave a string
  based on rounding x to 17 decimal digits.

  .. maybe add an example?

  The rounding library responsible for this improvement works on
  Windows, and on Unix platforms using the gcc, icc, or suncc
  compilers.  There may be a small number of platforms where correct
  operation of this code cannot be guaranteed, so the code is not
  used on such systems.  You can find out which code is being used
  by checking :data:`sys.float_repr_style`,  which will be ``short``
  if the new code is in use and ``legacy`` if it isn't.

  Implemented by Eric Smith and Mark Dickinson, using David Gay's
  :file:`dtoa.c` library; :issue:`7117`.

* The :meth:`str.format` method now supports automatic numbering of the replacement
  fields.  This makes using :meth:`str.format` more closely resemble using
  ``%s`` formatting::

    >>> '{}:{}:{}'.format(2009, 04, 'Sunday')
    '2009:4:Sunday'
    >>> '{}:{}:{day}'.format(2009, 4, day='Sunday')
    '2009:4:Sunday'

  The auto-numbering takes the fields from left to right, so the first ``{...}``
  specifier will use the first argument to :meth:`str.format`, the next
  specifier will use the next argument, and so on.  You can't mix auto-numbering
  and explicit numbering -- either number all of your specifier fields or none
  of them -- but you can mix auto-numbering and named fields, as in the second
  example above.  (Contributed by Eric Smith; :issue:`5237`.)

  Complex numbers now correctly support usage with :func:`format`,
  and default to being right-aligned.
  Specifying a precision or comma-separation applies to both the real
  and imaginary parts of the number, but a specified field width and
  alignment is applied to the whole of the resulting ``1.5+3j``
  output.  (Contributed by Eric Smith; :issue:`1588` and :issue:`7988`.)

  The 'F' format code now always formats its output using uppercase characters,
  so it will now produce 'INF' and 'NAN'.
  (Contributed by Eric Smith; :issue:`3382`.)

* The :func:`int` and :func:`long` types gained a ``bit_length``
  method that returns the number of bits necessary to represent
  its argument in binary::

      >>> n = 37
      >>> bin(n)
      '0b100101'
      >>> n.bit_length()
      6
      >>> n = 2**123-1
      >>> n.bit_length()
      123
      >>> (n+1).bit_length()
      124

  (Contributed by Fredrik Johansson and Victor Stinner; :issue:`3439`.)

* Conversions from long integers and regular integers to floating
  point now round differently, returning the floating-point number
  closest to the number.  This doesn't matter for small integers that
  can be converted exactly, but for large numbers that will
  unavoidably lose precision, Python 2.7 now approximates more
  closely.  For example, Python 2.6 computed the following::

    >>> n = 295147905179352891391
    >>> float(n)
    2.9514790517935283e+20
    >>> n - long(float(n))
    65535L

  Python 2.7's floating-point result is larger, but much closer to the
  true value::

    >>> n = 295147905179352891391
    >>> float(n)
    2.9514790517935289e+20
    >>> n - long(float(n))
    -1L

  (Implemented by Mark Dickinson; :issue:`3166`.)

  Integer division is also more accurate in its rounding behaviours.  (Also
  implemented by Mark Dickinson; :issue:`1811`.)

* It's now possible for a subclass of the built-in :class:`unicode` type
  to override the :meth:`__unicode__` method.  (Implemented by
  Victor Stinner; :issue:`1583863`.)

* The :class:`bytearray` type's :meth:`~bytearray.translate` method now accepts
  ``None`` as its first argument.  (Fixed by Georg Brandl;
  :issue:`4759`.)

  .. bytearray doesn't seem to be documented

* When using ``@classmethod`` and ``@staticmethod`` to wrap
  methods as class or static methods, the wrapper object now
  exposes the wrapped function as their :attr:`__func__` attribute.
  (Contributed by Amaury Forgeot d'Arc, after a suggestion by
  George Sakkis; :issue:`5982`.)

* A new encoding named "cp720", used primarily for Arabic text, is now
  supported.  (Contributed by Alexander Belchenko and Amaury Forgeot
  d'Arc; :issue:`1616979`.)

* The :class:`file` object will now set the :attr:`filename` attribute
  on the :exc:`IOError` exception when trying to open a directory
  on POSIX platforms (noted by Jan Kaliszewski; :issue:`4764`), and
  now explicitly checks for and forbids writing to read-only file objects
  instead of trusting the C library to catch and report the error
  (fixed by Stefan Krah; :issue:`5677`).

* The Python tokenizer now translates line endings itself, so the
  :func:`compile` built-in function can now accept code using any
  line-ending convention.  Additionally, it no longer requires that the
  code end in a newline.

* Extra parentheses in function definitions are illegal in Python 3.x,
  meaning that you get a syntax error from ``def f((x)): pass``.  In
  Python3-warning mode, Python 2.7 will now warn about this odd usage.
  (Noted by James Lingard; :issue:`7362`.)

* It's now possible to create weak references to old-style class
  objects.  New-style classes were always weak-referenceable.  (Fixed
  by Antoine Pitrou; :issue:`8268`.)

* When a module object is garbage-collected, the module's dictionary is
  now only cleared if no one else is holding a reference to the
  dictionary (:issue:`7140`).

.. ======================================================================

.. _new-27-interpreter:

Interpreter Changes
-------------------------------

A new environment variable, :envvar:`PYTHONWARNINGS`,
allows controlling warnings.  It should be set to a string
containing warning settings, equivalent to those
used with the :option:`-W` switch, separated by commas.
(Contributed by Brian Curtin; :issue:`7301`.)

For example, the following setting will print warnings every time
they occur, but turn warnings from the :mod:`Cookie` module into an
error.  (The exact syntax for setting an environment variable varies
across operating systems and shells, so it may be different for you.)

::

  export PYTHONWARNINGS=all,error:::Cookie:0


.. ======================================================================


Optimizations
-------------

Several performance enhancements have been added:

.. * A new :program:`configure` option, :option:`--with-computed-gotos`,
   compiles the main bytecode interpreter loop using a new dispatch
   mechanism that gives speedups of up to 20%, depending on the system
   and benchmark.  The new mechanism is only supported on certain
   compilers, such as gcc, SunPro, and icc.

* A new opcode was added to perform the initial setup for
  :keyword:`with` statements, looking up the :meth:`__enter__` and
  :meth:`__exit__` methods.  (Contributed by Benjamin Peterson.)

* The garbage collector now performs better for one common usage
  pattern: when many objects are being allocated without deallocating
  any of them.  This would previously take quadratic
  time for garbage collection, but now the number of full garbage collections
  is reduced as the number of objects on the heap grows.
  The new logic is to only perform a full garbage collection pass when
  the middle generation has been collected 10 times and when the
  number of survivor objects from the middle generation exceeds 10% of
  the number of objects in the oldest generation.  (Suggested by Martin
  von Löwis and implemented by Antoine Pitrou; :issue:`4074`.)

* The garbage collector tries to avoid tracking simple containers
  which can't be part of a cycle. In Python 2.7, this is now true for
  tuples and dicts containing atomic types (such as ints, strings,
  etc.). Transitively, a dict containing tuples of atomic types won't
  be tracked either. This helps reduce the cost of each
  garbage collection by decreasing the number of objects to be
  considered and traversed by the collector.
  (Contributed by Antoine Pitrou; :issue:`4688`.)

