Misc/NEWS

+++++++++++
Python News
+++++++++++

What's New in Python 3.2 Alpha 3?
=================================

*Release date: XX-Oct-2010*

Core and Builtins
-----------------

- Issue #9901: Destroying the GIL in Py_Finalize() can fail if some other
  threads are still running.  Instead, reinitialize the GIL on a second
  call to Py_Initialize().

- Issue #9252: PyImport_Import no longer uses a fromlist hack to return the
  module that was imported, but instead gets the module from sys.modules.

- Issue #9212: The range type_items now provides index() and count()
  methods, to conform to the Sequence ABC.  Patch by Daniel Urban and
  Daniel Stutzbach.

- Issue #7994: Issue a PendingDeprecationWarning if object.__format__
  is called with a non-empty format string. This is an effort to
  future-proof user code. If a derived class does not currently
  implement __format__ but later adds its own __format__, it would
  most likely break user code that had supplied a format string. This
  will be changed to a DeprecationWaring in Python 3.3 and it will be
  an error in Python 3.4.

- Issue #9828: Destroy the GIL in Py_Finalize(), so that it gets properly
  re-created on a subsequent call to Py_Initialize().  The problem (a crash)
  wouldn't appear in 3.1 or 2.7 where the GIL's structure is more trivial.

- Issue #9210: Configure option --with-wctype-functions was removed.  Using the
  functions from the libc caused the methods .upper() and lower() to become
  locale aware and created subtly wrong results.

- Issue #9738: PyUnicode_FromFormat() and PyErr_Format() raise an error on
  a non-ASCII byte in the format string.

- Issue #4617: Previously it was illegal to delete a name from the local
  namespace if it occurs as a free variable in a nested block.  This limitation
  of the compiler has been lifted, and a new opcode introduced (DELETE_DEREF).

- Issue #9804: ascii() now always represents unicode surrogate pairs as
  a single ``\UXXXXXXXX``, regardless of whether the character is printable
  or not.  Also, the "backslashreplace" error handler now joins surrogate
  pairs into a single character on UCS-2 builds.

- Issue #9757: memoryview objects get a release() method to release the
  underlying buffer (previously this was only done when deallocating the
  memoryview), and gain support for the context management protocol.

- Issue #9797: pystate.c wrongly assumed that zero couldn't be a valid
  thread-local storage key.

Library
-------

- Issue #9877: Expose sysconfig.get_makefile_filename()

- logging: Added hasHandlers() method to Logger and LoggerAdapter.

- Issue #1686: Fix string.Template when overriding the pattern attribute.

- Issue #9854: SocketIO objects now observe the RawIOBase interface in
  non-blocking mode: they return None when an operation would block (instead
  of raising an exception).

- Issue #1730136: Fix the comparison between a tk.font.Font and an object of
  another kind.

- Issue #9441: logging has better coverage for rotating file handlers.

- Issue #9865:  collections.OrderedDict now has a __sizeof__ method.

- Issue #9854: The default read() implementation in io.RawIOBase now
  handles non-blocking readinto() returning None correctly.

- Issue #1552: socket.socketpair() now returns regular socket.socket
  objects supporting the whole socket API (rather than the "raw"
  _socket.socket objects).

- Issue #9853: Fix the signature of SSLSocket.recvfrom() and
  SSLSocket.sendto() to match the corresponding socket methods.

- Issue 9840: Added a decorator to reprlib for wrapping __repr__ methods
  to make them handle recursive calls within the same thread.

- logging: Enhanced HTTPHandler with secure and credentials initializers.

- Issue #767645: Set os.path.supports_unicode_filenames to True on Mac OS X.

- Issue #9837: The read() method of ZipExtFile objects (as returned by
  ZipFile.open()) could return more bytes than requested.

- Issue #9826: OrderedDict.__repr__ can now handle self-referential
  values:   d['x'] = d.

- Issue #9825: Using __del__ in the definition of collections.OrderedDict made
  it possible for the user to create self-referencing ordered dictionaries
  which become permanently uncollectable GC garbage. Reinstated the Py3.1
  approach of using weakref proxies so that reference cycles never get created
  in the first place.

- Issue #9579, #9580: Fix os.confstr() for value longer than 255 bytes and
  encode the value with filesystem encoding and surrogateescape (instead of
  utf-8 in strict mode). Patch written by David Watson.

- Issue #9632: Remove sys.setfilesystemencoding() function: use
  PYTHONFSENCODING environment variable to set the filesystem encoding at
  Python startup. sys.setfilesystemencoding() creates inconsistencies because
  it is unable to reencode all filenames in all objects.

- Issue #9410: Various optimizations to the pickle module, leading to
  speedups up to 4x (depending on the benchmark).  Mostly ported from
  Unladen Swallow; initial patch by Alexandre Vassalotti.

- The pprint module now supports printing OrderedDicts in their given
  order (formerly, it would sort the keys).

- Logging: Added QueueHandler class to facilitate logging usage with
  multiprocessing.

- Issue #9707: Rewritten reference implementation of threading.local which
  is friendlier towards reference cycles.  This change is not normally
  visible since an optimized C implementation (_thread._local) is used
  instead.

- Issue #6394: os.getppid() is now supported on Windows.  Note that it will
  still return the id of the parent process after it has exited.  This process
  id may even have been reused by another unrelated process.

- Issue #9792: In case of connection failure, socket.create_connection()
  would swallow the exception and raise a new one, making it impossible
  to fetch the original errno, or to filter timeout errors.  Now the
  original error is re-raised.

- Issue #9758: When fcntl.ioctl() was called with mutable_flag set to True,
  and the passed buffer was exactly 1024 bytes long, the buffer wouldn't
  be updated back after the system call.  Original patch by Brian Brazil.

- Updates to the random module:

  * Document which parts of the module are guaranteed to stay the same
    across versions and which parts are subject to change.

  * Update the seed() method to use all of the bits in a string
    instead of just the hash value.  This makes better use of the
    seed value and assures the seeding is platform independent.
    Issue #7889.

  * Improved the random()-->integer algorithm used in choice(),
    shuffle(), sample(), randrange(), and randint().  Formerly, it
    used int(n*random()) which has a slight bias whenever n is not
    a power of two.  Issue #9025.

  * Improved documentation of arguments to randrange().  Issue #9379.

- collections.OrderedDict now supports a new method for repositioning
  keys to either end.

- Issue #9754: Similarly to assertRaises and assertRaisesRegexp, unittest
  test cases now also have assertWarns and assertWarnsRegexp methods to
  check that a given warning type was triggered by the code under test.

- Issue #5506: BytesIO objects now have a getbuffer() method exporting a
  view of their contents without duplicating them.  The view is both readable
  and writable.

- Issue #7566: Implement os.path.sameopenfile for Windows.

- Issue #9293: I/O streams now raise ``io.UnsupportedOperation`` when an
  unsupported operation is attempted (for example, writing to a file open
  only for reading).

- hashlib has two new constant attributes: algorithms_guaranteed and
  algorithms_avaiable that respectively list the names of hash algorithms
  guaranteed to exist in all Python implementations and the names of hash
  algorithms available in the current process.

C-API
-----

- Issue #9834: Don't segfault in PySequence_GetSlice, PySequence_SetSlice, or
  PySequence_DelSlice when the object doesn't have any mapping operations
  defined.

Tools/Demos
-----------

- Issue #9188: The gdb extension now handles correctly narrow (UCS2) as well
  as wide (UCS4) unicode builds for both the host interpreter (embedded
  inside gdb) and the interpreter under test.

Tests
-----

- Issue #9894: Do not hardcode ENOENT in test_subprocess.

- Issue #9315: Added tests for the trace module.  Patch by Eli Bendersky.

- Issue #9323: Make test.regrtest.__file__ absolute, this was not always the
  case when running profile or trace, for example.

Build
-----

- Issue #9552: Avoid unnecessary rebuild of OpenSSL. (Windows)

- Issue #9810: Compile bzip2 source files in python's project file
  directly. It used to be built with bzip2's makefile.

- Issue #9848: Stopping trying to build _weakref in setup.py as it is a
  built-in module.

- Issue #9806: python-config now has an ``--extension-suffix`` option that
  outputs the suffix for dynamic libraries including the ABI version name
  defined by PEP 3149.

- Issue #941346: Improve the build process under AIX and allow Python to
  be built as a shared library.  Patch by Sébastien Sablé.

- Issue #4026: Make the fcntl extension build under AIX.  Patch by Sébastien
  Sablé.


What's New in Python 3.2 Alpha 2?
=================================

*Release date: 05-Sep-2010*

Core and Builtins
-----------------

- Issue #9225: Remove the ROT_FOUR and DUP_TOPX opcode, the latter replaced by
  the new (and simpler) DUP_TOP_TWO.  Performance isn't changed, but our
  bytecode is a bit simplified.  Patch by Demur Rumed.

- Issue #9766: Rename poorly named variables exposed by _warnings to prevent
  confusion with the proper variables names from 'warnings' itself.

- Issue #9212: dict_keys and dict_items now provide the isdisjoint() method, to
  conform to the Set ABC.  Patch by Daniel Urban.

- Issue #9737: Fix a crash when trying to delete a slice or an item from a
  memoryview object.

- Issue #9549: sys.setdefaultencoding() and PyUnicode_SetDefaultEncoding() are
  now removed, since their effect was inexistent in 3.x (the default encoding is
  hardcoded to utf-8 and cannot be changed).

- Issue #7415: PyUnicode_FromEncodedObject() now uses the new buffer API
  properly.  Patch by Stefan Behnel.

- Issue #5553: The Py_LOCAL_INLINE macro now results in inlining on most
  platforms.  Previously, it inlined only when using Microsoft Visual C.

- Issue #9712: Fix tokenize on identifiers that start with non-ascii names.

- Issue #9688: __basicsize__ and __itemsize__ must be accessed as Py_ssize_t.

- Issue #9684: Added a definition for SIZEOF_WCHAR_T to PC/pyconfig.h, to match
  the pyconfig.h generated by configure on other systems.

- Issue #9666: Only catch AttributeError in hasattr(). All other exceptions that
  occur during attribute lookup are now propagated to the caller.

- Issue #8622: Add PYTHONFSENCODING environment variable to override the
  filesystem encoding.

- Issue #5127: The C functions that access the Unicode Database now accept and
  return characters from the full Unicode range, even on narrow unicode builds
  (Py_UNICODE_TOLOWER, Py_UNICODE_ISDECIMAL, and others).  A visible difference
  in Python is that unicodedata.numeric() now returns the correct value for
  large code points, and repr() may consider more characters as printable.

- Issue #9425: Create PyModule_GetFilenameObject() function to get the filename
  as a unicode object, instead of a byte string. Function needed to support
  unencodable filenames. Deprecate PyModule_GetFilename() in favor on the new
  function.

- Issue #8063: Call _PyGILState_Init() earlier in Py_InitializeEx().

- Issue #9612: The set object is now 64-bit clean under Windows.

- Issue #8202: sys.argv[0] is now set to '-m' instead of '-c' when searching for
  the module file to be executed with the -m command line option.

- Issue #9599: Create PySys_FormatStdout() and PySys_FormatStderr() functions to
  write a message formatted by PyUnicode_FromFormatV() to sys.stdout and
  sys.stderr.

- Issue #9542: Create PyUnicode_FSDecoder() function, a ParseTuple converter:
  decode bytes objects to unicode using PyUnicode_DecodeFSDefaultAndSize(); str
  objects are output as-is.

- Issue #9203: Computed gotos are now enabled by default on supported compilers
  (which are detected by the configure script).  They can still be disable
  selectively by specifying --without-computed-gotos.

- Issue #9425: Create PyErr_WarnFormat() function, similar to PyErr_WarnEx() but
  use PyUnicode_FromFormatV() to format the warning message.

- Issue #8530: Prevent stringlib fastsearch from reading beyond the front of an
  array.

- Issue #5319: Print an error if flushing stdout fails at interpreter shutdown.

- Issue #9337: The str() of a float or complex number is now identical to its
  repr().

- Issue #9416: Fix some issues with complex formatting where the output with no
  type specifier failed to match the str output:

    - format(complex(-0.0, 2.0), '-') omitted the real part from the output,
    - format(complex(0.0, 2.0), '-') included a sign and parentheses.

Extensions
----------

- Issue #8734: Avoid crash in msvcrt.get_osfhandle() when an invalid file
  descriptor is provided.  Patch by Pascal Chambon.

- Issue #7736: Release the GIL around calls to opendir() and closedir() in the
  posix module.  Patch by Marcin Bachry.

- Issue #4835: make PyLong_FromSocket_t() and PyLong_AsSocket_t() private to the
  socket module, and fix the width of socket descriptors to be correctly
  detected under 64-bit Windows.

- Issue #1027206: Support IDNA in gethostbyname, gethostbyname_ex, getaddrinfo
  and gethostbyaddr.  getnameinfo is now restricted to numeric addresses as
  input.

- Issue #9214: Set operations on a KeysView or ItemsView in collections now
  correctly return a set.  Patch by Eli Bendersky.

- Issue #5737: Add Solaris-specific mnemonics in the errno module.  Patch by
  Matthew Ahrens.

- Restore GIL in nis_cat in case of error. Decode NIS data to fs encoding, using
  the surrogate error handler.

- Issue #665761: ``functools.reduce()`` will no longer mask exceptions other
  than ``TypeError`` raised by the iterator argument.

- Issue #9570: Use PEP 383 decoding in os.mknod and os.mkfifo.

- Issue #6915: Under Windows, os.listdir() didn't release the Global Interpreter
  Lock around all system calls.  Original patch by Ryan Kelly.

- Issue #8524: Add a detach() method to socket objects, so as to put the socket
  into the closed state without closing the underlying file descriptor.

- Issue #477863: Print a warning at shutdown if gc.garbage is not empty.

- Issue #6869: Fix a refcount problem in the _ctypes extension.

- Issue #5504: ctypes should now work with systems where mmap can't be
  PROT_WRITE and PROT_EXEC.

- Issue #9507: Named tuple repr will now automatically display the right name in
  a tuple subclass.

- Issue #9324: Add parameter validation to signal.signal on Windows in order to
  prevent crashes.

- Issue #9526: Remove some outdated (int) casts that were preventing the array
  module from working correctly with arrays of more than 2**31 elements.

- Fix memory leak in ssl._ssl._test_decode_cert.

- Issue #8065: Fix memory leak in readline module (from failure to free the
  result of history_get_history_state()).

- Issue #9450: Fix memory leak in readline.replace_history_item and
  readline.remove_history_item for readline version >= 5.0.

- Issue #8105: Validate file descriptor passed to mmap.mmap on Windows.

- Issue #8046: Add context manager protocol support and .closed property to mmap
  objects.

Library
-------

- Issue #7451: Improve decoding performance of JSON objects, and reduce the
  memory consumption of said decoded objects when they use the same strings as
  keys.

- Issue #1100562: Fix deep-copying of objects derived from the list and dict
  types.  Patch by Michele Orrù and Björn Lindqvist.

- Issue #9753: Fixed socket.dup, which did not always work correctly on Windows.

- Issue #9421: Made the get<type> methods consistently accept the vars and
  default arguments on all parser classes.

- Issue #7005: Fixed output of None values for RawConfigParser.write and
  ConfigParser.write.

- Issue #8990: array.fromstring() and array.tostring() get renamed to
  frombytes() and tobytes(), respectively, to avoid confusion.  Furthermore,
  array.frombytes(), array.extend() as well as the array.array() constructor now
  accept bytearray objects.  Patch by Thomas Jollans.

- Issue #808164: Fixed socket.close to avoid references to globals, to avoid
  issues when socket.close is called from a __del__ method.

- Issue #9706: ssl module provides a better error handling in various
  circumstances.

- Issue #1868: Eliminate subtle timing issues in thread-local objects by getting
  rid of the cached copy of thread-local attribute dictionary.

- Issue #1512791: In setframerate() in the wave module, non-integral frame rates
  are rounded to the nearest integer.

- Issue #8797: urllib2 does a retry for Basic Authentication failure instead of
  falling into recursion.

- Issue #1194222: email.utils.parsedate now returns RFC2822 compliant four
  character years even if the message contains RFC822 two character years.