* Long integers are now stored internally either in base 2**15 or in base
  2**30, the base being determined at build time.  Previously, they
  were always stored in base 2**15.  Using base 2**30 gives
  significant performance improvements on 64-bit machines, but
  benchmark results on 32-bit machines have been mixed.  Therefore,
  the default is to use base 2**30 on 64-bit machines and base 2**15
  on 32-bit machines; on Unix, there's a new configure option
  :option:`--enable-big-digits` that can be used to override this default.

  Apart from the performance improvements this change should be
  invisible to end users, with one exception: for testing and
  debugging purposes there's a new structseq :data:`sys.long_info` that
  provides information about the internal format, giving the number of
  bits per digit and the size in bytes of the C type used to store
  each digit::

     >>> import sys
     >>> sys.long_info
     sys.long_info(bits_per_digit=30, sizeof_digit=4)

  (Contributed by Mark Dickinson; :issue:`4258`.)

  Another set of changes made long objects a few bytes smaller: 2 bytes
  smaller on 32-bit systems and 6 bytes on 64-bit.
  (Contributed by Mark Dickinson; :issue:`5260`.)

* The division algorithm for long integers has been made faster
  by tightening the inner loop, doing shifts instead of multiplications,
  and fixing an unnecessary extra iteration.
  Various benchmarks show speedups of between 50% and 150% for long
  integer divisions and modulo operations.
  (Contributed by Mark Dickinson; :issue:`5512`.)
  Bitwise operations are also significantly faster (initial patch by
  Gregory Smith; :issue:`1087418`).

* The implementation of ``%`` checks for the left-side operand being
  a Python string and special-cases it; this results in a 1-3%
  performance increase for applications that frequently use ``%``
  with strings, such as templating libraries.
  (Implemented by Collin Winter; :issue:`5176`.)

* List comprehensions with an ``if`` condition are compiled into
  faster bytecode.  (Patch by Antoine Pitrou, back-ported to 2.7
  by Jeffrey Yasskin; :issue:`4715`.)

* Converting an integer or long integer to a decimal string was made
  faster by special-casing base 10 instead of using a generalized
  conversion function that supports arbitrary bases.
  (Patch by Gawain Bolton; :issue:`6713`.)

* The :meth:`split`, :meth:`replace`, :meth:`rindex`,
  :meth:`rpartition`, and :meth:`rsplit` methods of string-like types
  (strings, Unicode strings, and :class:`bytearray` objects) now use a
  fast reverse-search algorithm instead of a character-by-character
  scan.  This is sometimes faster by a factor of 10.  (Added by
  Florent Xicluna; :issue:`7462` and :issue:`7622`.)

* The :mod:`pickle` and :mod:`cPickle` modules now automatically
  intern the strings used for attribute names, reducing memory usage
  of the objects resulting from unpickling.  (Contributed by Jake
  McGuire; :issue:`5084`.)

* The :mod:`cPickle` module now special-cases dictionaries,
  nearly halving the time required to pickle them.
  (Contributed by Collin Winter; :issue:`5670`.)

.. ======================================================================

New and Improved Modules
========================

As in every release, Python's standard library received a number of
enhancements and bug fixes.  Here's a partial list of the most notable
changes, sorted alphabetically by module name. Consult the
:file:`Misc/NEWS` file in the source tree for a more complete list of
changes, or look through the Subversion logs for all the details.

* The :mod:`bdb` module's base debugging class :class:`~bdb.Bdb`
  gained a feature for skipping modules.  The constructor
  now takes an iterable containing glob-style patterns such as
  ``django.*``; the debugger will not step into stack frames
  from a module that matches one of these patterns.
  (Contributed by Maru Newby after a suggestion by
  Senthil Kumaran; :issue:`5142`.)

* The :mod:`binascii` module now supports the buffer API, so it can be
  used with :class:`memoryview` instances and other similar buffer objects.
  (Backported from 3.x by Florent Xicluna; :issue:`7703`.)

* Updated module: the :mod:`bsddb` module has been updated from 4.7.2devel9
  to version 4.8.4 of
  `the pybsddb package <http://www.jcea.es/programacion/pybsddb.htm>`__.
  The new version features better Python 3.x compatibility, various bug fixes,
  and adds several new BerkeleyDB flags and methods.
  (Updated by Jesús Cea Avión; :issue:`8156`.  The pybsddb
  changelog can be browsed at http://hg.jcea.es/pybsddb/file/tip/ChangeLog.)

* The :mod:`bz2` module's :class:`~bz2.BZ2File` now supports the context
  management protocol, so you can write ``with bz2.BZ2File(...) as f: ...``.
  (Contributed by Hagen Fuerstenau; :issue:`3860`.)

* New class: the :class:`~collections.Counter` class in the :mod:`collections`
  module is useful for tallying data.  :class:`~collections.Counter` instances
  behave mostly like dictionaries but return zero for missing keys instead of
  raising a :exc:`KeyError`:

  .. doctest::
     :options: +NORMALIZE_WHITESPACE

     >>> from collections import Counter
     >>> c = Counter()
     >>> for letter in 'here is a sample of english text':
     ...   c[letter] += 1
     ...
     >>> c
     Counter({' ': 6, 'e': 5, 's': 3, 'a': 2, 'i': 2, 'h': 2,
     'l': 2, 't': 2, 'g': 1, 'f': 1, 'm': 1, 'o': 1, 'n': 1,
     'p': 1, 'r': 1, 'x': 1})
     >>> c['e']
     5
     >>> c['z']
     0

  There are three additional :class:`~collections.Counter` methods:
  :meth:`~collections.Counter.most_common` returns the N most common
  elements and their counts.  :meth:`~collections.Counter.elements`
  returns an iterator over the contained elements, repeating each
  element as many times as its count.
  :meth:`~collections.Counter.subtract` takes an iterable and
  subtracts one for each element instead of adding; if the argument is
  a dictionary or another :class:`Counter`, the counts are
  subtracted. ::

    >>> c.most_common(5)
    [(' ', 6), ('e', 5), ('s', 3), ('a', 2), ('i', 2)]
    >>> c.elements() ->
       'a', 'a', ' ', ' ', ' ', ' ', ' ', ' ',
       'e', 'e', 'e', 'e', 'e', 'g', 'f', 'i', 'i',
       'h', 'h', 'm', 'l', 'l', 'o', 'n', 'p', 's',
       's', 's', 'r', 't', 't', 'x'
    >>> c['e']
    5
    >>> c.subtract('very heavy on the letter e')
    >>> c['e']    # Count is now lower
    -1

  Contributed by Raymond Hettinger; :issue:`1696199`.

  .. revision 79660

  The new :class:`~collections.OrderedDict` class is described in the earlier
  section :ref:`pep-0372`.

  The :class:`~collections.namedtuple` class now has an optional *rename* parameter.
  If *rename* is true, field names that are invalid because they've
  been repeated or that aren't legal Python identifiers will be
  renamed to legal names that are derived from the field's
  position within the list of fields:

     >>> from collections import namedtuple
     >>> T = namedtuple('T', ['field1', '$illegal', 'for', 'field2'], rename=True)
     >>> T._fields
     ('field1', '_1', '_2', 'field2')

  (Added by Raymond Hettinger; :issue:`1818`.)