- Issue #8750: Fixed MutableSet's methods to correctly handle reflexive
  operations on its self, namely x -= x and x ^= x.

- Issue #9129: smtpd.py is vulnerable to DoS attacks deriving from missing error
  handling when accepting a new connection.

- Issue #9601: ftplib now provides a workaround for non-compliant
  implementations such as IIS shipped with Windows server 2003 returning invalid
  response codes for MKD and PWD commands.

- Issue #658749: asyncore's connect() method now correctly interprets winsock
  errors.

- Issue #9501: Fixed logging regressions in cleanup code.

- Fix functools.total_ordering() to skip methods inherited from object.

- Issue #9572: Importlib should not raise an exception if a directory it thought
  it needed to create was done concurrently by another process.

- Issue #9617: Signals received during a low-level write operation aren't
  ignored by the buffered IO layer anymore.

- Issue #843590: Make "macintosh" an alias to the "mac_roman" encoding.

- Create os.fsdecode(): decode from the filesystem encoding with surrogateescape
  error handler, or strict error handler on Windows.

- Issue #3488: Provide convenient shorthand functions ``gzip.compress`` and
  ``gzip.decompress``.  Original patch by Anand B. Pillai.

- Issue #8807: poplib.POP3_SSL class now accepts a context parameter, which is a
  ssl.SSLContext object allowing bundling SSL configuration options,
  certificates and private keys into a single (potentially long-lived)
  structure.

- Issue #8866: parameters passed to socket.getaddrinfo can now be specified as
  single keyword arguments.

- Address XXX comment in dis.py by having inspect.py prefer to reuse the dis.py
  compiler flag values over defining its own.

- Issue #9147: Added dis.code_info() which is similar to show_code() but returns
  formatted code information in a string rather than displaying on screen.

- Issue #9567: functools.update_wrapper now adds a __wrapped__ attribute
  pointing to the original callable.

- Issue #3445: functools.update_wrapper now tolerates missing attributes on
  wrapped callables.

- Issue #5867: Add abc.abstractclassmethod and abc.abstractstaticmethod.

- Issue #9605: posix.getlogin() decodes the username with file filesystem
  encoding and surrogateescape error handler. Patch written by David Watson.

- Issue #9604: posix.initgroups() encodes the username using the fileystem
  encoding and surrogateescape error handler. Patch written by David Watson.

- Issue #9603: posix.ttyname() and posix.ctermid() decode the terminal name
  using the filesystem encoding and surrogateescape error handler. Patch written
  by David Watson.

- Issue #7647: The posix module now has the ST_RDONLY and ST_NOSUID constants,
  for use with the statvfs() function.  Patch by Adam Jackson.

- Issue #8688: MANIFEST files created by distutils now include a magic comment
  indicating they are generated.  Manually maintained MANIFESTs without this
  marker will not be overwritten or removed.

- Issue #7467: when reading a file from a ZIP archive, its CRC is checked and a
  BadZipfile error is raised if it doesn't match (as used to be the case in
  Python 2.5 and earlier).

- Issue #9550: a BufferedReader could issue an additional read when the original
  read request had been satisfied, which could block indefinitely when the
  underlying raw IO channel was e.g. a socket.  Report and original patch by
  Jason V. Miller.

- Issue #3757: thread-local objects now support cyclic garbage collection.
  Thread-local objects involved in reference cycles will be deallocated timely
  by the cyclic GC, even if the underlying thread is still running.

- Issue #9452: Add read_file, read_string, and read_dict to the configparser
  API; new source attribute to exceptions.

- Issue #6231: Fix xml.etree.ElementInclude to include the tail of the current
  node.

- Issue #8047: Fix the xml.etree serializer to return bytes by default.  Use
  ``encoding="unicode"`` to generate a Unicode string.

- Issue #8280: urllib2's Request method will remove fragments in the url.  This
  is how it is supposed to work, wget and curl do the same.  Previous behavior
  was wrong.

- Issue #6683: For SMTP logins we now try all authentication methods advertised
  by the server.  Many servers are buggy and advertise authentication methods
  they do not support in reality.

- Issue #8814: function annotations (the ``__annotations__`` attribute) are now
  included in the set of attributes copied by default by functools.wraps and
  functools.update_wrapper.  Patch by Terrence Cole.

- Issue #2944: asyncore doesn't handle connection refused correctly.

- Issue #4184: Private attributes on smtpd.SMTPChannel made public and deprecate
  the private attributes. Add tests for smtpd module.

- Issue #3196: email header decoding is now forgiving if an RFC2047 encoded word
  encoded in base64 is lacking padding.

- Issue #9444: Argparse now uses the first element of prefix_chars as the option
  character for the added 'h/help' option if prefix_chars does not contain a
  '-', instead of raising an error.

- Issue #7372: Fix pstats regression when stripping paths from profile data
  generated with the profile module.

- Issue #9428: Fix running scripts with the profile/cProfile modules from the
  command line.

- Issue #7781: Fix restricting stats by entry counts in the pstats interactive
  browser.

- Issue #9209: Do not crash in the pstats interactive browser on invalid regular
  expressions.

- Update collections.OrderedDict to match the implementation in Py2.7 (based on
  lists instead of weakly referenced Link objects).

- Issue #8397: Raise an error when attempting to mix iteration and regular reads
  on a BZ2File object, rather than returning incorrect results.

- Issue #9448: Fix a leak of OS resources (mutexes or semaphores) when
  re-initializing a buffered IO object by calling its ``__init__`` method.

- Issue #1713: Fix os.path.ismount(), which returned true for symbolic links
  across devices.

- Issue #8826: Properly load old-style "expires" attribute in http.cookies.

- Issue #1690103: Fix initial namespace for code run with trace.main().

- Issue #7395: Fix tracebacks in pstats interactive browser.

- Issue #8230: Fix Lib/test/sortperf.py.

- Issue #8620: when a Cmd is fed input that reaches EOF without a final newline,
  it no longer truncates the last character of the last command line.

- Issue #5146: Handle UID THREAD command correctly in imaplib.

- Issue #5147: Fix the header generated for cookie files written by
  http.cookiejar.MozillaCookieJar.

- Issue #8198: In pydoc, output all help text to the correct stream when
  sys.stdout is reassigned.

- Issue #7909: Do not touch paths with the special prefixes ``\\.\`` or ``\\?\``
  in ntpath.normpath().

- Issue #1286: Allow using fileinput.FileInput as a context manager.

- Add lru_cache() decorator to the functools module.

Tools/Demos
-----------

- Fix ``Tools/scripts/checkpyc.py`` after PEP 3147.

- Issue #8867: Fix ``Tools/scripts/serve.py`` to work with files containing
  non-ASCII content.

Tests
-----

- Issue #9601: Provide a test case for ftplib.parse257.

- Issue #8857: Provide a test case for socket.getaddrinfo.

- Issue #7564: Skip test_ioctl if another process is attached to /dev/tty.

- Issue #8433: Fix test_curses failure with newer versions of ncurses.

- Issue #9496: Provide a test suite for the rlcompleter module.  Patch by
  Michele Orrù.

- Issue #8687: provide a test suite for sched.py module.

Build
-----

- Issue #1303434: Generate ZIP file containing all PDBs.

- Issue #9193: PEP 3149 is accepted.

- Issue #3101: Helper functions _add_one_to_index_C() and _add_one_to_index_F()
  become _Py_add_one_to_index_C() and _Py_add_one_to_index_F(), respectively.

- Issue #9700: define HAVE_BROKEN_POSIX_SEMAPHORES under AIX 6.x.  Patch by
  Sébastien Sablé.

- Don't run pgen twice when using make -j.


What's New in Python 3.2 Alpha 1?
=================================

*Release date: 01-Aug-2010*

Core and Builtins
-----------------

- Issue #8991: convertbuffer() rejects discontigious buffers.

- Issue #7616: Fix copying of overlapping memoryview slices with the Intel
  compiler.

- Issue #8413: structsequence now subclasses tuple.

- Issue #8271: during the decoding of an invalid UTF-8 byte sequence, only the
  start byte and the continuation byte(s) are now considered invalid, instead of
  the number of bytes specified by the start byte.  E.g.:
  '\xf1\x80AB'.decode('utf-8', 'replace') now returns u'\ufffdAB' and replaces
  with U+FFFD only the start byte ('\xf1') and the continuation byte ('\x80')
  even if '\xf1' is the start byte of a 4-bytes sequence.  Previous versions
  returned a single u'\ufffd'.

- Issue #9011: A negated imaginary literal (e.g., "-7j") now has real part -0.0
  rather than 0.0.  So "-7j" is now exactly equivalent to "-(7j)".

- Be more specific in error messages about positional arguments.

- Issue #8949: "z" format of PyArg_Parse*() functions doesn't accept bytes
  objects, as described in the documentation.

- Issue #6543: Write the traceback in the terminal encoding instead of utf-8.
  Fix the encoding of the modules filename.  Patch written by Amaury Forgeot
  d'Arc.

- Issue #9011: Remove buggy and unnecessary (in 3.x) ST->AST compilation code
  dealing with unary minus applied to a constant.  The removed code was mutating
  the ST, causing a second compilation to fail.

- Issue #850997: mbcs encoding (Windows only) handles errors argument: strict
  mode raises unicode errors.  The encoder only supports "strict" and "replace"
  error handlers, the decoder only supports "strict" and "ignore" error
  handlers.  Patch written by Mark Hammond.

- Issue #8850: Remove "w" and "w#" formats from PyArg_Parse*() functions, use
  "w*" format instead. Add tests for "w*" format.

- Issue #8592: PyArg_Parse*() functions raise a TypeError for "y", "u" and "Z"
  formats if the string contains a null byte/character.  Write unit tests for
  string formats.

- Issue #7490: To facilitate sharing of doctests between 2.x and 3.x test
  suites, the IGNORE_EXCEPTION_DETAIL directive now also ignores the module
  location of the raised exception.

- Issue #8969: On Windows, use mbcs codec in strict mode to encode and decode
  filenames and enable os.fsencode().

- Issue #9058: Remove assertions about INT_MAX in UnicodeDecodeError.

- Issue #8941: Decoding big endian UTF-32 data in UCS-2 builds could crash the
  interpreter with characters outside the Basic Multilingual Plane (higher than
  0x10000).

- Issue #8950: (See also issue #5080).  Py_ArgParse*() functions now raise
  TypeError instead of giving a DeprecationWarning when a float is parsed using
  the 'L' code (for long long).  (All other integer codes already raise
  TypeError in this case.)

- Issue #8922: Normalize the encoding name in PyUnicode_AsEncodedString() to
  enable shortcuts for upper case encoding name. Add also a shortcut for
  "iso-8859-1" in PyUnicode_AsEncodedString() and PyUnicode_Decode().

- Issue #8838: Remove codecs.charbuffer_encode() function.  The buffer protocol
  doesn't support "char buffer" anymore in Python 3.

- Issue #8339: Remove "t#" format of PyArg_Parse*() functions, use "s#" or "s*"
  instead.  codecs.charbuffer_encode() now accepts modifiable buffer objects
  like bytearray.

- Issue #8837: Remove "O?" format of PyArg_Parse*() functions.  The format is no
  used anymore and it was never documented.

- In str.format(), raise a ValueError when indexes to arguments are too large.

- Issue #2844: Make int('42', n) consistently raise ValueError for invalid
  integers n (including n = -909).

- Issue #8188: Introduce a new scheme for computing hashes of numbers (instances
  of int, float, complex, decimal.Decimal and fractions.Fraction) that makes it
  easy to maintain the invariant that hash(x) == hash(y) whenever x and y have
  equal value.

- Issue #8748: Fix two issues with comparisons between complex and integer
  objects.  (1) The comparison could incorrectly return True in some cases
  (2**53+1 == complex(2**53) == 2**53), breaking transitivity of equality.
  (2) The comparison raised an OverflowError for large integers, leading to
  unpredictable exceptions when combining integers and complex objects in sets
  or dicts.

- Issue #8766: Initialize _warnings module before importing the first module.
  Fix a crash if an empty directory called "encodings" exists in sys.path.

- Issue #8589: Decode PYTHONWARNINGS environment variable with the file system
  encoding and surrogateescape error handler instead of the locale encoding to
  be consistent with os.environ.  Add PySys_AddWarnOptionUnicode() function.

- PyObject_Dump() encodes unicode objects to utf8 with backslashreplace (instead
  of strict) error handler to escape surrogates.

- Issue #8715: Create PyUnicode_EncodeFSDefault() function: Encode a Unicode
  object to Py_FileSystemDefaultEncoding with the "surrogateescape" error
  handler, and return bytes.  If Py_FileSystemDefaultEncoding is not set, fall
  back to UTF-8.

- Enable shortcuts for common encodings in PyUnicode_AsEncodedString() for any
  error handler, not only the default error handler (strict).

- Issue #8610: Load file system codec at startup, and display a fatal error on
  failure.  Set the file system encoding to utf-8 (instead of None) if getting
  the locale encoding failed, or if nl_langinfo(CODESET) function is missing.

- PyFile_FromFd() uses PyUnicode_DecodeFSDefault() instead of
  PyUnicode_FromString() to support surrogates in the filename and use the right
  encoding.

- Issue #7507: Quote "!" in pipes.quote(); it is special to some shells.

- PyUnicode_DecodeFSDefaultAndSize() uses surrogateescape error handler.

- Issue #8419: Prevent the dict constructor from accepting non-string keyword
  arguments.

- Issue #8124: PySys_WriteStdout() and PySys_WriteStderr() don't execute
  indirectly Python signal handlers anymore because mywrite() ignores exceptions
  (KeyboardInterrupt).

- Issue #8092: Fix PyUnicode_EncodeUTF8() to support error handler producing
  unicode string (eg. backslashreplace).

- Issue #8485: PyUnicode_FSConverter() doesn't accept byteearray objects
  anymore, you have to convert your bytearray filenames to bytes.

- Issue #7332: Remove the 16KB stack-based buffer in
  PyMarshal_ReadLastObjectFromFile, which doesn't bring any noticeable benefit
  compared to the dynamic memory allocation fallback.  Patch by Charles-François
  Natali.

- Issue #8417: Raise an OverflowError when an integer larger than sys.maxsize is
  passed to bytes or bytearray.

- Issue #7301: Add environment variable $PYTHONWARNINGS.

- Issue #8329: Don't return the same lists from select.select when no fds are
  changed.

- Issue #8259: 1L << (2**31) no longer produces an 'outrageous shift error' on
  64-bit machines.  The shift count for either left or right shift is permitted
  to be up to sys.maxsize.

- Ensure that tokenization of identifiers is not affected by locale.

- Issue #1222585: Added LDCXXSHARED for C++ support. Patch by Arfrever.

- Raise a TypeError when trying to delete a T_STRING_INPLACE struct member.

- Issue #8211: Save/restore CFLAGS around AC_PROG_CC in configure.in, in case it
  is set.

- Issue #8226: sys.setfilesystemencoding() raises a LookupError if the encoding
  is unknown.

- Issue #1583863: A str subclass can now override the __str__ method.

- Issue #8014: Setting a T_UINT or T_PYSSIZET attribute of an object with
  PyMemberDefs could produce an internal error; raise TypeError instead.

- Issue #7845: Rich comparison methods on the complex type now return
  NotImplemented rather than raising a TypeError when comparing with an
  incompatible type; this allows user-defined classes to implement their own
  comparisons with complex.

- Issue #3137: Don't ignore errors at startup, especially a keyboard interrupt
  (SIGINT). If an error occurs while importing the site module, the error is
  printed and Python exits. Initialize the GIL before importing the site module.

- Issue #7173: Generator finalization could invalidate sys.exc_info().

- Issue #7544: Preallocate thread memory before creating the thread to avoid a
  fatal error in low memory condition.

- Issue #7820: The parser tokenizer restores all bytes in the right if the BOM
  check fails.

- Handle errors from looking up __prepare__ correctly.

- Issue #5939: Add additional runtime checking to ensure a valid capsule in
  Modules/_ctypes/callproc.c.

- Issue #7309: Fix unchecked attribute access when converting
  UnicodeEncodeError, UnicodeDecodeError, and UnicodeTranslateError to strings.

- Issue #6902: Fix problem with built-in types format incorrectly with 0
  padding.

- Issue #7988: Fix default alignment to be right aligned for complex.__format__.
  Now it matches other numeric types.

- Issue #5988: Remove deprecated functions PyOS_ascii_formatd,
  PyOS_ascii_strtod, and PyOS_ascii_atof.  Use PyOS_double_to_string and
  PyOS_string_to_double instead.  See issue #5835 for the original deprecations.

- Issue #7385: Fix a crash in `MemoryView_FromObject` when `PyObject_GetBuffer`
  fails.  Patch by Florent Xicluna.

- Issue #7788: Fix an interpreter crash produced by deleting a list slice with
  very large step value.

- Issue #7766: Change sys.getwindowsversion() return value to a named tuple and
  add the additional members returned in an OSVERSIONINFOEX structure.  The new
  members are service_pack_major, service_pack_minor, suite_mask, and
  product_type.

- Issue #7561: Operations on empty bytearrays (such as `int(bytearray())`) could
  crash in many places because of the PyByteArray_AS_STRING() macro returning
  NULL.  The macro now returns a statically allocated empty string instead.

- Issue #6690: Optimize the bytecode for expressions such as `x in {1, 2, 3}`,
  where the right hand operand is a set of constants, by turning the set into a
  frozenset and pre-building it as a constant.  The comparison operation is made
  against the constant instead of building a new set each time it is executed (a
  similar optimization already existed which turned a list of constants into a
  pre-built tuple).  Patch and additional tests by Dave Malcolm.

- Issue #7622: Improve the split(), rsplit(), splitlines() and replace() methods
  of bytes, bytearray and unicode objects by using a common implementation based
  on stringlib's fast search.  Patch by Florent Xicluna.

- Issue #7632: Fix various str -> float conversion bugs present in 2.7 alpha 2,
  including: (1) a serious 'wrong output' bug that could occur for long (> 40
  digit) input strings, (2) a crash in dtoa.c that occurred in debug builds when
  parsing certain long numeric strings corresponding to subnormal values, (3) a
  memory leak for some values large enough to cause overflow, and (4) a number
  of flaws that could lead to incorrectly rounded results.

- The __complex__ method is now looked up on the class of instances to make it
  consistent with other special methods.

- Issue #7462: Implement the stringlib fast search algorithm for the `rfind`,
  `rindex`, `rsplit` and `rpartition` methods.  Patch by Florent Xicluna.

- Issue #7604: Deleting an unset slotted attribute did not raise an
  AttributeError.

- Issue #7534: Fix handling of IEEE specials (infinities, nans, negative zero)
  in ** operator.  The behaviour now conforms to that described in C99 Annex F.

- Issue #1811: improve accuracy and cross-platform consistency for true division
  of integers: the result of a/b is now correctly rounded for ints a and b (at
  least on IEEE 754 platforms), and in particular does not depend on the
  internal representation of an int.

- Issue #6834: replace the implementation for the 'python' and 'pythonw'
  executables on OSX.

  These executables now work properly with the arch(1) command: ``arch -ppc
  python`` will start a universal binary version of python in PPC mode (unlike
  previous releases).

- Issue #7466: Segmentation fault when the garbage collector is called in the
  middle of populating a tuple.  Patch by Florent Xicluna.

- Issue #7419: setlocale() could crash the interpreter on Windows when called
  with invalid values.

- Issue #6077: On Windows, files opened with tempfile.TemporaryFile in "wt+"
  mode would appear truncated on the first '0x1a' byte (aka. Ctrl+Z).

- Issue #7085: Fix crash when importing some extensions in a thread on MacOSX
  10.6.

- Issue #1757126: Fix the cyrillic-asian alias for the ptcp154 encoding.

- Issue #6970: Remove redundant calls when comparing objects that don't
  implement the relevant rich comparison methods.

- Issue #7298: Fixes for range and reversed(range(...)).  Iteration over
  range(a, b, c) incorrectly gave an empty iterator when a, b and c fit in C
  long but the length of the range did not.  Also fix several cases where
  reversed(range(a, b, c)) gave wrong results, and fix a refleak for
  reversed(range(a, b, c)) with large arguments.

- Issue #7244: itertools.izip_longest() no longer ignores exceptions raised
  during the formation of an output tuple.

- Issue #3297: On wide unicode builds, do not split unicode characters into
  surrogates.

- Remove length limitation when constructing a complex number from a string.

- Issue #1087418: Boost performance of bitwise operations for longs.

- Support for AtheOS has been completely removed from the code base. It was
  disabled since Python 3.0.

- Support for several legacy threading libraries has been disabled. These
  libraries are: Mach C threads, SunOS LWP, GNU pth, Irix threads. Support code
  will be entirely removed in 3.3.

- Support for OSF* has been disabled. If nobody stands up, support will be
  removed in 3.3. See <http://bugs.python.org/issue8606>.

- Peephole constant folding had missed UNARY_POSITIVE.

- Issue #1722344: threading._shutdown() is now called in Py_Finalize(), which
  fixes the problem of some exceptions being thrown at shutdown when the
  interpreter is killed. Patch by Adam Olsen.

- Issue #7147: Remove support for compiling Python without complex number
  support.

- Issue #7120: logging: Removed import of multiprocessing which is causing crash
  in GAE.

- Issue #1754094: Improve the stack depth calculation in the compiler.  There
  should be no other effect than a small decrease in memory use.  Patch by
  Christopher Tur Lesniewski-Laas.

- Issue #7065: Fix a crash in bytes.maketrans and bytearray.maketrans when using
  byte values greater than 127.  Patch by Derk Drukker.

- Issue #1571184: The Unicode database contains properties for more characters.
  The tables for code points representing numeric values, white spaces or line
  breaks are now generated from the official Unicode Character Database files,
  and include information from the Unihan.txt file.

- Issue #7019: Raise ValueError when unmarshalling bad long data, instead of
  producing internally inconsistent Python longs.

- Issue #6990: Fix threading.local subclasses leaving old state around after a
  reference cycle GC which could be recycled by new locals.

- Issue #5460: Fix an ambiguity in the grammar.

- Issue #1766304: Improve performance of membership tests on range objects.

- Issue #6713: Improve performance of integer -> string conversions.

- Issue #6846: Fix bug where bytearray.pop() returns negative integers.

- Issue #6750: A text file opened with io.open() could duplicate its output when
  writing from multiple threads at the same time.

- Issue #6707: dir() on an uninitialized module caused a crash.

- Issue #6540: Fixed crash for bytearray.translate() with invalid parameters.

- Issue #6573: set.union() stopped processing inputs if an instance of self
  occurred in the argument chain.

- Issue #6070: On posix platforms import no longer copies the execute bit from
  the .py file to the .pyc file if it is set.

- Issue #1616979: Added the cp720 (Arabic DOS) encoding.

- Issue #6428: Since Python 3.0, the __bool__ method must return a bool object,
  and not an int.  Fix the corresponding error message, and the documentation.

- The deprecated PyCObject has been removed.

- Issue #6347: Include inttypes.h as well as stdint.h in pyport.h.  This fixes a
  build failure on HP-UX: int32_t and uint32_t are defined in inttypes.h instead
  of stdint.h on that platform.

- Issue #6373: Fixed a SystemError when encoding with the latin-1 codec and the
  'surrogateescape' error handler, a string which contains unpaired surrogates.

- Issue #4856: Remove checks for win NT.

- Issue #6687: PyBytes_FromObject() no longer accepts an integer as its argument
  to construct a null-initialized bytes object.

- Issue #1023290: Add from_bytes() and to_bytes() methods to integers.  These
  methods allow the conversion of integers to bytes, and vice-versa.

- Issue #7382: Fix bug in bytes.__getnewargs__ that prevented bytes instances
  from being copied with copy.copy(), and bytes subclasses from being pickled
  properly.

- Code objects now support weak references.

- Issue #7072: isspace(0xa0) is true on Mac OS X.

- Issue #8084: PEP 370 now conforms to system conventions for framework builds
  on MacOS X. That is, "python setup.py install --user" will install into
  "~/Library/Python/2.7" instead of "~/.local".

C-API
-----

- Issue #2443: A new macro, `Py_VA_COPY`, copies the state of the
  variable argument list.  `Py_VA_COPY` is equivalent to C99
  `va_copy`, but available on all python platforms.

- PySlice_GetIndicesEx now clips the step to [-PY_SSIZE_T_MAX, PY_SSIZE_T_MAX]
  instead of [-PY_SSIZE_T_MAX-1, PY_SSIZE_T_MAX].  This makes it safe to do
  "step = -step" when reversing a slice.

- Issue #5753: A new C API function, `PySys_SetArgvEx`, allows embedders of the
  interpreter to set sys.argv without also modifying sys.path.  This helps fix
  `CVE-2008-5983
  <http://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2008-5983>`_.

- Add PyArg_ValidateKeywordArguments, which checks if all keyword arguments are
  strings in an efficient manner.

- Issue #8276: PyEval_CallObject() is now only available in macro form.  The
  function declaration, which was kept for backwards compatibility reasons, is
  now removed (the macro was introduced in 1997!).

- Issue #7767: New function PyLong_AsLongLongAndOverflow added, analogous to
  PyLong_AsLongAndOverflow.

- Make PyUnicode_CompareWithASCIIString return not equal if the Python string
  has '\0' at the end.

- Issue #5080: The argument parsing functions PyArg_ParseTuple,
  PyArg_ParseTupleAndKeywords, PyArg_VaParse, PyArg_VaParseTupleAndKeywords and
  PyArg_Parse now raise a DeprecationWarning for float arguments passed with the
  'L' format code.  This will become a TypeError in a future version of Python,
  to match the behaviour of the other integer format codes.

- Issue #7033: Function ``PyErr_NewExceptionWithDoc()`` added.

- Issue #7414: 'C' code wasn't being skipped properly (for keyword arguments) in
  PyArg_ParseTupleAndKeywords.

- Issue #7228: Add '%lld' and '%llu' support to PyString_FromFormat(V) and
  PyErr_Format, on machines with HAVE_LONG_LONG defined.

- Issue #6151: Made PyDescr_COMMON conform to standard C (like PyObject_HEAD in
  PEP 3123).  The PyDescr_TYPE and PyDescr_NAME macros should be should used for
  accessing the d_type and d_name members of structures using PyDescr_COMMON.

- Issue #6405: Remove duplicate type declarations in descrobject.h.

- The code flags for old __future__ features are now available again.

- Issue #5954: Add a PyFrame_GetLineNumber() function to replace most uses of
  PyCode_Addr2Line().

- Issue #5959: Add a PyCode_NewEmpty() function to create a new empty code
  object at a specified file, function, and line number.

- Issue #1419652: Change the first argument to PyImport_AppendInittab() to
  ``const char *`` as the string is stored beyond the call.

- Issue #2422: When compiled with the ``--with-valgrind`` option, the pymalloc
  allocator will be automatically disabled when running under Valgrind.  This
  gives improved memory leak detection when running under Valgrind, while taking
  advantage of pymalloc at other times.

Library
-------

- In pdb, when Ctrl-C is entered while defining commands for a breakpoint, the
  old commands are restored.

- For traceback debugging, the pdb listing now also shows the locations where
  the exception was originally (re)raised, if it differs from the last line
  executed (e.g. in case of finally clauses).

- The pdb command "source" has been added.  It displays the source code for a
  given object, if possible.

- The pdb command "longlist" has been added.  It displays the whole source code
  for the current function.

- Issue #1503502: Make pdb.Pdb easier to subclass by putting message and error
  output into methods.

- Issue #809887: Make the output of pdb's breakpoint deletions more consistent;
  emit a message when a breakpoint is enabled or disabled.

- Issue #5294: Fix the behavior of pdb's "continue" command when called in the
  top-level debugged frame.

- Issue #5727: Restore the ability to use readline when calling into pdb in
  doctests.

- Issue #6719: In pdb, do not stop somewhere in the encodings machinery if the
  source file to be debugged is in a non-builtin encoding.

- Issue #8048: Prevent doctests from failing when sys.displayhook has been
  reassigned.

- Issue #8015: In pdb, do not crash when an empty line is entered as a
  breakpoint command.

- In pdb, allow giving a line number to the "until" command.

- Issue #1437051: For pdb, allow "continue" and related commands in .pdbrc
  files.  Also, add a command-line option "-c" that runs a command as if given
  in .pdbrc.

- Issue #4179: In pdb, allow "list ." as a command to return to the currently
  debugged line.

- Issue #4108: In urllib.robotparser, if there are multiple ``User-agent: *``
  entries, consider the first one.

- Issue #6630: Allow customizing regex flags when subclassing the
  string.Template class.

- Issue #9411: Allow specifying an encoding for config files in the configparser
  module.

- Issue #1682942: Improvements to configparser: support alternate delimiters,
  alternate comment prefixes and empty lines in values.

- Issue #9354: Provide getsockopt() in asyncore's file_wrapper.

- Issue #8966: ctypes: Remove implicit bytes-unicode conversion.

- Issue #9378: python -m pickle <pickle file> will now load and display the
  first object in the pickle file.

- Issue #4770: Restrict binascii module to accept only bytes (as specified).
  And fix the email package to encode to ASCII instead of ``raw-unicode-escape``
  before ASCII-to-binary decoding.

- Issue #9384: ``python -m tkinter`` will now display a simple demo applet.

- The default size of the re module's compiled regular expression cache has been
  increased from 100 to 500 and the cache replacement policy has changed from
  simply clearing the entire cache on overflow to forgetting the least recently
  used cached compiled regular expressions.  This is a performance win for
  applications that use a lot of regular expressions and limits the impact of
  the performance hit anytime the cache is exceeded.

- Issue #7113: Speed up loading in configparser. Patch by Łukasz Langa.

- Issue #9032: XML-RPC client retries the request on EPIPE error.  The EPIPE
  error occurs when the server closes the socket and the client sends a big
  XML-RPC request.

- Issue #4629: getopt raises an error if an argument ends with "=", whereas
  getopt doesn't accept a value (eg. --help= is rejected if getopt uses
  ['help='] long options).

- Issue #7989: Added pure python implementation of the `datetime` module.  The C
  module is renamed to `_datetime` and if available, overrides all classes
  defined in datetime with fast C impementation.  Python implementation is based
  on the original python prototype for the datetime module by Tim Peters with
  minor modifications by the PyPy project.  The test suite now tests `datetime`
  module with and without `_datetime` acceleration using the same test cases.

- Issue #7895: platform.mac_ver() no longer crashes after calling os.fork().

- Issue #9323: Fixed a bug in trace.py that resulted in loosing the name of the
  script being traced.  Patch by Eli Bendersky.

- Issue #9282: Fixed --listfuncs option of trace.py.  Thanks Eli Bendersky for
  the patch.

- Issue #3704: http.cookiejar was not properly handling URLs with a / in the
  parameters.

- Issue #9268: ``pickletools.dis()`` now has an optional *annotate* argument
  which controls printing of opcode descriptions in ``dis()`` output.

- Issue #1555570: email no longer inserts extra blank lines when a \r\n combo
  crosses an 8192 byte boundary.

- Issue #9243: Fix sndhdr module and add unit tests, contributed by James Lee.

- ``ast.literal_eval()`` now allows byte literals.

- Issue #9137: Fix issue in MutableMapping.update, which incorrectly treated
  keyword arguments called 'self' or 'other' specially.

- ``ast.literal_eval()`` now allows set literals.

- Issue #9164: Ensure that sysconfig handles duplicate -arch flags in CFLAGS.

- Issue #7646: The fnmatch pattern cache no longer grows without bound.

- Issue #9136: Fix 'dictionary changed size during iteration' RuntimeError
  produced when profiling the decimal module.  This was due to a dangerous
  iteration over 'locals()' in Context.__init__.

- Fix extreme speed issue in Decimal.pow when the base is an exact power of 10
  and the exponent is tiny (for example, ``Decimal(10) **
  Decimal('1e-999999999')``).

- Issue #9186: Fix math.log1p(-1.0) to raise ValueError, not OverflowError.

- Issue #9130: Fix validation of relative imports in parser module.

- Issue #9128: Fix validation of class decorators in parser module.