  The :class:`~collections.deque` data type now has a
  :meth:`~collections.deque.count` method that returns the number of
  contained elements equal to the supplied argument *x*, and a
  :meth:`~collections.deque.reverse` method that reverses the elements
  of the deque in-place.  :class:`deque` also exposes its maximum
  length as the read-only :attr:`~collections.deque.maxlen` attribute.
  (Both features added by Raymond Hettinger.)

* The :mod:`copy` module's :func:`~copy.deepcopy` function will now
  correctly copy bound instance methods.  (Implemented by
  Robert Collins; :issue:`1515`.)

* The :mod:`ctypes` module now always converts ``None`` to a C NULL
  pointer for arguments declared as pointers.  (Changed by Thomas
  Heller; :issue:`4606`.)  The underlying `libffi library
  <http://sourceware.org/libffi/>`__ has been updated to version
  3.0.9, containing various fixes for different platforms.  (Updated
  by Matthias Klose; :issue:`8142`.)

* New method: the :mod:`datetime` module's :class:`~datetime.timedelta` class
  gained a :meth:`~datetime.timedelta.total_seconds` method that returns the
  number of seconds in the duration.  (Contributed by Brian Quinlan; :issue:`5788`.)

* New method: the :class:`~decimal.Decimal` class gained a
  :meth:`~decimal.Decimal.from_float` class method that performs an exact
  conversion of a floating-point number to a :class:`~decimal.Decimal`.
  Note that this is an **exact** conversion that strives for the
  closest decimal approximation to the floating-point representation's value;
  the resulting decimal value will therefore still include the inaccuracy,
  if any.
  For example, ``Decimal.from_float(0.1)`` returns
  ``Decimal('0.1000000000000000055511151231257827021181583404541015625')``.
  (Implemented by Raymond Hettinger; :issue:`4796`.)

  Most of the methods of the :class:`~decimal.Context` class now accept integers
  as well as :class:`~decimal.Decimal` instances; the only exceptions are the
  :meth:`~decimal.Context.canonical` and :meth:`~decimal.Context.is_canonical`
  methods.  (Patch by Juan José Conti; :issue:`7633`.)

  The constructor for :class:`~decimal.Decimal` now accepts
  floating-point numbers (added by Raymond Hettinger; :issue:`8257`)
  and non-European Unicode characters such as Arabic-Indic digits
  (contributed by Mark Dickinson; :issue:`6595`).

  When using :class:`~decimal.Decimal` instances with a string's
  :meth:`~str.format` method, the default alignment was previously
  left-alignment.  This has been changed to right-alignment, which seems
  more sensible for numeric types.  (Changed by Mark Dickinson; :issue:`6857`.)

* The :mod:`difflib` module now produces output that is more
  compatible with modern :command:`diff`/:command:`patch` tools thanks
  to two changes: 1) the header giving the filename now uses a tab
  character instead of spaces as a separator, and 2) the date format
  used is now ISO-8601 style, ``2005-01-26 23:30:50``.  (Fixed by
  Anatoly Techtonik; :issue:`7585`.)

* The :class:`~fractions.Fraction` class now accepts a single float or
  :class:`~decimal.Decimal` instance, or two rational numbers, as
  arguments to its constructor.  (Implemented by Mark Dickinson;
  rationals added in :issue:`5812`, and float/decimal in
  :issue:`8294`.)

  An oversight was fixed, making the :class:`Fraction` match the other
  numeric types; ordering comparisons (``<``, ``<=``, ``>``, ``>=``) between
  fractions and complex numbers now raise a :exc:`TypeError`.

  .. revision 79455

* New class: a new :class:`~ftplib.FTP_TLS` class in
  the :mod:`ftplib` module provides secure FTP
  connections using TLS encapsulation of authentication as well as
  subsequent control and data transfers.
  (Contributed by Giampaolo Rodola', :issue:`2054`.)

  The :meth:`~ftplib.FTP.storbinary` method for binary uploads can now restart
  uploads thanks to an added *rest* parameter (patch by Pablo Mouzo;
  :issue:`6845`.)

* New class decorator: :func:`total_ordering` in the :mod:`functools`
  module takes a class that defines an :meth:`__eq__` method and one of
  :meth:`__lt__`, :meth:`__le__`, :meth:`__gt__`, or :meth:`__ge__`,
  and generates the missing comparison methods.  Since the
  :meth:`__cmp__` method is being deprecated in Python 3.x,
  this decorator makes it easier to define ordered classes.
  (Added by Raymond Hettinger; :issue:`5479`.)

  New function: :func:`cmp_to_key` will take an old-style comparison
  function that expects two arguments and return a new callable that
  can be used as the *key* parameter to functions such as
  :func:`sorted`, :func:`min` and :func:`max`, etc.  The primary
  intended use is to help with making code compatible with Python 3.x.
  (Added by Raymond Hettinger.)

* New function: the :mod:`gc` module's :func:`~gc.is_tracked` returns
  true if a given instance is tracked by the garbage collector, false
  otherwise. (Contributed by Antoine Pitrou; :issue:`4688`.)

* The :mod:`gzip` module's :class:`~gzip.GzipFile` now supports the context
  management protocol, so you can write ``with gzip.GzipFile(...) as f: ...``
  (contributed by Hagen Fuerstenau; :issue:`3860`), and it now implements
  the :class:`io.BufferedIOBase` ABC, so you can wrap it with
  :class:`io.BufferedReader` for faster processing
  (contributed by Nir Aides; :issue:`7471`).
  It's also now possible to override the modification time
  recorded in a gzipped file by providing an optional timestamp to
  the constructor.  (Contributed by Jacques Frechet; :issue:`4272`.)

  Files in gzip format can be padded with trailing zero bytes; the
  :mod:`gzip` module will now consume these trailing bytes.  (Fixed by
  Tadek Pietraszek and Brian Curtin; :issue:`2846`.)

* New attribute: the :mod:`hashlib` module now has an :attr:`~hashlib.hashlib.algorithms`
  attribute containing a tuple naming the supported algorithms.
  In Python 2.7, ``hashlib.algorithms`` contains
  ``('md5', 'sha1', 'sha224', 'sha256', 'sha384', 'sha512')``
  (Contributed by Carl Chenet; :issue:`7418`.)

* The default :class:`~httplib.HTTPResponse` class used by the :mod:`httplib` module now
  supports buffering, resulting in much faster reading of HTTP responses.
  (Contributed by Kristjan Valur Jonsson; :issue:`4879`.)

  The :class:`~httplib.HTTPConnection` and :class:`~httplib.HTTPSConnection` classes
  now support a *source_address* parameter, a ``(host, port)`` 2-tuple
  giving the source address that will be used for the connection.
  (Contributed by Eldon Ziegler; :issue:`3972`.)

* The :mod:`imaplib` module now supports IPv6 addresses.
  (Contributed by Derek Morr; :issue:`1655`.)

* New function: the :mod:`inspect` module's :func:`~inspect.getcallargs`
  takes a callable and its positional and keyword arguments,
  and figures out which of the callable's parameters will receive each argument,
  returning a dictionary mapping argument names to their values.  For example::

    >>> from inspect import getcallargs
    >>> def f(a, b=1, *pos, **named):
    ...     pass
    >>> getcallargs(f, 1, 2, 3)
    {'a': 1, 'named': {}, 'b': 2, 'pos': (3,)}
    >>> getcallargs(f, a=2, x=4)
    {'a': 2, 'named': {'x': 4}, 'b': 1, 'pos': ()}
    >>> getcallargs(f)
    Traceback (most recent call last):
    ...
    TypeError: f() takes at least 1 argument (0 given)

  Contributed by George Sakkis; :issue:`3135`.