- Issue #9094: python -m pickletools will now disassemble pickle files listed in
  the command line arguments.  See output of python -m pickletools -h for more
  details.

- Issue #5468: urlencode to handle bytes type and other encodings in its query
  parameter. Patch by Dan Mahn.

- Issue #7673: Fix security vulnerability (CVE-2010-2089) in the audioop module,
  ensure that the input string length is a multiple of the frame size.

- Issue #6507: Accept source strings in dis.dis().  Original patch by Daniel
  Urban.

- Issue #7829: Clearly document that the dis module is exposing an
  implementation detail that is not stable between Python VMs or releases.

- Issue #6589: cleanup asyncore.socket_map in case smtpd.SMTPServer constructor
  raises an exception.

- Issue #9110: Addition of ContextDecorator to contextlib, for creating APIs
  that act as both context managers and decorators. contextmanager changes to
  use ContextDecorator.

- Implement importlib.abc.SourceLoader and deprecate PyLoader and PyPycLoader
  for removal in Python 3.4.

- Issue #9064: pdb's "up" and "down" commands now accept an optional argument
  giving the number of frames to go.

- Issue #9018: os.path.normcase() now raises a TypeError if the argument is not
  ``str`` or ``bytes``.

- Issue #9075: In the ssl module, remove the setting of a ``debug`` flag on an
  OpenSSL structure.

- Issue #8682: The ssl module now temporary increments the reference count of a
  socket object got through ``PyWeakref_GetObject``, so as to avoid possible
  deallocation while the object is still being used.

- Issue #1368368: FancyURLOpener class changed to throw an Exception on wrong
  password instead of presenting an interactive prompt.  Older behavior can be
  obtained by passing retry=True to http_error_xxx methods of FancyURLOpener.

- Issue #8720: Fix regression caused by fix for #4050 by making getsourcefile
  smart enough to find source files in the linecache.

- Issue #5610: feedparser no longer eats extra characters at the end of a body
  part if the body part ends with a ``\r\n``.

- Issue #8986: math.erfc was incorrectly raising OverflowError for values
  between -27.3 and -30.0 on some platforms.

- Issue #8784: Set tarfile default encoding to 'utf-8' on Windows.

- Issue #8966: If a ctypes structure field is an array of c_char, convert its
  value to bytes instead of str (as done for c_char and c_char_p).

- Issue #8188: Comparisons between Decimal and Fraction objects are now
  permitted, returning a result based on the exact numerical values of the
  operands.  This builds on issue #2531, which allowed Decimal-to-float
  comparisons; all comparisons involving numeric types (bool, int, float,
  complex, Decimal, Fraction) should now act as expected.

- Issue #8897: Fix sunau module, use bytes to write the header. Patch written by
  Thomas Jollans.

- Issue #8899: time.struct_time now has class and atribute docstrings.

- Issue #6470: Drop UNC prefix in FixTk.

- Issue #4768: base64 encoded email body parts were incorrectly stored as binary
  strings.  They are now correctly converted to strings.

- Issue #8833: tarfile created hard link entries with a size field != 0 by
  mistake.

- Charset.body_encode now correctly handles base64 encoding by encoding with the
  output_charset before calling base64mime.encode.  Passes the tests from 2.x
  issue #1368247.

- Issue #8845: sqlite3 Connection objects now have a read-only in_transaction
  attribute that is True iff there are uncommitted changes.

- Issue #1289118: datetime.timedelta objects can now be multiplied by float and
  divided by float and int objects.  Results are rounded to the nearest multiple
  of timedelta.resolution with ties resolved using round-half-to-even method.

- Issue #7150: Raise OverflowError if the result of adding or subtracting
  timedelta from date or datetime falls outside of the MINYEAR:MAXYEAR range.

- Issue #8806: add SSL contexts support to ftplib.

- Issue #4769: Fix main() function of the base64 module, use sys.stdin.buffer
  and sys.stdout.buffer (instead of sys.stdin and sys.stdout) to use the bytes
  API.

- Issue #8770: Now sysconfig displays information when it's called as a script.
  Initial idea by Sridhar Ratnakumar.

- Issue #6662: Fix parsing of malformatted charref (&#bad;), patch written by
  Fredrik Håård.

- Issue #8540: Decimal module: rename the Context._clamp attribute to
  Context.clamp and make it public.  This is useful in creating contexts that
  correspond to the decimal interchange formats specified in IEEE 754.

- Issue #6268: Fix seek() method of codecs.open(), don't read or write the BOM
  twice after seek(0). Fix also reset() method of codecs, UTF-16, UTF-32 and
  StreamWriter classes.

- Issue #3798: sys.exit(message) writes the message to sys.stderr file, instead
  of the C file stderr, to use stderr encoding and error handler.

- Issue #8782: Add a trailing newline in linecache.updatecache to the last line
  of files without one.

- Issue #8729: Return NotImplemented from collections.Mapping.__eq__ when
  comparing to a non-mapping.

- Issue #8774: tabnanny uses the encoding cookie (#coding:...) to use the
  correct encoding.

- Issue #4870: Add an `options` attribute to SSL contexts, as well as several
  ``OP_*`` constants to the `ssl` module.  This allows to selectively disable
  protocol versions, when used in combination with `PROTOCOL_SSLv23`.

- Issue #8759: Fixed user paths in sysconfig for posix and os2 schemes.

- Issue #8663: distutils.log emulates backslashreplace error handler. Fix
  compilation in a non-ASCII directory if stdout encoding is ASCII (eg. if
  stdout is not a TTY).

- Issue #8513: os.get_exec_path() supports b'PATH' key and bytes value.
  subprocess.Popen() and os._execvpe() support bytes program name. Add
  os.supports_bytes_environ flag: True if the native OS type of the environment
  is bytes (eg. False on Windows).

- Issue #8633: tarfile is now able to read and write archives with "raw" binary
  pax headers as described in POSIX.1-2008.

- Issue #1285086: Speed up urllib.parse functions: quote, quote_from_bytes,
  unquote, unquote_to_bytes.

- Issue #8688: Distutils now recalculates MANIFEST everytime.

- Issue #8477: ssl.RAND_egd() and ssl._test_decode_cert() support str with
  surrogates and bytes for the filename.

- Issue #8550: Add first class ``SSLContext`` objects to the ssl module.

- Issue #8681: Make the zlib module's error messages more informative when the
  zlib itself doesn't give any detailed explanation.

- The audioop module now supports sound fragments of length greater than 2**31
  bytes on 64-bit machines, and is PY_SSIZE_T_CLEAN.

- Issue #4972: Add support for the context manager protocol to the ftplib.FTP
  class.

- Issue #8664: In py_compile, create __pycache__ when the compiled path is
  given.

- Issue #8514: Add os.fsencode() function (Unix only): encode a string to bytes
  for use in the file system, environment variables or the command line.

- Issue #8571: Fix an internal error when compressing or decompressing a chunk
  larger than 1GB with the zlib module's compressor and decompressor objects.

- Issue #8603: Support bytes environmental variables on Unix: Add os.environb
  mapping and os.getenvb() function. os.unsetenv() encodes str argument to the
  file system encoding with the surrogateescape error handler (instead of
  utf8/strict) and accepts bytes. posix.environ keys and values are now bytes.

- Issue #8573: asyncore _strerror() function might throw ValueError.

- Issue #8483: asyncore.dispatcher's __getattr__ method produced confusing error
  messages when accessing undefined class attributes because of the cheap
  inheritance with the underlying socket object.  The cheap inheritance has been
  deprecated.

- Issue #4265: shutil.copyfile() was leaking file descriptors when disk fills.
  Patch by Tres Seaver.

- Issue #8390: tarfile uses surrogateescape as the default error handler
  (instead of replace in read mode or strict in write mode).

- Issue #7755: Use an unencumbered audio file for tests.

- Issue #8621: uuid.uuid4() returned the same sequence of values in the parent
  and any children created using ``os.fork`` on MacOS X 10.6.

- Issue #8567: Fix precedence of signals in Decimal module: when a Decimal
  operation raises multiple signals and more than one of those signals is
  trapped, the specification determines the order in which the signals should be
  handled.  In many cases this order wasn't being followed, leading to the wrong
  Python exception being raised.

- Issue #7865: The close() method of :mod:`io` objects should not swallow
  exceptions raised by the implicit flush().  Also qensure that calling close()
  several times is supported.  Patch by Pascal Chambon.

- Issue #4687: Fix accuracy of garbage collection runtimes displayed with
  gc.DEBUG_STATS.

- Issue #8354: The siginterrupt setting is now preserved for all signals, not
  just SIGCHLD.

- Issue #7192: webbrowser.get("firefox") now works on Mac OS X, as does
  webbrowser.get("safari").

- Issue #8464: tarfile no longer creates files with execute permissions set when
  mode="w|" is used.

- Issue #7834: Fix connect() of Bluetooth L2CAP sockets with recent versions of
  the Linux kernel.  Patch by Yaniv Aknin.

- Issue #8295: Added shutil.unpack_archive.

- Issue #6312: Fixed http HEAD request when the transfer encoding is chunked.
  It should correctly return an empty response now.

- Issue #8546: Reject None given as the buffering argument to _pyio.open.

- Issue #8549: Fix compiling the _ssl extension under AIX.  Patch by
  Sridhar Ratnakumar.

- Issue #6656: fix locale.format_string to handle escaped percents
  and mappings.

- Issue #2302: Fix a race condition in SocketServer.BaseServer.shutdown, where
  the method could block indefinitely if called just before the event loop
  started running.  This also fixes the occasional freezes witnessed in
  test_httpservers.

- Issue #8524: When creating an SSL socket, the timeout value of the original
  socket wasn't retained (instead, a socket with a positive timeout would be
  turned into a non-blocking SSL socket).

- Issue #5103: SSL handshake would ignore the socket timeout and block
  indefinitely if the other end didn't respond.

- The do_handshake() method of SSL objects now adjusts the blocking mode of the
  SSL structure if necessary (as other methods already do).

- Issue #8391: os.execvpe() and os.getenv() supports unicode with surrogates and
  bytes strings for environment keys and values.

- Issue #8467: Pure Python implementation of subprocess encodes the error
  message using surrogatepass error handler to support surrogates in the
  message.

- Issue #8468: bz2.BZ2File() accepts str with surrogates and bytes filenames.

- Issue #8451: Syslog module now uses basename(sys.argv[0]) instead of the
  string "python" as the *ident*.  openlog() arguments are all optional and
  keywords.

- Issue #8108: Fix the unwrap() method of SSL objects when the socket has a
  non-infinite timeout.  Also make that method friendlier with applications
  wanting to continue using the socket in clear-text mode, by disabling
  OpenSSL's internal readahead.  Thanks to Darryl Miles for guidance.

- Issue #8496: make mailcap.lookup() always return a list, rather than an
  iterator.  Patch by Gregory Nofi.

- Issue #8195: Fix a crash in sqlite Connection.create_collation() if the
  collation name contains a surrogate character.

- Issue #8484: Load all ciphers and digest algorithms when initializing the _ssl
  extension, such that verification of some SSL certificates doesn't fail
  because of an "unknown algorithm".

- Issue #6547: Added the ignore_dangling_symlinks option to shutil.copytree.

- Issue #1540112: Now allowing the choice of a copy function in shutil.copytree.

- Issue #4814: timeout parameter is now applied also for connections resulting
  from PORT/EPRT commands.

- Issue #8463: added missing reference to bztar in shutil's documentation.

- Issue #7154: urllib.request can now detect the proxy settings on OSX 10.6 (as
  long as the user didn't specify 'automatic proxy configuration').

- Issue #3817: ftplib.FTP.abort() method now considers 225 a valid response code
  as stated in RFC-959 at chapter 5.4.

- Issue #8394: _ctypes.dlopen() accepts bytes, bytearray and str with
  surrogates.

- Issue #850728: Add a *timeout* parameter to the `acquire()` method of
  `threading.Semaphore` objects.  Original patch by Torsten Landschoff.

- Issue #8322: Add a *ciphers* argument to SSL sockets, so as to change the
  available cipher list.  Helps fix test_ssl with OpenSSL 1.0.0.

- Issue #8393: subprocess accepts bytes, bytearray and str with surrogates for
  the current working directory.

- Issue #7606: XML-RPC traceback stored in X-traceback is now encoded to ASCII
  using backslashreplace error handler.

- Issue #8412: os.system() now accepts bytes, bytearray and str with surrogates.

- Issue #2987: RFC2732 support for urlparse (IPv6 addresses). Patch by Tony
  Locke and Hans Ulrich Niedermann.

- Issue #5277: Fix quote counting when parsing RFC 2231 encoded parameters.

- Issue #7316: The acquire() method of lock objects in the :mod:`threading`
  module now takes an optional timeout argument in seconds.  Timeout support
  relies on the system threading library, so as to avoid a semi-busy wait loop.

- Issue #8383: pickle and pickletools use surrogatepass error handler when
  encoding unicode as utf8 to support lone surrogates and stay compatible with
  Python 2.x and 3.x.

- Issue #7585: difflib context and unified diffs now place a tab between
  filename and date, conforming to the 'standards' they were originally designed
  to follow.  This improves compatibility with patch tools.

- Issue #7472: Fixed typo in email.encoders module; messages using ISO-2022
  character sets will now consistently use a Content-Transfer-Encoding of 7bit
  rather than sometimes being marked as 8bit.

- Issue #8375: test_distutils now checks if the temporary directory are still
  present before it cleans them.

- Issue #8374: Update the internal alias table in the :mod:`locale` module to
  cover recent locale changes and additions.

- Issue #8321: Give access to OpenSSL version numbers from the `ssl` module,
  using the new attributes `ssl.OPENSSL_VERSION`, `ssl.OPENSSL_VERSION_INFO` and
  `ssl.OPENSSL_VERSION_NUMBER`.

- Add functools.total_ordering() and functools.cmp_to_key().

- Issue #8257: The Decimal construct now accepts a float instance directly,
  converting that float to a Decimal of equal value:

     >>> Decimal(1.1)
     Decimal('1.100000000000000088817841970012523233890533447265625')

- Issue #8294: The Fraction constructor now accepts Decimal and float instances
  directly.

- Issue #7279: Comparisons involving a Decimal signaling NaN now signal
  InvalidOperation instead of returning False.  (Comparisons involving a quiet
  NaN are unchanged.)  Also, Decimal quiet NaNs are now hashable; Decimal
  signaling NaNs remain unhashable.

- Issue #2531: Comparison operations between floats and Decimal instances now
  return a result based on the numeric values of the operands; previously they
  returned an arbitrary result based on the relative ordering of id(float) and
  id(Decimal).  See also issue #8188, which adds Decimal-to-Fraction
  comparisons.

- Added a subtract() method to collections.Counter().

- Issue #8233: When run as a script, py_compile.py optionally takes a single
  argument `-` which tells it to read files to compile from stdin.  Each line is
  read on demand and the named file is compiled immediately.  (Original patch by
  Piotr Ożarowski).

- Backwards incompatible change: Unicode codepoints line tabulation (0x0B) and
  form feed (0x0C) are now considered linebreaks, as specified in Unicode
  Standard Annex #14.  See issue #7643.  http://www.unicode.org/reports/tr14/

- Comparisons using one of <, <=, >, >= between a complex instance and a
  Fractions instance now raise TypeError instead of returning True/False.  This
  makes Fraction <=> complex comparisons consistent with int <=> complex, float
  <=> complex, and complex <=> complex comparisons.

- Issue #8139: ossaudiodev didn't initialize its types properly, therefore some
  methods (such as oss_mixer_device.fileno()) were not available.  Initial patch
  by Bertrand Janin.

- Issue #8205: Remove the "Modules" directory from sys.path when Python is
  running from the build directory (POSIX only).

- Issue #7512: shutil.copystat() could raise an OSError when the filesystem
  didn't support chflags() (for example ZFS under FreeBSD).  The error is now
  silenced.

- Issue #7860: platform.uname now reports the correct 'machine' type when Python
  is running in WOW64 mode on 64 bit Windows.

- Issue #3890, #8222: Fix recv() and recv_into() on non-blocking SSL sockets.
  Also, enable the SSL_MODE_AUTO_RETRY flag on SSL sockets, so that blocking
  reads and writes are always retried by OpenSSL itself.

- Issue #4282: Fix the main function of the profile module for a non-ASCII
  script, open the file in binary mode and not in text mode with the default
  (utf8) encoding.

- Issue #8179: Fix macpath.realpath() on a non-existing path.

- Issue #8024: Update the Unicode database to 5.2.

- Issue #8168: py_compile now handles files with utf-8 BOMS.

- ``tokenize.detect_encoding`` now returns ``'utf-8-sig'`` when a UTF-8 BOM is
  detected.

- Issue #6716/2: Backslash-replace error output in compilall.

- Issue #4961: Inconsistent/wrong result of askyesno function in tkMessageBox
  with Tcl/Tk-8.5.

- Issue #8140: extend compileall to compile single files.  Add -i option.

- Issue #7356: ctypes.util: Make parsing of ldconfig output independent of the
  locale.

- The internals of the subprocess module on POSIX systems have been replaced by
  an extension module (_posixsubprocess) so that the fork()+exec() can be done
  safely without the possibility of deadlock in multithreaded applications.

- subprocess.Popen now has restore_signals and start_new_session features.  The
  default of restore_signals=True is a new behavior compared to earlier Python
  versions.  This means that signals such as SIGPIPE are not ignored by default
  in subprocesses launched by Python (Issue #1652).

- Issue #6472: The xml.etree package is updated to ElementTree 1.3.  The
  cElementTree module is updated too.

- Issue #7774: Set sys.executable to an empty string if argv[0] has been set to
  an non existent program name and Python is unable to retrieve the real program
  name.

- Issue #7880: Fix sysconfig when the python executable is a symbolic link.

- Issue #6509: fix re.sub to work properly when the pattern, the string, and the
  replacement were all bytes.  Patch by Antoine Pitrou.

- The sqlite3 module was updated to pysqlite 2.6.0. This fixes several obscure
  bugs and allows loading SQLite extensions from shared libraries.

- Issue #1054943: Fix ``unicodedata.normalize('NFC', text)`` for the Public
  Review Issue #29 (http://unicode.org/review/pr-29.html).

- Issue #7494: fix a crash in _lsprof (cProfile) after clearing the profiler,
  reset also the pointer to the current pointer context.

- Issue #7232: Add support for the context manager protocol to the TarFile
  class.

- Issue #7250: Fix info leak of os.environ across multi-run uses of
  wsgiref.handlers.CGIHandler.

- Issue #1729305: Fix doctest to handle encode error with "backslashreplace".

- Issue #691291: codecs.open() should not convert end of lines on reading and
  writing.

- Issue #7869: logging: improved diagnostic for format-time errors.

- Issue #7868: logging: added loggerClass attribute to Manager.

- logging: Implemented PEP 391.

- Issue #1537721: Add a writeheader() method to csv.DictWriter.

- Issue #7959: ctypes callback functions are now registered correctly with the
  cycle garbage collector.

- Issue #5801: removed spurious empty lines in wsgiref.

- Issue #6666: fix bug in trace.py that applied the list of directories to be
  ignored only to the first file.  Noted by Bogdan Opanchuk.

- Issue #7597: curses.use_env() can now be called before initscr().  Noted by
  Kan-Ru Chen.

- Issue #7310: fix the __repr__ of os.environ to show the environment variables.

- Issue #7970: email.Generator.flatten now correctly flattens message/rfc822
  messages parsed by email.Parser.HeaderParser.

- Issue #7361: Importlib was not properly checking the number of bytes in
  bytecode file when it was less then 8 bytes.

- Issue #7633: In the decimal module, Context class methods (with the exception
  of canonical and is_canonical) now accept instances of int and long wherever a
  Decimal instance is accepted, and implicitly convert that argument to Decimal.
  Previously only some arguments were converted.

- Issue #7835: shelve should no longer produce mysterious warnings during
  interpreter shutdown.

- Issue #2746: Don't escape ampersands and angle brackets ("&", "<", ">") in XML
  processing instructions and comments.  These raw characters are allowed by the
  XML specification, and are necessary when outputting e.g.  PHP code in a
  processing instruction.  Patch by Neil Muller.

- Issue #6233: ElementTree failed converting unicode characters to XML entities
  when they could't be represented in the requested output encoding.  Patch by
  Jerry Chen.

- Issue #6003: add an argument to ``zipfile.Zipfile.writestr`` to specify the
  compression type.

- Issue #4772: Raise a ValueError when an unknown Bluetooth protocol is
  specified, rather than fall through to AF_PACKET (in the `socket` module).
  Also, raise ValueError rather than TypeError when an unknown TIPC address type
  is specified.  Patch by Brian Curtin.

- Issue #6939: Fix file I/O objects in the `io` module to keep the original file
  position when calling `truncate()`.  It would previously change the file
  position to the given argument, which goes against the tradition of
  ftruncate() and other truncation APIs.  Patch by Pascal Chambon.

- Issue #7610: Reworked implementation of the internal
  :class:`zipfile.ZipExtFile` class used to represent files stored inside an
  archive.  The new implementation is significantly faster and can be wrapped in
  a :class:`io.BufferedReader` object for more speedups.  It also solves an
  issue where interleaved calls to `read()` and `readline()` give wrong results.
  Patch by Nir Aides.

- Issue #6963: Added "maxtasksperchild" argument to multiprocessing.Pool,
  allowing for a maximum number of tasks within the pool to be completed by the
  worker before that worker is terminated, and a new one created to replace it.

- Issue #7792: Registering non-classes to ABCs raised an obscure error.

- Issue #7785: Don't accept bytes in FileIO.write().

- Removed the functions 'verify' and 'vereq' from Lib/test/support.py.

- Issue #7773: Fix an UnboundLocalError in platform.linux_distribution() when
  the release file is empty.

- Issue #7561: Fix crashes when using bytearray objects with the posix
  module.

- Issue #1670765: Prevent email.generator.Generator from re-wrapping headers in
  multipart/signed MIME parts, which fixes one of the sources of invalid
  modifications to such parts by Generator.

- Issue #7703: Add support for the new buffer API to `binascii.a2bhqx`.  Patch
  by Florent Xicluna, along with some additional tests.

- Issue #7701: Fix crash in binascii.b2a_uu() in debug mode when given a 1-byte
  argument.  Patch by Victor Stinner.

- Issue #3299: Fix possible crash in the _sre module when given bad argument
  values in debug mode.  Patch by Victor Stinner.

- Issue #2846: Add support for gzip.GzipFile reading zero-padded files.  Patch
  by Brian Curtin.

- Issue #7681: Use floor division in appropiate places in the wave module.

- Issue #5372: Drop the reuse of .o files in Distutils' ccompiler (since
  Extension extra options may change the output without changing the .c
  file). Initial patch by Collin Winter.

- Issue #7617: Make sure distutils.unixccompiler.UnixCCompiler recognizes gcc
  when it has a fully qualified configuration prefix. Initial patch by Arfrever.

- Issue #7105: Make WeakKeyDictionary and WeakValueDictionary robust against the
  destruction of weakref'ed objects while iterating.

- Issue #7455: Fix possible crash in cPickle on invalid input.  Patch by Victor
  Stinner.

- Issue #1628205: Socket file objects returned by socket.socket.makefile() now
  properly handles EINTR within the read, readline, write & flush methods.  The
  socket.sendall() method now properly handles interrupted system calls.

- Issue #7471: Improve the performance of GzipFile's buffering mechanism, and
  make it implement the `io.BufferedIOBase` ABC to allow for further speedups by
  wrapping it in an `io.BufferedReader`.  Patch by Nir Aides.

- Issue #3972: http.client.HTTPConnection now accepts an optional source_address
  parameter to allow specifying where your connections come from.

- socket.create_connection now accepts an optional source_address parameter.

- Issue #5511: now zipfile.ZipFile can be used as a context manager.  Initial
  patch by Brian Curtin.

- Issue #7556: Make sure Distutils' msvc9compile reads and writes the MSVC XML
  Manifest file in text mode so string patterns can be used in regular
  expressions.

- Issue #7552: Removed line feed in the base64 Authorization header in the
  Distutils upload command to avoid an error when PyPI reads it.  This occurs on
  long passwords. Initial patch by JP St. Pierre.

- Issue #7231: urllib2 cannot handle https with proxy requiring auth.  Patch by
  Tatsuhiro Tsujikawa.

- Issue #4757: `zlib.compress` and other methods in the zlib module now raise a
  TypeError when given an `str` object (rather than a `bytes`-like object).
  Patch by Victor Stinner and Florent Xicluna.

- Issue #7349: Make methods of file objects in the io module accept None as an
  argument where file-like objects (ie StringIO and BytesIO) accept them to mean
  the same as passing no argument.

- Issue #7357: tarfile no longer suppresses fatal extraction errors by default.

- Issue #5949: added check for correct lineends in input from IMAP server in
  imaplib.

- Add count() and reverse() methods to collections.deque().

- Fix variations of extending deques:  d.extend(d)  d.extendleft(d)  d+=d

- Issue #6986: Fix crash in the JSON C accelerator when called with the wrong
  parameter types.  Patch by Victor Stinner.

- Issue #7457: added a read_pkg_file method to
  distutils.dist.DistributionMetadata.

- logging: Added optional `secure` parameter to SMTPHandler, to enable use of
  TLS with authentication credentials.

- Issue #1923: Fixed the removal of meaningful spaces when PKG-INFO is generated
  in Distutils.  Patch by Stephen Emslie.

- Issue #4120: Drop reference to CRT from manifest when building extensions with
  msvc9compiler.

- Issue #7333: The `posix` module gains an `initgroups()` function providing
  access to the initgroups(3) C library call on Unix systems which implement it.
  Patch by Jean-Paul Calderone.

- Issue #7408: Fixed distutils.tests.sdist so it doesn't check for group
  ownership when the group is not forced, because the group may be different
  from the user's group and inherit from its container when the test is run.

- Issue #4486: When an exception has an explicit cause, do not print its
  implicit context too.  This affects the `traceback` module as well as built-in
  exception printing.

- Issue #1515: Enable use of deepcopy() with instance methods.  Patch by Robert
  Collins.

- Issue #7403: logging: Fixed possible race condition in lock creation.

- Issue #6845: Add restart support for binary upload in ftplib.  The
  `storbinary()` method of FTP and FTP_TLS objects gains an optional `rest`
  argument.  Patch by Pablo Mouzo.

- Issue #5788: `datetime.timedelta` objects get a new `total_seconds()` method
  returning the total number of seconds in the duration.  Patch by Brian
  Quinlan.

- Issue #7133: SSL objects now support the new buffer API.

- Issue #1488943: difflib.Differ() doesn't always add hints for tab characters.

- Issue #6123: tarfile now opens empty archives correctly and consistently
  raises ReadError on empty files.

- Issue #7354: distutils.tests.test_msvc9compiler - dragfullwindows can be 2.

- Issue #5037: Proxy the __bytes__ special method instead to __bytes__ instead
  of __str__.

- Issue #7341: Close the internal file object in the TarFile constructor in case
  of an error.

- Issue #7293: distutils.test_msvc9compiler is fixed to work on any fresh
  Windows box. Help provided by David Bolen.

- Issue #2054: ftplib now provides an FTP_TLS class to do secure FTP using TLS
  or SSL.  Patch by Giampaolo Rodola'.

- Issue #7328: pydoc no longer corrupts sys.path when run with the '-m' switch.

- Issue #4969: The mimetypes module now reads the MIME database from the
  registry under Windows.  Patch by Gabriel Genellina.

- Issue #6816: runpy now provides a run_path function that allows Python code to
  execute file paths that refer to source or compiled Python files as well as
  zipfiles, directories and other valid sys.path entries that contain a
  __main__.py file.  This allows applications that run other Python scripts to
  support the same flexibility as the CPython command line itself.

- Issue #7318: multiprocessing now uses a timeout when it fails to establish a
  connection with another process, rather than looping endlessly.  The default
  timeout is 20 seconds, which should be amply sufficient for local connections.

- Issue #7197: Allow unittest.TextTestRunner objects to be pickled and
  unpickled.  This fixes crashes under Windows when trying to run
  test_multiprocessing in verbose mode.

- Issue #7893: ``unittest.TextTestResult`` is made public and a ``resultclass``
  argument added to the TextTestRunner constructor allowing a different result
  class to be used without having to subclass.

- Issue #7588: ``unittest.TextTestResult.getDescription`` now includes the test
  name in failure reports even if the test has a docstring.

- Issue #3001: Add a C implementation of recursive locks which is used by
  default when instantiating a `threading.RLock` object. This makes recursive
  locks as fast as regular non-recursive locks (previously, they were slower by
  10x to 15x).

- Issue #7282: Fix a memory leak when an RLock was used in a thread other than
  those started through `threading.Thread` (for example, using
  `_thread.start_new_thread()`).

- Issue #7187: Importlib would not silence the IOError raised when trying to
  write new bytecode when it was made read-only.

- Issue #7264: Fix a possible deadlock when deallocating thread-local objects
  which are part of a reference cycle.

- Issue #7211: Allow 64-bit values for the `ident` and `data` fields of kevent
  objects on 64-bit systems.  Patch by Michael Broghton.

- Issue #6896: mailbox.Maildir now invalidates its internal cache each time a
  modification is done through it.  This fixes inconsistencies and test failures
  on systems with slightly bogus mtime behaviour.

- Issue #7246 & Issue #7208: getpass now properly flushes input before reading
  from stdin so that existing input does not confuse it and lead to incorrect
  entry or an IOError.  It also properly flushes it afterwards to avoid the
  terminal echoing the input afterwards on OSes such as Solaris.

- Issue #7233: Fix a number of two-argument Decimal methods to make sure that
  they accept an int or long as the second argument.  Also fix buggy handling of
  large arguments (those with coefficient longer than the current precision) in
  shift and rotate.

- Issue #4750: Store the basename of the original filename in the gzip FNAME
  header as required by RFC 1952.

- Issue #1180: Added a new global option to ignore ~/.pydistutils.cfg in
  Distutils.

- Issue #7218: Fix test_site for win32, the directory comparison was done with
  an uppercase.

- Issue #7205: Fix a possible deadlock when using a BZ2File object from
  several threads at once.

- Issue #7077: logging: SysLogHandler now treats Unicode as per RFC 5424.

- Issue #7099: Decimal.is_normal now returns True for numbers with exponent
  larger than emax.

- Issue #7080: locale.strxfrm() raises a MemoryError on 64-bit non-Windows
  platforms, and assorted locale fixes by Derk Drukker.

- Issue #5833: Fix extra space character in readline completion with the GNU
  readline library version 6.0.

- Issue #6894: Fixed the issue urllib2 doesn't respect "no_proxy" environment.

- Issue #7086: Added TCP support to SysLogHandler, and tidied up some
  anachronisms in the code which were a relic of 1.5.2 compatibility.

- Issue #7082: When falling back to the MIME 'name' parameter, the correct place
  to look for it is the Content-Type header.

- Make tokenize.detect_coding() normalize utf-8 and iso-8859-1 variants like the
  builtin tokenizer.

- Issue #7048: Force Decimal.logb to round its result when that result is too
  large to fit in the current precision.

- Issue #6236, #6348: Fix various failures in the I/O library under AIX and
  other platforms, when using a non-gcc compiler. Patch by Derk Drukker.

- Issue #4606: Passing 'None' if ctypes argtype is set to POINTER(...)  does now
  always result in NULL.

- Issue #5042: Structure sub-subclass does now initialize correctly with base
  class positional arguments.

- Issue #6882: Import uuid creates zombies processes.

- Issue #6635: Fix profiler printing usage message.

- Issue #6856: Add a filter keyword argument to TarFile.add().

- Issue #6888: pdb's alias command was broken when no arguments were given.

- Issue #6857: Default format() alignment should be '>' for Decimal instances.

- Issue #6795: int(Decimal('nan')) now raises ValueError instead of returning
  NaN or raising InvalidContext.  Also, fix infinite recursion in
  long(Decimal('nan')).

- Issue #6850: Fix bug in Decimal._parse_format_specifier for formats with no
  type specifier.

- Issue #6239: ctypes.c_char_p return value must return bytes.