* Updated module: The :mod:`io` library has been upgraded to the version shipped with
  Python 3.1.  For 3.1, the I/O library was entirely rewritten in C
  and is 2 to 20 times faster depending on the task being performed.  The
  original Python version was renamed to the :mod:`_pyio` module.

  One minor resulting change: the :class:`io.TextIOBase` class now
  has an :attr:`errors` attribute giving the error setting
  used for encoding and decoding errors (one of ``'strict'``, ``'replace'``,
  ``'ignore'``).

  The :class:`io.FileIO` class now raises an :exc:`OSError` when passed
  an invalid file descriptor.  (Implemented by Benjamin Peterson;
  :issue:`4991`.)  The :meth:`~io.IOBase.truncate` method now preserves the
  file position; previously it would change the file position to the
  end of the new file.  (Fixed by Pascal Chambon; :issue:`6939`.)

* New function: ``itertools.compress(data, selectors)`` takes two
  iterators.  Elements of *data* are returned if the corresponding
  value in *selectors* is true::

    itertools.compress('ABCDEF', [1,0,1,0,1,1]) =>
      A, C, E, F

  .. maybe here is better to use >>> list(itertools.compress(...)) instead

  New function: ``itertools.combinations_with_replacement(iter, r)``
  returns all the possible *r*-length combinations of elements from the
  iterable *iter*.  Unlike :func:`~itertools.combinations`, individual elements
  can be repeated in the generated combinations::

    itertools.combinations_with_replacement('abc', 2) =>
      ('a', 'a'), ('a', 'b'), ('a', 'c'),
      ('b', 'b'), ('b', 'c'), ('c', 'c')

  Note that elements are treated as unique depending on their position
  in the input, not their actual values.

  The :func:`itertools.count` function now has a *step* argument that
  allows incrementing by values other than 1.  :func:`~itertools.count` also
  now allows keyword arguments, and using non-integer values such as
  floats or :class:`~decimal.Decimal` instances.  (Implemented by Raymond
  Hettinger; :issue:`5032`.)

  :func:`itertools.combinations` and :func:`itertools.product` were
  previously raising :exc:`ValueError` for values of *r* larger than
  the input iterable.  This was deemed a specification error, so they
  now return an empty iterator.  (Fixed by Raymond Hettinger; :issue:`4816`.)

* Updated module: The :mod:`json` module was upgraded to version 2.0.9 of the
  simplejson package, which includes a C extension that makes
  encoding and decoding faster.
  (Contributed by Bob Ippolito; :issue:`4136`.)

  To support the new :class:`collections.OrderedDict` type, :func:`json.load`
  now has an optional *object_pairs_hook* parameter that will be called
  with any object literal that decodes to a list of pairs.
  (Contributed by Raymond Hettinger; :issue:`5381`.)

* New functions: the :mod:`math` module gained
  :func:`~math.erf` and :func:`~math.erfc` for the error function and the complementary error function,
  :func:`~math.expm1` which computes ``e**x - 1`` with more precision than
  using :func:`~math.exp` and subtracting 1,
  :func:`~math.gamma` for the Gamma function, and
  :func:`~math.lgamma` for the natural log of the Gamma function.
  (Contributed by Mark Dickinson and nirinA raseliarison; :issue:`3366`.)

* The :mod:`multiprocessing` module's :class:`Manager*` classes
  can now be passed a callable that will be called whenever
  a subprocess is started, along with a set of arguments that will be
  passed to the callable.
  (Contributed by lekma; :issue:`5585`.)

  The :class:`~multiprocessing.Pool` class, which controls a pool of worker processes,
  now has an optional *maxtasksperchild* parameter.  Worker processes
  will perform the specified number of tasks and then exit, causing the
  :class:`~multiprocessing.Pool` to start a new worker.  This is useful if tasks may leak
  memory or other resources, or if some tasks will cause the worker to
  become very large.
  (Contributed by Charles Cazabon; :issue:`6963`.)

* The :mod:`nntplib` module now supports IPv6 addresses.
  (Contributed by Derek Morr; :issue:`1664`.)

* New functions: the :mod:`os` module wraps the following POSIX system
  calls: :func:`~os.getresgid` and :func:`~os.getresuid`, which return the
  real, effective, and saved GIDs and UIDs;
  :func:`~os.setresgid` and :func:`~os.setresuid`, which set
  real, effective, and saved GIDs and UIDs to new values;
  :func:`~os.initgroups`.  (GID/UID functions
  contributed by Travis H.; :issue:`6508`.  Support for initgroups added
  by Jean-Paul Calderone; :issue:`7333`.)

  The :func:`os.fork` function now re-initializes the import lock in
  the child process; this fixes problems on Solaris when :func:`~os.fork`
  is called from a thread.  (Fixed by Zsolt Cserna; :issue:`7242`.)

* In the :mod:`os.path` module, the :func:`~os.path.normpath` and
  :func:`~os.path.abspath` functions now preserve Unicode; if their input path
  is a Unicode string, the return value is also a Unicode string.
  (:meth:`~os.path.normpath` fixed by Matt Giuca in :issue:`5827`;
  :meth:`~os.path.abspath` fixed by Ezio Melotti in :issue:`3426`.)

* The :mod:`pydoc` module now has help for the various symbols that Python
  uses.  You can now do ``help('<<')`` or ``help('@')``, for example.
  (Contributed by David Laban; :issue:`4739`.)

* The :mod:`re` module's :func:`~re.split`, :func:`~re.sub`, and :func:`~re.subn`
  now accept an optional *flags* argument, for consistency with the
  other functions in the module.  (Added by Gregory P. Smith.)

* New function: in the :mod:`shutil` module, :func:`~shutil.make_archive`
  takes a filename, archive type (zip or tar-format), and a directory
  path, and creates an archive containing the directory's contents.
  (Added by Tarek Ziadé.)

  :mod:`shutil`'s :func:`~shutil.copyfile` and :func:`~shutil.copytree`
  functions now raise a :exc:`~shutil.SpecialFileError` exception when
  asked to copy a named pipe.  Previously the code would treat
  named pipes like a regular file by opening them for reading, and
  this would block indefinitely.  (Fixed by Antoine Pitrou; :issue:`3002`.)

* New functions: in the :mod:`site` module, three new functions
  return various site- and user-specific paths.
  :func:`~site.getsitepackages` returns a list containing all
  global site-packages directories, and
  :func:`~site.getusersitepackages` returns the path of the user's
  site-packages directory.
  :func:`~site.getuserbase` returns the value of the :envvar:`USER_BASE`
  environment variable, giving the path to a directory that can be used
  to store data.
  (Contributed by Tarek Ziadé; :issue:`6693`.)

  The :mod:`site` module now reports exceptions occurring
  when the :mod:`sitecustomize` module is imported, and will no longer
  catch and swallow the :exc:`KeyboardInterrupt` exception.  (Fixed by
  Victor Stinner; :issue:`3137`.)

* The :mod:`socket` module's :class:`~ssl.SSL` objects now support the
  buffer API, which fixed a test suite failure (fix by Antoine Pitrou;
  :issue:`7133`).  :class:`SSL` objects also now automatically set
  OpenSSL's :cmacro:`SSL_MODE_AUTO_RETRY`, which will prevent an error
  code being returned from :meth:`recv` operations that trigger an SSL
  renegotiation (fix by Antoine Pitrou; :issue:`8222`).