- Issue #6838: Use a list to accumulate the value instead of repeatedly
  concatenating strings in http.client's HTTPResponse._read_chunked providing a
  significant speed increase when downloading large files servend with a
  Transfer-Encoding of 'chunked'.

- Trying to import a submodule from a module that is not a package, ImportError
  should be raised, not AttributeError.

- When the globals past to importlib.__import__() has __package__ set to None,
  fall back to computing what __package__ should be instead of giving up.

- Raise a TypeError when the name of a module to be imported for
  importlib.__import__ is not a string (was raising an AttributeError before).

- Allow the fromlist passed into importlib.__import__ to be any iterable.

- Have importlib raise ImportError if None is found in sys.modules.

- Issue #6054: Do not normalize stored pathnames in tarfile.

- Issue #6794: Fix Decimal.compare_total and Decimal.compare_total_mag: NaN
  payloads are now ordered by integer value rather than lexicographically.

- Issue #1356969: Add missing info methods in tix.HList.

- Issue #1522587: New constants and methods for the tix.Grid widget.

- Issue #1250469: Fix the return value of tix.PanedWindow.panes.

- Issue #1119673: Do not override tkinter.Text methods when creating a
  ScrolledText.

- Issue #6665: Fix fnmatch to properly match filenames with newlines in them.

- Issue #1135: Add the XView and YView mix-ins to avoid duplicating the xview*
  and yview* methods.

- Issue #6629: Fix a data corruption issue in the new I/O library, which could
  occur when writing to a BufferedRandom object (e.g. a file opened in "rb+" or
  "wb+" mode) after having buffered a certain amount of data for reading. This
  bug was not present in the pure Python implementation.

- Issue #6622: Fix "local variable 'secret' referenced before assignment" bug in
  POP3.apop.

- Issue #2715: Remove remnants of Carbon.File from binhex module.

- Issue #6595: The Decimal constructor now allows arbitrary Unicode decimal
  digits in input, as recommended by the standard.  Previously it was restricted
  to accepting [0-9].

- Issue #6106: telnetlib.Telnet.process_rawq doesn't handle default WILL/WONT
  DO/DONT correctly.

- Issue #1424152: Fix for http.client, urllib.request to support SSL while
  working through proxy.  Original patch by Christopher Li, changes made by
  Senthil Kumaran.

- Add importlib.abc.ExecutionLoader to represent the PEP 302 protocol for
  loaders that allow for modules to be executed. Both importlib.abc.PyLoader and
  PyPycLoader inherit from this class and provide implementations in relation to
  other methods required by the ABCs.

- importlib.abc.PyLoader did not inherit from importlib.abc.ResourceLoader like
  the documentation said it did even though the code in PyLoader relied on the
  abstract method required by ResourceLoader.

- Issue #6431: Make Fraction type return NotImplemented when it doesn't know how
  to handle a comparison without loss of precision.  Also add correct handling
  of infinities and nans for comparisons with float.

- Issue #6415: Fixed warnings.warn segfault on bad formatted string.

- Issue #6358: The exit status of a command started with os.popen() was reported
  differently than it did with python 2.x.

- Issue #6323: The pdb debugger did not exit when running a script with a syntax
  error.

- Issue #3392: The subprocess communicate() method no longer fails in select()
  when file descriptors are large; communicate() now uses poll() when possible.

- Issue #6369: Fix an RLE decompression bug in the binhex module.

- Issue #6344: Fixed a crash of mmap.read() when passed a negative argument.

- The deprecated function string.maketrans has been removed.

- Issue #4005: Fixed a crash of pydoc when there was a zip file present in
  sys.path.

- Issue #6218: io.StringIO and io.BytesIO instances are now picklable.

- The os.get_exec_path() function to return the list of directories that will be
  searched for an executable when launching a subprocess was added.

- Issue #7481: When a threading.Thread failed to start it would leave the
  instance stuck in initial state and present in threading.enumerate().

- Issue #1068268: The subprocess module now handles EINTR in internal os.waitpid
  and os.read system calls where appropriate.

- Issue #6729: Added ctypes.c_ssize_t to represent ssize_t.

- Issue #6247: The argparse module has been added to the standard library.

- Issue #8235: _socket: Add the constant ``SO_SETFIB``.  SO_SETFIB is a socket
  option available on FreeBSD 7.1 and newer.

- Issue #9315: Fix for the trace module to record correct class name
  for tracing methods.

Extension Modules
-----------------

- Issue #9422: Fix memory leak when re-initializing a struct.Struct object.

- Issue #7900: The getgroups(2) system call on MacOSX behaves rather oddly
  compared to other unix systems. In particular, os.getgroups() does not reflect
  any changes made using os.setgroups() but basicly always returns the same
  information as the id command. os.getgroups() can now return more than 16
  groups on MacOSX.

- Issue #6095: Make directory argument to os.listdir optional.

- Issue #9277: Fix bug in struct.pack for bools in standard mode (e.g.,
  struct.pack('>?')): if conversion to bool raised an exception then that
  exception wasn't properly propagated on machines where char is unsigned.

- Issue #5180: Fixed a bug that prevented loading 2.x pickles in 3.x python when
  they contain instances of old-style classes.

- Issue #9165: Add new functions math.isfinite and cmath.isfinite, to accompany
  existing isinf and isnan functions.

- Issue #1578269: Implement os.symlink for Windows 6.0+.  Patch by Jason
  R. Coombs.

- In struct.pack, correctly propogate exceptions from computing the truth of an
  object in the '?' format.

- Issue #9000: datetime.timezone objects now have eval-friendly repr.

- In the math module, correctly lookup __trunc__, __ceil__, and __floor__ as
  special methods.

- Issue #9005: Prevent utctimetuple() from producing year 0 or year 10,000.
  Prior to this change, timezone adjustment in utctimetuple() could produce
  tm_year value of 0 or 10,000.  Now an OverflowError is raised in these edge
  cases.

- Issue #6641: The ``datetime.strptime`` method now supports the ``%z``
  directive.  When the ``%z`` directive is present in the format string, an
  aware ``datetime`` object is returned with ``tzinfo`` bound to a
  ``datetime.timezone`` instance constructed from the parsed offset.  If both
  ``%z`` and ``%Z`` are present, the data in ``%Z`` field is used for timezone
  name, but ``%Z`` data without ``%z`` is discarded.

- Issue #5094: The ``datetime`` module now has a simple concrete class
  implementing ``datetime.tzinfo`` interface.  Instances of the new class,
  ``datetime.timezone``, return fixed name and UTC offset from their
  ``tzname(dt)`` and ``utcoffset(dt)`` methods.  The ``dst(dt)`` method always
  returns ``None``.  A class attribute, ``utc`` contains an instance
  representing the UTC timezone.  Original patch by Rafe Kaplan.

- Issue #8973: Add __all__ to struct module; this ensures that help(struct)
  includes documentation for the struct.Struct class.

- Issue #3129: Trailing digits in struct format string are no longer ignored.
  For example, "1" or "ilib123" are now invalid formats and cause
  ``struct.error`` to be raised.  Patch by Caleb Deveraux.

- Issue #7384: If the system readline library is linked against ncurses, the
  curses module must be linked against ncurses as well. Otherwise it is not safe
  to load both the readline and curses modules in an application.

- Issue #2810: Fix cases where the Windows registry API returns ERROR_MORE_DATA,
  requiring a re-try in order to get the complete result.

- Issue #8692: Optimize math.factorial: replace the previous naive algorithm
  with an improved 'binary-split' algorithm that uses fewer multiplications and
  allows many of the multiplications to be performed using plain C integer
  arithmetic instead of PyLong arithmetic.  Also uses a lookup table for small
  arguments.

- Issue #8674: Fixed a number of incorrect or undefined-behaviour-inducing
  overflow checks in the audioop module.

- Issue #8644: The accuracy of td.total_seconds() has been improved (by
  calculating with integer arithmetic instead of float arithmetic internally):
  the result is now always correctly rounded, and is equivalent to ``td /
  timedelta(seconds=1)``.

- Issue #2706: Allow division of a timedelta by another timedelta: timedelta /
  timedelta, timedelta % timedelta, timedelta // timedelta and divmod(timedelta,
  timedelta) are all supported.

- Issue #8314: Fix unsigned long long bug in libffi on Sparc v8.

- Issue #8300: When passing a non-integer argument to struct.pack with any
  integer format code, struct.pack first attempts to convert the non-integer
  using its __index__ method.  If that method is non-existent or raises
  TypeError it goes on to try the __int__ method, as described below.

- Issue #8142: Update libffi to the 3.0.9 release.

- Issue #6949: Allow the _dbm extension to be built with db 4.8.x.

- Issue #6544: Fix a reference leak in the kqueue implementation's error
  handling.

- Stop providing crtassem.h symbols when compiling with Visual Studio 2010, as
  msvcr100.dll is not a platform assembly anymore.

- Issue #6508: Add posix.{getresuid,getresgid,setresuid,setresgid}.

- Issue #7078: Set struct.__doc__ from _struct.__doc__.

- Issue #3366: Add erf, erfc, expm1, gamma, lgamma functions to math module.

- Issue #6877: It is now possible to link the readline extension to the libedit
  readline emulation on OSX 10.5 or later.

- Issue #6848: Fix curses module build failure on OS X 10.6.

- Fix a segfault that could be triggered by expat with specially formed input.

- Issue #6561: '\d' in a regex now matches only characters with Unicode category
  'Nd' (Number, Decimal Digit).  Previously it also matched characters with
  category 'No'.

- Issue #4509: Array objects are no longer modified after an operation failing
  due to the resize restriction in-place when the object has exported buffers.

- Issue #2389: Array objects are now pickled in a portable manner.

- Expat: Fix DoS via XML document with malformed UTF-8 sequences
  (CVE_2009_3560).

- Issue #7242: On Solaris 9 and earlier calling os.fork() from within a thread
  could raise an incorrect RuntimeError about not holding the import lock.  The
  import lock is now reinitialized after fork.

- Issue #7999: os.setreuid() and os.setregid() would refuse to accept a -1
  parameter on some platforms such as OS X.

- Build the ossaudio extension on GNU/kFreeBSD.

- Issue #7347: winreg: Add CreateKeyEx and DeleteKeyEx, as well as fix a bug in
  the return value of QueryReflectionKey.

- Issue #7567: PyCurses_setupterm: Don't call ``setupterm`` twice.

Build
-----

- Use OpenSSL 1.0.0a on Windows.

- Issue #9280: Make sharedinstall depend on sharedmods.

- Issue #9189: Make a user-specified CFLAGS, CPPFLAGS, or LDFLAGS setting
  override the configure and makefile defaults, without deleting options the
  user didn't intend to override.  Developers should no longer need to specify
  OPT or EXTRA_CFLAGS, although those variables are still present for
  backward-compatibility.

- Issue #8854: Fix finding Visual Studio 2008 on Windows x64.

- Issue #1759169, #8864: Drop _XOPEN_SOURCE on Solaris, define it for
  multiprocessing only.

- Issue #8625: Turn off optimization in --with-pydebug builds with gcc.
  (Optimization was unintentionally turned on in gcc --with-pydebug builds as a
  result of the issue #1628484 fix, combined with autoconf's strange choice of
  default CFLAGS produced by AC_PROG_CC for gcc.)

- Issue #3646: It is now easily possible to install a Python framework into your
  home directory on MacOSX, see Mac/README for more information.

- Issue #3928: os.mknod() now available in Solaris, also.

- Issue #3326: Build Python without -fno-strict-aliasing when the gcc does not
  give false warnings.

- Issue #1628484: The Makefile doesn't ignore the CFLAGS environment variable
  anymore.  It also forwards the LDFLAGS settings to the linker when building a
  shared library.

- Issue #6716: Quote -x arguments of compileall in MSI installer.  Exclude 2to3
  tests from compileall.

- Issue #3920, #7903: Define _BSD_SOURCE on OpenBSD 4.4 through 4.9.

- Issue #7632: When Py_USING_MEMORY_DEBUGGER is defined, disable the private
  memory allocation scheme in dtoa.c and use PyMem_Malloc and PyMem_Free
  instead.  Also disable caching of powers of 5.

- Issue #6491: Allow --with-dbmliborder to specify that no dbms will be built.

- Issue #6943: Use pkg-config to find the libffi headers when the
  --with-system-ffi flag is used.

- Issue #7609: Add a --with-system-expat option that causes the system's expat
  library to be used for the pyexpat module instead of the one included with
  Python.

- Issue #7589: Only build the nis module when the correct header files are
  found.

- Switch to OpenSSL 0.9.8l and sqlite 3.6.21 on Windows.

- Issue #5792: Extend the short float repr support to x86 systems using
  icc or suncc.

- Issue #6603: Change READ_TIMESTAMP macro in ceval.c so that it compiles
  correctly under gcc on x86-64.  This fixes a reported problem with the
  --with-tsc build on x86-64.

- Issue #6802: Fix build issues on MacOSX 10.6.

- Issue #6244: Allow detect_tkinter to look for Tcl/Tk 8.6.

- Issue #4601: 'make install' did not set the appropriate permissions on
  directories.

- Issue #5390: Add uninstall icon independent of whether file extensions are
  installed.

- Issue #7541: When using ``python-config`` with a framework install the
  compiler might use the wrong library.

- python-config now supports multiple options on the same command line.

- Issue #8509: Fix quoting in help strings and code snippets in configure.in.

- Issue #8510: Update to autoconf2.65.

Documentation
-------------

- Issue #9817: Add expat COPYING file; add expat, libffi and expat licenses
  to Doc/license.rst.

- Issue #9524: Document that two CTRL* signals are meant for use only
  with os.kill.

- Issue #9255: Document that the 'test' package is meant for internal Python use
  only.

- A small WSGI server was added as Tools/scripts/serve.py, and is used to
  implement a local documentation server via 'make serve' in the doc directory.

- Updating `Using Python` documentation to include description of CPython's -J
  and -X options.

- Document that importing a module that has None in sys.modules triggers an
  ImportError.

- Issue #6556: Fixed the Distutils configuration files location explanation for
  Windows.

- Update python manual page (options -B, -O0, -s, environment variables
  PYTHONDONTWRITEBYTECODE, PYTHONNOUSERSITE).

- Issue #8909: Added the size of the bitmap used in the installer created by
  distutils' bdist_wininst. Patch by Anatoly Techtonik.

Tests
-----

- Issue #9251: test_threaded_import didn't fail when run through regrtest if the
  import lock was disabled.

- Issue #8605: Skip test_gdb if Python is compiled with optimizations.

- Issue #7449: Skip test_socketserver if threading support is disabled.

- Issue #8672: Add a zlib test ensuring that an incomplete stream can be handled
  by a decompressor object without errors (it returns incomplete uncompressed
  data).

- Issue #8533: regrtest uses backslashreplace error handler for stdout to avoid
  UnicodeEncodeError (write non-ASCII character to stdout using ASCII encoding).

- Issue #8576: Remove use of find_unused_port() in test_smtplib and
  test_multiprocessing.  Patch by Paul Moore.

- Issue #7449: Fix many tests to support Python compiled without thread
  support. Patches written by Jerry Seutter.

- Issue #8108: test_ftplib's non-blocking SSL server now has proper handling of
  SSL shutdowns.

- Issues #8279, #8330, #8437, #8480, #8495: Fix test_gdb failures, patch written
  by Dave Malcolm.

- Issue #3864: Skip three test_signal tests on freebsd6 because they fail if any
  thread was previously started, most likely due to a platform bug.

- Issue #8193: Fix test_zlib failure with zlib 1.2.4.

- Issue #8248: Add some tests for the bool type.  Patch by Gregory Nofi.

- Issue #8263: Now regrtest.py will report a failure if it receives a
  KeyboardInterrupt (SIGINT).

- Issue #8180 and #8207: Fix test_pep277 on OS X and add more tests for special
  Unicode normalization cases.

- Issue #7783: test.support.open_urlresource invalidates the outdated files from
  the local cache.

- Issue #7849: Now the utility ``check_warnings`` verifies if the warnings are
  effectively raised.

- The four path modules (genericpath, macpath, ntpath, posixpath) share a common
  TestCase for some tests: test_genericpath.CommonTest.

- Print platform information when running the whole test suite, or using the
  --verbose flag.