  The version of OpenSSL being used is now available as the module
  attributes :attr:`OPENSSL_VERSION` (a string),
  :attr:`OPENSSL_VERSION_INFO` (a 5-tuple), and
  :attr:`OPENSSL_VERSION_NUMBER` (an integer).  (Added by Antoine
  Pitrou; :issue:`8321`.)

  The :func:`~socket.create_connection` function
  gained a *source_address* parameter, a ``(host, port)`` 2-tuple
  giving the source address that will be used for the connection.
  (Contributed by Eldon Ziegler; :issue:`3972`.)

  The :meth:`~socket.socket.recv_into` and :meth:`~socket.socket.recvfrom_into`
  methods will now write into objects that support the buffer API, most usefully
  the :class:`bytearray` and :class:`memoryview` objects.  (Implemented by
  Antoine Pitrou; :issue:`8104`.)

* The :mod:`SocketServer` module's :class:`~SocketServer.TCPServer` class now
  has a :attr:`~SocketServer.TCPServer.disable_nagle_algorithm` class attribute.
  The default value is False; if overridden to be True,
  new request connections will have the TCP_NODELAY option set to
  prevent buffering many small sends into a single TCP packet.
  (Contributed by Kristjan Valur Jonsson; :issue:`6192`.)

* Updated module: the :mod:`sqlite3` module has been updated to
  version 2.6.0 of the `pysqlite package <http://code.google.com/p/pysqlite/>`__. Version 2.6.0 includes a number of bugfixes, and adds
  the ability to load SQLite extensions from shared libraries.
  Call the ``enable_load_extension(True)`` method to enable extensions,
  and then call :meth:`~sqlite3.Connection.load_extension` to load a particular shared library.
  (Updated by Gerhard Häring.)

* The :mod:`struct` module will no longer silently ignore overflow
  errors when a value is too large for a particular integer format
  code (one of ``bBhHiIlLqQ``); it now always raises a
  :exc:`struct.error` exception.  (Changed by Mark Dickinson;
  :issue:`1523`.)  The :func:`~struct.pack` function will also
  attempt to use :meth:`__index__` to convert and pack non-integers
  before trying the :meth:`__int__` method or reporting an error.
  (Changed by Mark Dickinson; :issue:`8300`.)

* New function: the :mod:`subprocess` module's
  :func:`~subprocess.check_output` runs a command with a specified set of arguments
  and returns the command's output as a string when the command runs without
  error, or raises a :exc:`~subprocess.CalledProcessError` exception otherwise.

  ::

    >>> subprocess.check_output(['df', '-h', '.'])
    'Filesystem     Size   Used  Avail Capacity  Mounted on\n
    /dev/disk0s2    52G    49G   3.0G    94%    /\n'

    >>> subprocess.check_output(['df', '-h', '/bogus'])
      ...
    subprocess.CalledProcessError: Command '['df', '-h', '/bogus']' returned non-zero exit status 1

  (Contributed by Gregory P. Smith.)

  The :mod:`subprocess` module will now retry its internal system calls
  on receiving an :const:`EINTR` signal.  (Reported by several people; final
  patch by Gregory P. Smith in :issue:`1068268`.)

* New function: :func:`~symtable.is_declared_global` in the :mod:`symtable` module
  returns true for variables that are explicitly declared to be global,
  false for ones that are implicitly global.
  (Contributed by Jeremy Hylton.)

* The ``sys.version_info`` value is now a named tuple, with attributes
  named :attr:`major`, :attr:`minor`, :attr:`micro`,
  :attr:`releaselevel`, and :attr:`serial`.  (Contributed by Ross
  Light; :issue:`4285`.)

  :func:`sys.getwindowsversion` also returns a named tuple,
  with attributes named :attr:`major`, :attr:`minor`, :attr:`build`,
  :attr:`platform`, :attr:`service_pack`, :attr:`service_pack_major`,
  :attr:`service_pack_minor`, :attr:`suite_mask`, and
  :attr:`product_type`.  (Contributed by Brian Curtin; :issue:`7766`.)

* The :mod:`tarfile` module's default error handling has changed, to
  no longer suppress fatal errors.  The default error level was previously 0,
  which meant that errors would only result in a message being written to the
  debug log, but because the debug log is not activated by default,
  these errors go unnoticed.  The default error level is now 1,
  which raises an exception if there's an error.
  (Changed by Lars Gustäbel; :issue:`7357`.)

  :mod:`tarfile` now supports filtering the :class:`~tarfile.TarInfo`
  objects being added to a tar file.  When you call :meth:`~tarfile.TarFile.add`,
  instance, you may supply an optional *filter* argument
  that's a callable.  The *filter* callable will be passed the
  :class:`~tarfile.TarInfo` for every file being added, and can modify and return it.
  If the callable returns ``None``, the file will be excluded from the
  resulting archive.  This is more powerful than the existing
  *exclude* argument, which has therefore been deprecated.
  (Added by Lars Gustäbel; :issue:`6856`.)
  The :class:`~tarfile.TarFile` class also now supports the context manager protocol.
  (Added by Lars Gustäbel; :issue:`7232`.)

* The :meth:`~threading.Event.wait` method of the :class:`threading.Event` class
  now returns the internal flag on exit.  This means the method will usually
  return true because :meth:`~threading.Event.wait` is supposed to block until the
  internal flag becomes true.  The return value will only be false if
  a timeout was provided and the operation timed out.
  (Contributed by Tim Lesher; :issue:`1674032`.)

* The Unicode database provided by the :mod:`unicodedata` module is
  now used internally to determine which characters are numeric,
  whitespace, or represent line breaks.  The database also
  includes information from the :file:`Unihan.txt` data file (patch
  by Anders Chrigström and Amaury Forgeot d'Arc; :issue:`1571184`)
  and has been updated to version 5.2.0 (updated by
  Florent Xicluna; :issue:`8024`).

* The :class:`~UserDict.UserDict` class is now a new-style class.  (Changed by
  Benjamin Peterson.)

* The ElementTree library, :mod:`xml.etree`, no longer escapes
  ampersands and angle brackets when outputting an XML processing
  instruction (which looks like ``<?xml-stylesheet href="#style1"?>``)
  or comment (which looks like ``<!-- comment -->``).
  (Patch by Neil Muller; :issue:`2746`.)

* The :mod:`zipfile` module's :class:`~zipfile.ZipFile` now supports the context
  management protocol, so you can write ``with zipfile.ZipFile(...) as f: ...``.
  (Contributed by Brian Curtin; :issue:`5511`.)

  :mod:`zipfile` now supports archiving empty directories and
  extracts them correctly.  (Fixed by Kuba Wieczorek; :issue:`4710`.)
  Reading files out of an archive is now faster, and interleaving
  :meth:`~zipfile.ZipFile.read` and :meth:`~zipfile.ZipFile.readline` now works correctly.
  (Contributed by Nir Aides; :issue:`7610`.)

  The :func:`~zipfile.is_zipfile` function now
  accepts a file object, in addition to the path names accepted in earlier
  versions.  (Contributed by Gabriel Genellina; :issue:`4756`.)

  The :meth:`~zipfile.ZipFile.writestr` method now has an optional *compress_type* parameter
  that lets you override the default compression method specified in the
  :class:`~zipfile.ZipFile` constructor.  (Contributed by Ronald Oussoren;
  :issue:`6003`.)


New module: sysconfig
---------------------------------

XXX A new :mod:`sysconfig` module has been extracted from
:mod:`distutils` and put in the standard library.

The :mod:`sysconfig` module provides access to Python's configuration
information like the list of installation paths and the configuration
variables relevant for the current platform. (contributed by Tarek)

Updated module: ElementTree 1.3
---------------------------------

XXX write this.

.. ======================================================================
.. whole new modules get described in subsections here


Unit Testing Enhancements
---------------------------------

The :mod:`unittest` module was enhanced in several ways.
The progress messages now shows 'x' for expected failures
and 'u' for unexpected successes when run in verbose mode.
(Contributed by Benjamin Peterson.)
Test cases can raise the :exc:`~unittest.SkipTest` exception to skip a test.
(:issue:`1034053`.)

.. XXX describe test discovery (Contributed by Michael Foord; :issue:`6001`.)

The error messages for :meth:`~unittest.TestCase.assertEqual`,
:meth:`~unittest.TestCase.assertTrue`, and :meth:`~unittest.TestCase.assertFalse`
failures now provide more information.  If you set the
:attr:`~unittest.TestCase.longMessage` attribute of your :class:`~unittest.TestCase` classes to
True, both the standard error message and any additional message you
provide will be printed for failures.  (Added by Michael Foord; :issue:`5663`.)

The :meth:`~unittest.TestCase.assertRaises` method now
return a context handler when called without providing a callable
object to run.  For example, you can write this::

  with self.assertRaises(KeyError):
      {}['foo']

(Implemented by Antoine Pitrou; :issue:`4444`.)

.. rev 78774

Module- and class-level setup and teardown fixtures are now supported.
Modules can contain :func:`~unittest.setUpModule` and :func:`~unittest.tearDownModule`
functions.  Classes can have :meth:`~unittest.TestCase.setUpClass` and
:meth:`~unittest.TestCase.tearDownClass` methods that must be defined as class methods
(using ``@classmethod`` or equivalent).  These functions and
methods are invoked when the test runner switches to a test case in a
different module or class.

The methods :meth:`~unittest.TestCase.addCleanup` and
:meth:`~unittest.TestCase.doCleanups` were added.
:meth:`~unittest.TestCase.addCleanup` allows you to add cleanup functions that
will be called unconditionally (after :meth:`~unittest.TestCase.setUp` if
:meth:`~unittest.TestCase.setUp` fails, otherwise after :meth:`~unittest.TestCase.tearDown`). This allows
for much simpler resource allocation and deallocation during tests
(:issue:`5679`).

A number of new methods were added that provide more specialized
tests.  Many of these methods were written by Google engineers
for use in their test suites; Gregory P. Smith, Michael Foord, and
GvR worked on merging them into Python's version of :mod:`unittest`.

* :meth:`~unittest.TestCase.assertIsNone` and :meth:`~unittest.TestCase.assertIsNotNone` take one
  expression and verify that the result is or is not ``None``.

* :meth:`~unittest.TestCase.assertIs` and :meth:`~unittest.TestCase.assertIsNot`
  take two values and check whether the two values evaluate to the same object or not.
  (Added by Michael Foord; :issue:`2578`.)

* :meth:`~unittest.TestCase.assertIsInstance` and
  :meth:`~unittest.TestCase.assertNotIsInstance` check whether
  the resulting object is an instance of a particular class, or of
  one of a tuple of classes.  (Added by Georg Brandl; :issue:`7031`.)

* :meth:`~unittest.TestCase.assertGreater`, :meth:`~unittest.TestCase.assertGreaterEqual`,
  :meth:`~unittest.TestCase.assertLess`, and :meth:`~unittest.TestCase.assertLessEqual` compare
  two quantities.

* :meth:`~unittest.TestCase.assertMultiLineEqual` compares two strings, and if they're
  not equal, displays a helpful comparison that highlights the
  differences in the two strings.  This comparison is now used by
  default when Unicode strings are compared with :meth:`~unittest.TestCase.assertEqual`.

* :meth:`~unittest.TestCase.assertRegexpMatches` checks whether its first argument is a
  string matching a regular expression provided as its second argument.

  .. XXX add assertNotRegexpMatches see issue 8038

* :meth:`~unittest.TestCase.assertRaisesRegexp` checks whether a particular exception
  is raised, and then also checks that the string representation of
  the exception matches the provided regular expression.

* :meth:`~unittest.TestCase.assertIn` and :meth:`~unittest.TestCase.assertNotIn`
  tests whether *first* is or is not in  *second*.

* :meth:`~unittest.TestCase.assertItemsEqual` tests whether two provided sequences
  contain the same elements.

* :meth:`~unittest.TestCase.assertSetEqual` compares whether two sets are equal, and
  only reports the differences between the sets in case of error.

* Similarly, :meth:`~unittest.TestCase.assertListEqual` and :meth:`~unittest.TestCase.assertTupleEqual`
  compare the specified types and explain any differences without necessarily
  printing their full values; these methods are now used by default
  when comparing lists and tuples using :meth:`~unittest.TestCase.assertEqual`.
  More generally, :meth:`~unittest.TestCase.assertSequenceEqual` compares two sequences
  and can optionally check whether both sequences are of a
  particular type.

* :meth:`~unittest.TestCase.assertDictEqual` compares two dictionaries and reports the
  differences; it's now used by default when you compare two dictionaries
  using :meth:`~unittest.TestCase.assertEqual`.  :meth:`~unittest.TestCase.assertDictContainsSubset` checks whether
  all of the key/value pairs in *first* are found in *second*.

* :meth:`~unittest.TestCase.assertAlmostEqual` and :meth:`~unittest.TestCase.assertNotAlmostEqual` test
  whether *first* and *second* are approximately equal by computing
  their difference, rounding the result to an optionally-specified number
  of *places* (the default is 7), and comparing to zero.

* :meth:`~unittest.TestLoader.loadTestsFromName` properly honors the
  :attr:`~unittest.TestLoader.suiteClass` attribute of
  the :class:`~unittest.TestLoader`. (Fixed by Mark Roddy; :issue:`6866`.)

* A new hook lets you extend the :meth:`~unittest.TestCase.assertEqual` method to handle
  new data types.  The :meth:`~unittest.TestCase.addTypeEqualityFunc` method takes a type
  object and a function. The function will be used when both of the
  objects being compared are of the specified type.  This function
  should compare the two objects and raise an exception if they don't
  match; it's a good idea for the function to provide additional
  information about why the two objects are matching, much as the new
  sequence comparison methods do.

:func:`unittest.main` now takes an optional ``exit`` argument.  If
False, :func:`~unittest.main` doesn't call :func:`sys.exit`, allowing it to be
used from the interactive interpreter. (Contributed by J. Pablo
Fernández; :issue:`3379`.)

A new command-line switch, :option:`-f` or :option:`--failfast`, makes
test execution stop immediately when a test fails instead of
continuing to execute further tests.  (Suggested by Cliff Dyer and
implemented by Michael Foord; :issue:`8074`.)

.. XXX document the other new switches

:class:`~unittest.TestResult` has new :meth:`~unittest.TestResult.startTestRun` and
:meth:`~unittest.TestResult.stopTestRun` methods that are called immediately before
and after a test run.  (Contributed by Robert Collins; :issue:`5728`.)