- Issue #767675: enable test_pep277 on POSIX platforms with Unicode-friendly
  filesystem encoding.

- Issue #6292: for the moment at least, the test suite runs cleanly if python is
  run with the -OO flag.  Tests requiring docstrings are skipped.

- Issue #7712: test.support gained a new `temp_cwd` context manager which is now
  also used by regrtest to run all the tests in a temporary directory.  The
  original CWD is saved in `support.SAVEDCWD`.  Thanks to Florent Xicluna who
  helped with the patch.

- Issue #7924: Fix an intermittent 'XXX undetected error' failure in test_capi
  (only seen so far on platforms where the curses module wasn't built), due to
  an uncleared exception.

- Issue #7728: test_timeout was changed to use support.bind_port instead of a
  hard coded port.

- Issue #7376: Instead of running a self-test (which was failing) when called
  with no arguments, doctest.py now gives a usage message.

- Issue #7396: fix regrtest -s, which was broken by the -j enhancement.

- Issue #7498: test_multiprocessing now uses test.support.find_unused_port
  instead of a hardcoded port number in test_rapid_restart.

- Issue #7431: Use TESTFN in test_linecache instead of trying to create a file
  in the Lib/test directory, which might be read-only for the user running the
  tests.

- Issue #7324: Add a sanity check to regrtest argument parsing to catch the case
  of an option with no handler.

- Issue #7312: Add a -F flag to run the selected tests in a loop until a test
  fails.  Can be combined with -j.

- Issue #6551: test_zipimport could import and then destroy some modules of the
  encodings package, which would make other tests fail further down the road
  because the internally cached encoders and decoders would point to empty
  global variables.

- Issue #7295: Do not use a hardcoded file name in test_tarfile.

- Issue #7270: Add some dedicated unit tests for multi-thread synchronization
  primitives such as Lock, RLock, Condition, Event and Semaphore.

- Issue #7248 (part 2): Use a unique temporary directory for importlib source
  tests instead of tempfile.tempdir. This prevents the tests from sharing state
  between concurrent executions on the same system.

- Issue #7248: In importlib.test.source.util a try/finally block did not make
  sure that some referenced objects actually were created in the block before
  calling methods on the object.

- Issue #7222: Make thread "reaping" more reliable so that reference
  leak-chasing test runs give sensible results. The previous method of reaping
  threads could return successfully while some Thread objects were still
  referenced. This also introduces a new private function:
  :func:`_thread._count()`.

- Issue #7151: Fixed regrtest -j so that output to stderr from a test no longer
  runs the risk of causing the worker thread to fail.

- Issue #7055: test___all__ now greedily detects all modules which have an
  __all__ attribute, rather than using a hardcoded and incomplete list.

- Issue #7058: Added save/restore for things like sys.argv and cwd to
  runtest_inner in regrtest, with warnings if the called test modifies them, and
  a new section in the summary report at the end.

- Issue #7042: Fix test_signal (test_itimer_virtual) failure on OS X 10.6.

- Fixed tests in importlib.test.source.test_abc_loader that were masking the
  proper exceptions that should be raised for missing or improper code object
  bytecode.

- Removed importlib's custom test discovery code and switched to
  unittest.TestLoader.discover().

Tools/Demos
-----------

- Issue #5464, #8974: Implement plural forms in msgfmt.py.

- iobench (a file I/O benchmark) and ccbench (a concurrency benchmark) were
  added to the `Tools/` directory.  They were previously living in the sandbox.


What's New in Python 3.1?
=========================

*Release date: 27-June-2009*

Core and Builtins
-----------------

- Issue #6334: Fix bug in range length calculation for ranges with
  large arguments.

- Issue #6329: Fixed iteration for memoryview objects (it was being blocked
  because it wasn't recognized as a sequence).

Library
-------

- Issue #6126: Fixed pdb command-line usage.

- Issue #6314: logging: performs extra checks on the "level" argument.

- Issue #6274: Fixed possible file descriptors leak in subprocess.py

- Accessing io.StringIO.buffer now raises an AttributeError instead of
  io.UnsupportedOperation.

- Issue #6271: mmap tried to close invalid file handle (-1) when anonymous.
  (On Unix)

- Issue #1202: zipfile module would cause a struct.error when attempting to
  store files with a CRC32 > 2**31-1.

Extension Modules
-----------------

- Issue #5590: Remove unused global variable in pyexpat extension.


What's New in Python 3.1 Release Candidate 2?
=============================================

*Release date: 13-June-2009*

Core and Builtins
-----------------

- Fixed SystemError triggered by "range([], 1, -1)".

- Issue #5924: On Windows, a large PYTHONPATH environment variable
  (more than 255 characters) would be completely ignored.

- Issue #4547: When debugging a very large function, it was not always
  possible to update the lineno attribute of the current frame.

- Issue #5330: C functions called with keyword arguments were not reported by
  the various profiling modules (profile, cProfile). Patch by Hagen Fürstenau.

Library
-------

- Issue #6438: Fixed distutils.cygwinccompiler.get_versions : the regular
  expression string pattern was trying to match against a bytes returned by
  Popen. Tested under win32 to build the py-postgresql project.

- Issue #6258: Support AMD64 in bdist_msi.

- Issue #6195: fixed doctest to no longer try to read 'source' data from
  binary files.

- Issue #5262: Fixed bug in next rollover time computation in
  TimedRotatingFileHandler.

- Issue #6217: The C implementation of io.TextIOWrapper didn't include the
  errors property.  Additionally, the errors and encoding properties of StringIO
  are always None now.

- Issue #6137: The pickle module now translates module names when loading
  or dumping pickles with a 2.x-compatible protocol, in order to make data
  sharing and migration easier. This behaviour can be disabled using the
  new `fix_imports` optional argument.

- Removed the ipaddr module.

- Issue #3613: base64.{encode,decode}string are now called
  base64.{encode,decode}bytes which reflects what type they accept and return.
  The old names are still there as deprecated aliases.

- Issue #5767: Remove sgmlop support from xmlrpc.client.

- Issue #6150: Fix test_unicode on wide-unicode builds.

- Issue #6149: Fix initialization of WeakValueDictionary objects from non-empty
  parameters.

Windows
-------

- Issue #6221: Delete test registry key before running the test.

- Issue #6158: Package Sine-1000Hz-300ms.aif in MSI file.

C-API
-----

- Issue #5735: Python compiled with --with-pydebug should throw an
  ImportError when trying to import modules compiled without
  --with-pydebug, and vice-versa.


Build
-----

- Issue #6154: Make sure the intl library is added to LIBS if needed. Also
  added LIBS to OS X framework builds.

- Issue #5809: Specifying both --enable-framework and --enable-shared is
  an error. Configure now explicity tells you about this.



What's New in Python 3.1 release candidate 1?
=============================================

*Release date: 2009-05-30*

Core and Builtins
-----------------

- Issue #6097: Escape UTF-8 surrogates resulting from mbstocs conversion
  of the command line.

- Issue #6012: Add cleanup support to O& argument parsing.

- Issue #6089: Fixed str.format with certain invalid field specifiers
  that would raise SystemError.

- Issue #5982: staticmethod and classmethod now expose the wrapped
  function with __func__.

- Added support for multiple context managers in the same with-statement.
  Deprecated contextlib.nested() which is no longer needed.

- Issue #5829: complex("1e500") no longer raises OverflowError.  This
  makes it consistent with float("1e500") and interpretation of real
  and imaginary literals.

- Issue #3527: Removed Py_WIN_WIDE_FILENAMES which is not used any more.

- Issue #5994: the marshal module now has docstrings.

- Issue #5981: Fix three minor inf/nan issues in float.fromhex:
  (1) inf and nan strings with trailing whitespace were incorrectly
  rejected;  (2) parsing of strings representing infinities and nans
  was locale aware; and (3) the interpretation of fromhex('-nan')
  didn't match that of float('-nan').

Library
-------

- Issue #4859: Implement PEP 383 for pwd, spwd, and grp.

- smtplib 'login' and 'cram-md5' login are also fixed (see Issue #5259).

- Issue #6121: pydoc now ignores leading and trailing spaces in the
  argument to the 'help' function.

- Issue #6118: urllib.parse.quote_plus ignored the encoding and errors
  arguments for strings with a space in them.

- collections.namedtuple() was not working with the following field
  names:  cls, self, tuple, itemgetter, and property.

- In unittest, using a skipping decorator on a class is now equivalent to
  skipping every test on the class.  The ClassTestSuite class has been removed.

- Issue #6050: Don't fail extracting a directory from a zipfile if
  the directory already exists.

- Issue #1309352: fcntl now converts its third arguments to a C `long` rather
  than an int, which makes some operations possible under 64-bit Linux (e.g.
  DN_MULTISHOT with F_NOTIFY).

- Issue #5761: Add the name of the underlying file to the repr() of various
  IO objects.

- Issue #5259: smtplib plain auth login no longer gives a traceback.  Fix
  by Musashi Tamura, tests by Marcin Bachry.

- Issue #1983: Fix functions taking or returning a process identifier to use
  the dedicated C type ``pid_t`` instead of a C ``int``. Some platforms have
  a process identifier type wider than the standard C integer type.

- Issue #4066: smtplib.SMTP_SSL._get_socket now correctly returns the socket.
  Patch by Farhan Ahmad, test by Marcin Bachry.

- Issue #2116: Weak references and weak dictionaries now support copy()ing and
  deepcopy()ing.

- Issue #1655: Make imaplib IPv6-capable. Patch by Derek Morr.

- Issue #5918: Fix a crash in the parser module.

- Issue #1664: Make nntplib IPv6-capable. Patch by Derek Morr.

- Issue #5006: Better handling of unicode byte-order marks (BOM) in the io
  library. This means, for example, that opening an UTF-16 text file in
  append mode doesn't add a BOM at the end of the file if the file isn't
  empty.

- Issue #4050: inspect.findsource/getsource now raise an IOError if the 'source'
  file is a binary.  Patch by Brodie Rao, tests by Daniel Diniz.  This fix
  corrects a pydoc regression.

- Issue #5955: aifc's close method did not close the file it wrapped,
  now it does.  This also means getfp method now returns the real fp.

Installation
------------

- Issue #6047: fullinstall has been removed because Python 3's executable will
  now be known as python3.

- Lib/smtpd.py is no longer installed as a script.

Extension Modules
-----------------

- Issue #3061: Use wcsftime for time.strftime where available.

- Issue #4873: Fix resource leaks in error cases of pwd and grp.

- Issue #6093: Fix off-by-one error in locale.strxfrm.

- The _functools and _locale modules are now built into the libpython shared
  library instead of as extension modules.

Build
-----

- Issue #3585: Add pkg-config support. It creates a python-2.7.pc file
  and a python3.pc symlink in the $(LIBDIR)/pkgconfig directory. Patch by
  Clinton Roy.

Tests
-----

- Issue #5442: Tests for importlib were not properly skipping case-sensitivity
  tests on darwin even when the OS was installed on a case-sensitive
  filesystem. Also fixed tests that should not be run when
  sys.dont_write_bytecode is true.


What's New in Python 3.1 beta 1?
================================

*Release date: 2009-05-06*

Core and Builtins
-----------------

- Issue #5914: Add new C API function PyOS_string_to_double, and
  deprecate PyOS_ascii_strtod and PyOS_ascii_atof.

- Issue #3382: float.__format__, complex.__format__, and %-formatting
  no longer map 'F' to 'f'. Because of issue #5859 (below), this only
  affects nan -> NAN and inf -> INF.

- Issue #5799: ntpath (ie, os.path on Windows) fully supports UNC pathnames
  in all operations, including splitdrive, split, etc.  splitunc() now issues
  a PendingDeprecation warning.

- Issue #5920: For float.__format__, change the behavior with the
  empty presentation type (that is, not one of 'e', 'f', 'g', or 'n')
  to be like 'g' but with at least one decimal point and with a
  default precision of 12. Previously, the behavior the same but with
  a default precision of 6.  This more closely matches str(), and
  reduces surprises when adding alignment flags to the empty
  presentation type. This also affects the new complex.__format__ in
  the same way.

- Implement PEP 383, Non-decodable Bytes in System Character Interfaces.

- Issue #5890: in subclasses of 'property' the __doc__ attribute was
  shadowed by classtype's, even if it was None.  property now
  inserts the __doc__ into the subclass instance __dict__.

- Issue #4426: The UTF-7 decoder was too strict and didn't accept some legal
  sequences. Patch by Nick Barnes and Victor Stinner.

- Issue #3672: Reject surrogates in utf-8 codec; add surrogatepass error handler.

- Issue #5883: In the io module, the BufferedIOBase and TextIOBase ABCs have
  received a new method, detach().  detach() disconnects the underlying stream
  from the buffer or text IO and returns it.

- Issue #5859: Remove switch from '%f' to '%g'-style formatting for
  floats with absolute value over 1e50.  Also remove length
  restrictions for float formatting: '%.67f' % 12.34 and '%.120e' %
  12.34 no longer raise an exception.

- Issue #1588: Add complex.__format__. For example,
  format(complex(1, 2./3), '.5') now produces a sensible result.

- Issue #5864: Fix empty format code formatting for floats so that it
  never gives more than the requested number of significant digits.

- Issue #5793: Rationalize isdigit / isalpha / tolower, etc. Includes
  new Py_ISDIGIT / Py_ISALPHA / Py_TOLOWER, etc. in pctypes.h.

- Issue #5835: Deprecate PyOS_ascii_formatd.

- Issue #4971: Fix titlecase for characters that are their own
  titlecase, but not their own uppercase.

- Issue #5283: Setting __class__ in __del__ caused a segfault.

- Issue #5816: complex(repr(z)) now recovers z exactly, even when
  z involves nans, infs or negative zeros.

- Issue #3166: Make int -> float conversions correctly rounded.

- Issue #1869 (and many duplicates): make round(x, n) correctly
  rounded for a float x, by using the decimal <-> binary conversions
  from Python/dtoa.c.  As a consequence, (e.g.) round(x, 2) now
  consistently agrees with format(x, '.2f').

- Issue #5787: object.__getattribute__(some_type, "__bases__") segfaulted on
  some builtin types.

- Issue #5772: format(1e100, '<') produces '1e+100', not '1.0e+100'.

- Issue #5515: str.format() type 'n' combined with commas and leading
  zeros no longer gives odd results with ints and floats.

- Implement PEP 378, Format Specifier for Thousands Separator, for
  floats.

- The str function switches to exponential notation at
  1e11, not 1e12.  This avoids printing 13 significant digits in
  situations where only 12 of them are correct.  Example problem
  value: str(1e11 + 0.5).  (This minor issue has existed in 2.x for a
  long time.)

- Issue #1580: On most platforms, use a 'short' float repr: for a
  finite float x, repr(x) now outputs a string based on the shortest
  sequence of decimal digits that rounds to x.  Previous behaviour was
  to output 17 significant digits and then strip trailing zeros.
  Another minor difference is that the new repr switches to
  exponential notation at 1e16 instead of the previous 1e17; this
  avoids misleading output in some cases.

  There's a new sys attribute sys.float_repr_style, which takes
  the value 'short' to indicate that we're using short float repr,
  and 'legacy' if the short float repr isn't available for one
  reason or another.

  The float repr change involves incorporating David Gay's 'perfect
  rounding' code into the Python core (it's in Python/dtoa.c).  As a
  secondary consequence, all string-to-float and float-to-string
  conversions (including all float formatting operations) will be
  correctly rounded on these platforms.

  See issue #1580 discussions for details of platforms for which
  this change does not apply.

- Issue #5759: float() didn't call __float__ on str subclasses.

- The string.maketrans() function is deprecated; there is a new static method
  maketrans() on the bytes and bytearray classes.  This removes confusion about
  the types string.maketrans() is supposed to work with, and mirrors the
  methods available on the str class.

- Issue #2170: refactored xml.dom.minidom.normalize, increasing both
  its clarity and its speed.

- Issue #1113244: Py_XINCREF, Py_DECREF, Py_XDECREF: Add ``do { ... } while (0)``
  to avoid compiler warnings.

- Issue #3739: The unicode-internal encoder now reports the number of characters
  consumed like any other encoder (instead of the number of bytes).