With all these changes, the :file:`unittest.py` was becoming awkwardly
large, so the module was turned into a package and the code split into
several files (by Benjamin Peterson).  This doesn't affect how the
module is imported.


.. _importlib-section:

importlib: Importing Modules
------------------------------

Python 3.1 includes the :mod:`importlib` package, a re-implementation
of the logic underlying Python's :keyword:`import` statement.
:mod:`importlib` is useful for implementors of Python interpreters and
to users who wish to write new importers that can participate in the
import process.  Python 2.7 doesn't contain the complete
:mod:`importlib` package, but instead has a tiny subset that contains
a single function, :func:`~importlib.import_module`.

``import_module(name, package=None)`` imports a module.  *name* is
a string containing the module or package's name.  It's possible to do
relative imports by providing a string that begins with a ``.``
character, such as ``..utils.errors``.  For relative imports, the
*package* argument must be provided and is the name of the package that
will be used as the anchor for
the relative import.  :func:`~importlib.import_module` both inserts the imported
module into ``sys.modules`` and returns the module object.

Here are some examples::

    >>> from importlib import import_module
    >>> anydbm = import_module('anydbm')  # Standard absolute import
    >>> anydbm
    <module 'anydbm' from '/p/python/Lib/anydbm.py'>
    >>> # Relative import
    >>> sysconfig = import_module('..sysconfig', 'distutils.command')
    >>> sysconfig
    <module 'distutils.sysconfig' from '/p/python/Lib/distutils/sysconfig.pyc'>

:mod:`importlib` was implemented by Brett Cannon and introduced in
Python 3.1.


ttk: Themed Widgets for Tk
--------------------------

Tcl/Tk 8.5 includes a set of themed widgets that re-implement basic Tk
widgets but have a more customizable appearance and can therefore more
closely resemble the native platform's widgets.  This widget
set was originally called Tile, but was renamed to Ttk (for "themed Tk")
on being added to Tcl/Tck release 8.5.

XXX write a brief discussion and an example here.

The :mod:`ttk` module was written by Guilherme Polo and added in
:issue:`2983`.  An alternate version called ``Tile.py``, written by
Martin Franklin and maintained by Kevin Walzer, was proposed for
inclusion in :issue:`2618`, but the authors argued that Guilherme
Polo's work was more comprehensive.


Deprecations and Removals
=========================

* :func:`contextlib.nested`, which allows handling more than one context manager
  with one :keyword:`with` statement, has been deprecated; :keyword:`with`
  supports multiple context managers syntactically now.

.. ======================================================================


Build and C API Changes
=======================

Changes to Python's build process and to the C API include:

* The latest release of the GNU Debugger, GDB 7, can be `scripted
  using Python
  <http://sourceware.org/gdb/current/onlinedocs/gdb/Python.html>`__.
  When you begin debugging an executable program P, GDB will look for
  a file named ``P-gdb.py`` and automatically read it.  Dave Malcolm
  contributed a :file:`python-gdb.py` that adds a number of useful
  commands when debugging Python itself.  For example, there are
  ``py-up`` and ``py-down`` that go up or down one Python stack frame,
  which usually corresponds to several C stack frames.  ``py-print``
  prints the value of a Python variable, and ``py-bt`` prints the
  Python stack trace.  (Added as a result of :issue:`8032`.)

* If you use the :file:`.gdbinit` file provided with Python,
  the "pyo" macro in the 2.7 version now works correctly when the thread being
  debugged doesn't hold the GIL; the macro now acquires it before printing.
  (Contributed by Victor Stinner; :issue:`3632`.)

* :cfunc:`Py_AddPendingCall` is now thread-safe, letting any
  worker thread submit notifications to the main Python thread.  This
  is particularly useful for asynchronous IO operations.
  (Contributed by Kristjan Valur Jonsson; :issue:`4293`.)

* New function: :cfunc:`PyCode_NewEmpty` creates an empty code object;
  only the filename, function name, and first line number are required.
  This is useful to extension modules that are attempting to
  construct a more useful traceback stack.  Previously such
  extensions needed to call :cfunc:`PyCode_New`, which had many
  more arguments.  (Added by Jeffrey Yasskin.)

* New function: :cfunc:`PyErr_NewExceptionWithDoc` creates a new
  exception class, just as the existing :cfunc:`PyErr_NewException` does,
  but takes an extra ``char *`` argument containing the docstring for the
  new exception class.  (Added by the 'lekma' user on the Python bug tracker;
  :issue:`7033`.)

* New function: :cfunc:`PyFrame_GetLineNumber` takes a frame object
  and returns the line number that the frame is currently executing.
  Previously code would need to get the index of the bytecode
  instruction currently executing, and then look up the line number
  corresponding to that address.  (Added by Jeffrey Yasskin.)

* New functions: :cfunc:`PyLong_AsLongAndOverflow` and
  :cfunc:`PyLong_AsLongLongAndOverflow`  approximates a Python long
  integer as a C :ctype:`long` or :ctype:`long long`.
  If the number is too large to fit into
  the output type, an *overflow* flag is set and returned to the caller.
  (Contributed by Case Van Horsen; :issue:`7528` and :issue:`7767`.)

* New function: stemming from the rewrite of string-to-float conversion,
  a new :cfunc:`PyOS_string_to_double` function was added.  The old
  :cfunc:`PyOS_ascii_strtod` and :cfunc:`PyOS_ascii_atof` functions
  are now deprecated.

* New macros: the Python header files now define the following macros:
  :cmacro:`Py_ISALNUM`,
  :cmacro:`Py_ISALPHA`,
  :cmacro:`Py_ISDIGIT`,
  :cmacro:`Py_ISLOWER`,
  :cmacro:`Py_ISSPACE`,
  :cmacro:`Py_ISUPPER`,
  :cmacro:`Py_ISXDIGIT`,
  and :cmacro:`Py_TOLOWER`, :cmacro:`Py_TOUPPER`.
  All of these functions are analogous to the C
  standard macros for classifying characters, but ignore the current
  locale setting, because in
  several places Python needs to analyze characters in a
  locale-independent way.  (Added by Eric Smith;
  :issue:`5793`.)

  .. XXX these macros don't seem to be described in the c-api docs.

* New format codes: the :cfunc:`PyFormat_FromString`,
  :cfunc:`PyFormat_FromStringV`, and :cfunc:`PyErr_Format` now
  accepts ``%lld`` and ``%llu`` format codes for displaying values of
  C's :ctype:`long long` types.
  (Contributed by Mark Dickinson; :issue:`7228`.)

* The complicated interaction between threads and process forking has
  been changed.  Previously, the child process created by
  :func:`os.fork` might fail because the child is created with only a
  single thread running, the thread performing the :func:`os.fork`.
  If other threads were holding a lock, such as Python's import lock,
  when the fork was performed, the lock would still be marked as
  "held" in the new process.  But in the child process nothing would
  ever release the lock, since the other threads weren't replicated,
  and the child process would no longer be able to perform imports.

  Python 2.7 now acquires the import lock before performing an
  :func:`os.fork`, and will also clean up any locks created using the
  :mod:`threading` module.  C extension modules that have internal
  locks, or that call :cfunc:`fork()` themselves, will not benefit
  from this clean-up.

  (Fixed by Thomas Wouters; :issue:`1590864`.)

* The :cfunc:`Py_Finalize` function now calls the internal
  :func:`threading._shutdown` function; this prevents some exceptions from
  being raised when an interpreter shuts down.
  (Patch by Adam Olsen; :issue:`1722344`.)

* Global symbols defined by the :mod:`ctypes` module are now prefixed
  with ``Py``, or with ``_ctypes``.  (Implemented by Thomas
  Heller; :issue:`3102`.)

* New configure option: the :option:`--with-system-expat` switch allows
  building the :mod:`pyexpat` module to use the system Expat library.
  (Contributed by Arfrever Frehtes Taifersar Arahesis; :issue:`7609`.)

* New configure option: compiling Python with the
  :option:`--with-valgrind` option will now disable the pymalloc
  allocator, which is difficult for the Valgrind memory-error detector
  to analyze correctly.
  Valgrind will therefore be better at detecting memory leaks and
  overruns. (Contributed by James Henstridge; :issue:`2422`.)

* New configure option: you can now supply no arguments to
  :option:`--with-dbmliborder=` in order to build none of the various
  DBM modules.  (Added by Arfrever Frehtes Taifersar Arahesis;
  :issue:`6491`.)

* The :program:`configure` script now checks for floating-point rounding bugs
  on certain 32-bit Intel chips and defines a :cmacro:`X87_DOUBLE_ROUNDING`
  preprocessor definition.  No code currently uses this definition,
  but it's available if anyone wishes to use it.
  (Added by Mark Dickinson; :issue:`2937`.)

  :program:`configure` also now sets a :envvar:`LDCXXSHARED` Makefile
  variable for supporting C++ linking.  (Contributed by Arfrever
  Frehtes Taifersar Arahesis; :issue:`1222585`.)

* The build process now creates the necessary files for pkg-config
  support.  (Contributed by Clinton Roy; :issue:`3585`.)

* The build process now supports Subversion 1.7.  (Contributed by
  Arfrever Frehtes Taifersar Arahesis; :issue:`6094`.)


.. ======================================================================

Port-Specific Changes: Windows
-----------------------------------

* The :mod:`msvcrt` module now contains some constants from
  the :file:`crtassem.h` header file:
  :data:`CRT_ASSEMBLY_VERSION`,
  :data:`VC_ASSEMBLY_PUBLICKEYTOKEN`,
  and :data:`LIBRARIES_ASSEMBLY_NAME_PREFIX`.
  (Contributed by David Cournapeau; :issue:`4365`.)

* The new :cfunc:`_beginthreadex` API is used to start threads, and
  the native thread-local storage functions are now used.
  (Contributed by Kristjan Valur Jonsson; :issue:`3582`.)

* The :func:`os.kill` function now works on Windows.  The signal value
  can be the constants :const:`CTRL_C_EVENT`,
  :const:`CTRL_BREAK_EVENT`, or any integer.  The Control-C and
  Control-Break keystroke events can be sent to subprocesses; any
  other value will use the :cfunc:`TerminateProcess` API.
  (Contributed by Miki Tebeka; :issue:`1220212`.)

* The :func:`os.listdir` function now correctly fails
  for an empty path.  (Fixed by Hirokazu Yamamoto; :issue:`5913`.)

* The :mod:`mimelib` module will now read the MIME database from
  the Windows registry when initializing.
  (Patch by Gabriel Genellina; :issue:`4969`.)

.. ======================================================================

Port-Specific Changes: Mac OS X
-----------------------------------

* The path ``/Library/Python/2.7/site-packages`` is now appended to
  ``sys.path``, in order to share added packages between the system
  installation and a user-installed copy of the same version.
  (Changed by Ronald Oussoren; :issue:`4865`.)


Other Changes and Fixes
=======================

* Two benchmark scripts, :file:`iobench` and :file:`ccbench`, were
  added to the :file:`Tools` directory.  :file:`iobench` measures the
  speed of built-in file I/O objects (as returned by :func:`open`)
  while performing various operations, and :file:`ccbench` is a
  concurrency benchmark that tries to measure computing throughput,
  thread switching latency, and IO processing bandwidth when
  performing several tasks using a varying number of threads.

* When importing a module from a :file:`.pyc` or :file:`.pyo` file
  with an existing :file:`.py` counterpart, the :attr:`co_filename`
  attributes of the resulting code objects are overwritten when the
  original filename is obsolete.  This can happen if the file has been
  renamed, moved, or is accessed through different paths.  (Patch by
  Ziga Seilnacht and Jean-Paul Calderone; :issue:`1180193`.)

* The :file:`regrtest.py` script now takes a :option:`--randseed=`
  switch that takes an integer that will be used as the random seed
  for the :option:`-r` option that executes tests in random order.
  The :option:`-r` option also reports the seed that was used
  (Added by Collin Winter.)

* Another :file:`regrtest.py` switch is :option:`-j`, which
  takes an integer specifying how many tests run in parallel. This
  allows reducing the total runtime on multi-core machines.
  This option is compatible with several other options, including the
  :option:`-R` switch which is known to produce long runtimes.
  (Added by Antoine Pitrou, :issue:`6152`.)  This can also be used
  with a new :option:`-F` switch that runs selected tests in a loop
  until they fail.  (Added by Antoine Pitrou; :issue:`7312`.)

* When executed as a script, the :file:`py_compile.py` module now
  accepts ``'-'`` as an argument, which will read standard input for
  the list of filenames to be compiled.  (Contributed by Piotr
  Ożarowski; :issue:`8233`.)

.. ======================================================================

Porting to Python 2.7
=====================

This section lists previously described changes and other bugfixes
that may require changes to your code:

* When using :class:`Decimal` instances with a string's
  :meth:`format` method, the default alignment was previously
  left-alignment.  This has been changed to right-alignment, which might
  change the output of your programs.
  (Changed by Mark Dickinson; :issue:`6857`.)

  Another :meth:`format`-related change: the default precision used
  for floating-point and complex numbers was changed from 6 decimal
  places to 12, which matches the precision used by :func:`str`.
  (Changed by Eric Smith; :issue:`5920`.)

* Because of an optimization for the :keyword:`with` statement, the special
  methods :meth:`__enter__` and :meth:`__exit__` must belong to the object's
  type, and cannot be directly attached to the object's instance.  This
  affects new-style classes (derived from :class:`object`) and C extension
  types.  (:issue:`6101`.)

* The :meth:`readline` method of :class:`StringIO` objects now does
  nothing when a negative length is requested, as other file-like
  objects do.  (:issue:`7348`).

In the standard library:

* The ElementTree library, :mod:`xml.etree`, no longer escapes
  ampersands and angle brackets when outputting an XML processing
  instruction (which looks like `<?xml-stylesheet href="#style1"?>`)
  or comment (which looks like `<!-- comment -->`).
  (Patch by Neil Muller; :issue:`2746`.)

For C extensions:

* C extensions that use integer format codes with the ``PyArg_Parse*``
  family of functions will now raise a :exc:`TypeError` exception
  instead of triggering a :exc:`DeprecationWarning` (:issue:`5080`).

* Use the new :cfunc:`PyOS_string_to_double` function instead of the old
  :cfunc:`PyOS_ascii_strtod` and :cfunc:`PyOS_ascii_atof` functions,
  which are now deprecated.


.. ======================================================================


.. _acks27:

Acknowledgements
================

The author would like to thank the following people for offering
suggestions, corrections and assistance with various drafts of this
article: Ryan Lovett, R. David Murray, Hugh Secker-Walker.