Installation
------------

- Issue #5756: Install idle and pydoc with a 3 suffix.

Library
-------

- Issue #8203: Fix IDLE Credits dialog: view_file() uses its encoding argument.

- Issue #5311: bdist_msi can now build packages that do not depend on a
  specific Python version.

- Issue #5150: IDLE's format menu now has an option to strip trailing
  whitespace.

- Issue #5940: distutils.command.build_clib.check_library_list was not doing
  the right type checkings anymore.

- Issue #4875: On win32, ctypes.util.find_library does no longer
  return directories.

- Issue #5142: Add the ability to skip modules while stepping to pdb.

- Issue #1309567: Fix linecache behavior of stripping subdirectories when
  looking for files given by a relative filename.

- Issue #5923: Update the ``turtle`` module to version 1.1, add two new
  turtle demos in Demo/turtle.

- Issue #5692: In :class:`zipfile.Zipfile`, fix wrong path calculation when
  extracting a file to the root directory.

- Issue #5913: os.listdir() should fail for empty path on windows.

- Issue #5084: unpickling now interns the attribute names of pickled objects,
  saving memory and avoiding growth in size of subsequent pickles. Proposal
  and original patch by Jake McGuire.

- The json module now works exclusively with str and not bytes.

- Issue #3959: The ipaddr module has been added to the standard library.
  Contributed by Google.

- Issue #3002: ``shutil.copyfile()`` and ``shutil.copytree()`` now raise an
  error when a named pipe is encountered, rather than blocking infinitely.

- Issue #5857: tokenize.tokenize() now returns named tuples.

- Issue #4305: ctypes should now build again on mipsel-linux-gnu

- Issue #1734234: Massively speedup ``unicodedata.normalize()`` when the
  string is already in normalized form, by performing a quick check beforehand.
  Original patch by Rauli Ruohonen.

- Issue #5853: calling a function of the mimetypes module from several threads
  at once could hit the recursion limit if the mimetypes database hadn't been
  initialized before.

- Issue #5854: Updated __all__ to include some missing names and remove some
  names which should not be exported.

- Issue #3102:  All global symbols that the _ctypes extension defines
  are now prefixed with 'Py' or '_ctypes'.

- Issue #5041: ctypes does now allow pickling wide character.

- Issue #5812: For the two-argument form of the Fraction constructor,
  Fraction(m, n), m and n are permitted to be arbitrary Rational
  instances.

- Issue #5812: Fraction('1e6') is valid: more generally, any string
  that's valid for float() is now valid for Fraction(), with the
  exception of strings representing NaNs and infinities.

- Issue #5734: BufferedRWPair was poorly tested and had several glaring
  bugs. Patch by Brian Quinlan.

- Issue #1161031: fix readwrite select flag handling: POLLPRI now
  results in a handle_expt_event call, not handle_read_event, and POLLERR
  and POLLNVAL now call handle_close, not handle_expt_event.  Also,
  dispatcher now has an 'ignore_log_types' attribute for suppressing
  log messages, which is set to 'warning' by default.

- Issue #2703: SimpleXMLRPCDispatcher.__init__: Provide default values for
  new arguments introduced in 2.5.

- Issue #5828 (Invalid behavior of unicode.lower): Fixed bogus logic in
  makeunicodedata.py and regenerated the Unicode database (This fixes
  u'\u1d79'.lower() == '\x00').

Extension Modules
-----------------

- Issue #5881: Remove old undocumented compatibility interfaces in hashlib and
  pwd.

- Issue #5463: In struct module, remove deprecated float coercion
  for integer type codes: struct.pack('L', 0.3) should now raise
  an error.  The _PY_STRUCT_FLOAT_COERCE constant has been removed.
  The version number has been bumped to 0.3.

- Issue #5359: Readd the Berkeley DB detection code to allow _dbm be built
  using Berkeley DB.

Tests
-----

- Issue #5354: New test support function import_fresh_module() makes
  it easy to import both normal and optimised versions of modules.
  test_heapq and test_warnings have been adjusted to use it, tests for
  other modules with both C and Python implementations in the stdlib
  can be adjusted to use it over time.

- Issue #5837: Certain sequences of calls to set() and unset() for
  support.EnvironmentVarGuard objects restored the environment variables
  incorrectly on __exit__.

C-API
-----

- Issue #5630: A replacement PyCObject API, PyCapsule, has been added.


What's New in Python 3.1 alpha 2?
=================================

*Release date: 2009-4-4*

Core and Builtins
-----------------

- Implement PEP 378, Format Specifier for Thousands Separator, for
  integers.

- Issue #5666: Py_BuildValue's 'c' code should create byte strings.

- Issue #5499: The 'c' code for argument parsing functions now only accepts a
  byte, and the 'C' code only accepts a unicode character.

- Fix a problem in PyErr_NormalizeException that leads to "undetected errors"
  when hitting the recursion limit under certain circumstances.

- Issue #1665206: Remove the last eager import in _warnings.c and make it lazy.

- Fix a segfault when running test_exceptions with coverage, caused by
  insufficient checks in accessors of Exception.__context__.

- Issue #5604: non-ASCII characters in module name passed to
  imp.find_module() were converted to UTF-8 while the path is
  converted to the default filesystem encoding, causing nonsense.

- Issue #5126: str.isprintable() returned False for space characters.

- Issue #4865: On MacOSX /Library/Python/2.7/site-packages is added to
  the end sys.path, for compatibility with the system install of Python.

- Issue #4688: Add a heuristic so that tuples and dicts containing only
  untrackable objects are not tracked by the garbage collector. This can
  reduce the size of collections and therefore the garbage collection overhead
  on long-running programs, depending on their particular use of datatypes.

- Issue #5512: Rewrite PyLong long division algorithm (x_divrem) to
  improve its performance.  Long divisions and remainder operations
  are now between 50% and 150% faster.

- Issue #4258: Make it possible to use base 2**30 instead of base
  2**15 for the internal representation of integers, for performance
  reasons.  Base 2**30 is enabled by default on 64-bit machines.  Add
  --enable-big-digits option to configure, which overrides the
  default.  Add sys.int_info structseq to provide information about
  the internal format.

- Issue #4474: PyUnicode_FromWideChar now converts characters outside
  the BMP to surrogate pairs, on systems with sizeof(wchar_t) == 4
  and sizeof(Py_UNICODE) == 2.

- Issue #5237: Allow auto-numbered fields in str.format(). For
  example: '{} {}'.format(1, 2) == '1 2'.

- Issue #5392: when a very low recursion limit was set, the interpreter would
  abort with a fatal error after the recursion limit was hit twice.

- Issue #3845: In PyRun_SimpleFileExFlags avoid invalid memory access with
  short file names.

Library
-------

- Issue #2625: added missing items() call to the for loop in
  mailbox.MH.get_message().

- Issue #5640: Fix _multibytecodec so that CJK codecs don't repeat
  error substitutions from non-strict codec error callbacks in
  incrementalencoder and StreamWriter.

- Issue #5656: Fix the coverage reporting when running the test suite with
  the -T argument.

- Issue #5647: MutableSet.__iand__() no longer mutates self during iteration.

- Issue #5624: Fix the _winreg module name still used in several modules.

- Issue #5628: Fix io.TextIOWrapper.read() with a unreadable buffer.

- Issue #5619: Multiprocessing children disobey the debug flag and causes
  popups on windows buildbots. Patch applied to work around this issue.

- Issue #5400: Added patch for multiprocessing on netbsd compilation/support

- Issue #5387: Fixed mmap.move crash by integer overflow.

- Issue #5261: Patch multiprocessing's semaphore.c to support context
  manager use: "with multiprocessing.Lock()" works now.

- Issue #5236: Change time.strptime() to only take strings. Didn't work with
  bytes already but the failure was non-obvious.

- Issue #5177: Multiprocessing's SocketListener class now uses
  socket.SO_REUSEADDR on all connections so that the user no longer needs
  to wait 120 seconds for the socket to expire.

- Issue #5595: Fix UnboundedLocalError in ntpath.ismount().

- Issue #1174606: Calling read() without arguments of an unbounded file
  (typically /dev/zero under Unix) could crash the interpreter.

- The max_buffer_size arguments of io.BufferedWriter, io.BufferedRWPair, and
  io.BufferedRandom have been deprecated for removal in Python 3.2.

- Issue #5068: Fixed the tarfile._BZ2Proxy.read() method that would loop
  forever on incomplete input. That caused tarfile.open() to hang when used
  with mode 'r' or 'r:bz2' and a fileobj argument that contained no data or
  partial bzip2 compressed data.

- Issue #2110: Add support for thousands separator and 'n' type
  specifier to Decimal.__format__

- Fix Decimal.__format__ bug that swapped the meanings of the '<' and
  '>' alignment characters.

- The error detection code in FileIO.close() could fail to reflect the `errno`
  value, and report it as -1 instead.

- Issue #5016: FileIO.seekable() could return False if the file position
  was negative when truncated to a C int. Patch by Victor Stinner.

Extension Modules
-----------------

- Issue #5391: mmap now deals exclusively with bytes.

- Issue #5463: In struct module, remove deprecated overflow wrapping
  when packing an integer: struct.pack('=L', -1) now raises
  struct.error instead of returning b'\xff\xff\xff\xff'.  The
  _PY_STRUCT_RANGE_CHECKING and _PY_STRUCT_OVERFLOW_MASKING constants
  have been removed from the struct module.


What's New in Python 3.1 alpha 1
================================

*Release date: 2009-03-07*

Core and Builtins
-----------------

- The io module has been reimplemented in C for speed.

- Give dict views an informative __repr__.

- Issue #5247: Improve error message when unknown format codes are
  used when using str.format() with str, int, and float arguments.

- Issue #5249: time.strftime returned malformed string when format string
  contained non ascii character on windows.

- Issue #4626: compile(), exec(), and eval() ignore the coding cookie if the
  source has already been decoded into str.

- Issue #5186: Reduce hash collisions for objects with no __hash__ method by
  rotating the object pointer by 4 bits to the right.

- Issue #4575: Fix Py_IS_INFINITY macro to work correctly on x87 FPUs:
  it now forces its argument to double before testing for infinity.

- Issue #5137: Make len() correctly raise a TypeError when a __len__ method
  returns a non-number type.

- Issue #5182: Removed memoryview.__str__.

- Issue #1717: Removed builtin cmp() function, dropped tp_compare
  slot, the C API functions PyObject_Compare and PyUnicode_Compare and
  the type definition cmpfunc.  The tp_compare slot has been renamed
  to tp_reserved, and is reserved for future usage.

- Issue #1242657: the __len__() and __length_hint__() calls in several tools
  were suppressing all exceptions.  These include list() and bytearray().

- Issue #4707: round(x, n) now returns an integer if x is an integer.
  Previously it returned a float.

- Issue #4753: By enabling a configure option named '--with-computed-gotos'
  on compilers that support it (notably: gcc, SunPro, icc), the bytecode
  evaluation loop is compiled with a new dispatch mechanism which gives
  speedups of up to 20%, depending on the system, on various benchmarks.

- Issue #4874: Most builtin decoders now reject unicode input.

- Issue #4842: Don't allow trailing 'L' when constructing an integer
  from a string.

- Issue #4991: os.fdopen now raises an OSError for invalid file descriptors.

- Issue #4838: When a module is deallocated, free the memory backing the
  optional module state data.

- Issue #4910: Rename nb_long slot to nb_reserved, and change its
  type to ``(void *)``.

- Issue #4935: The overflow checking code in the expandtabs() method common
  to str, bytes and bytearray could be optimized away by the compiler, letting
  the interpreter segfault instead of raising an error.

- Issue #3720: Fix a crash when an iterator modifies its class and removes its
  __next__ method.

- Issue #4910: Builtin int() function and PyNumber_Long/PyNumber_Int API
  function no longer attempt to call the __long__ slot to convert an object
  to an integer.  Only the __int__ and __trunc__ slots are examined.

- Issue #4893: Use NT threading on CE.

- Issue #4915: Port sysmodule to Windows CE.

- Issue #4868: utf-8, utf-16 and latin1 decoding are now 2x to 4x faster. The
  common cases are optimized thanks to a dedicated fast path and a moderate
  amount of loop unrolling.

- Issue #4074: Change the criteria for doing a full garbage collection (i.e.
  collecting the oldest generation) so that allocating lots of objects without
  destroying them does not show quadratic performance. Based on a proposal by
  Martin von Löwis at
  http://mail.python.org/pipermail/python-dev/2008-June/080579.html.

- Issue #4604: Some objects of the I/O library could still be used after
  having been closed (for instance, a read() call could return some
  previously buffered data). Patch by Dmitry Vasiliev.

- Issue #4705: Fix the -u ("unbuffered binary stdout and stderr") command-line
  flag to work properly. Furthermore, when specifying -u, the text stdout
  and stderr streams have line-by-line buffering enabled (the default being
  to buffer arbitrary chunks of data).

- The internal table, _PyLong_DigitValue, is now an array of unsigned chars
  instead of ints (reducing its size from 4 to 8 times thereby reducing
  Python's overall memory).

- Issue #1180193: When importing a module from a .pyc (or .pyo) file with
  an existing .py counterpart, override the co_filename attributes of all
  code objects if the original filename is obsolete (which can happen if the
  file has been renamed, moved, or if it is accessed through different paths).
  Patch by Ziga Seilnacht and Jean-Paul Calderone.

- Issue #4580: Fix slicing of memoryviews when the item size is greater than
  one byte. Also fixes the meaning of len() so that it returns the number of
  items, rather than the size in bytes.

- Issue #4075: Use OutputDebugStringW in Py_FatalError.

- Issue #4747: When the terminal does not use utf-8, executing a script with
  non-ascii characters in its name could fail with a "SyntaxError: None" error.

- Issue #4797: IOError.filename was not set when ``_fileio.FileIO`` failed
  to open file with a bytes filename on Windows.

- Issue #3680: Reference cycles created through a dict, set or deque iterator
  did not get collected.

- Issue #4701: PyObject_Hash now implicitly calls PyType_Ready on types
  where the tp_hash and tp_dict slots are both NULL.

- Issue #4759: None is now allowed as the first argument of
  bytearray.translate().  It was always allowed for bytes.translate().

- Added test case to ensure attempts to read from a file opened for writing
  fail.

- Issue #3106: Speedup some comparisons (str/str and int/int).

- Issue #2183: Simplify and optimize bytecode for list, dict and set
  comprehensions. Original patch for list comprehensions by Neal Norwitz.

- Issue #2467: gc.DEBUG_STATS reported invalid elapsed times. Also, always
  print elapsed times, not only when some objects are uncollectable /
  unreachable. Original patch by Neil Schemenauer.

- Issue #3439: Add a bit_length method to int.

- Issue #2173: When getting device encoding, check that return value of
  nl_langinfo is not the empty string.  This was causing silent build
  failures on OS X.

- Issue #4597: Fixed several opcodes that weren't always propagating
  exceptions.

- Issue #4589: Fixed exception handling when the __exit__ function of a
  context manager returns a value that cannot be converted to a bool.

- Issue #4445: Replace "sizeof(PyBytesObject)" with
  "offsetof(PyBytesObject, ob_sval) + 1" when allocating memory for
  bytes instances.  On a typical machine this saves 3 bytes of memory
  (on average) per allocation of a bytes instance.

- Issue #4533: File read operation was dreadfully slow due to a slowly
  growing read buffer. Fixed by using the same growth rate algorithm as
  Python 2.x.

- Issue #4509: Various issues surrounding resize of bytearray objects to
  which there are buffer exports (e.g. memoryview instances).

- Issue #4233: Changed semantic of ``_fileio.FileIO``'s ``close()``
  method on file objects with closefd=False. The file descriptor is still
  kept open but the file object behaves like a closed file. The ``FileIO``
  object also got a new readonly attribute ``closefd``.

- Issue #4569: Interpreter crash when mutating a memoryview with an item size
  larger than 1.

- Issue #4748: Lambda generators no longer return a value.

- The re.sub(), re.subn() and re.split() functions now accept a flags parameter.

- Issue #5108: Handle %s like %S, %R and %A in PyUnicode_FromFormatV(): Call
  PyUnicode_DecodeUTF8() once, remember the result and output it in a second
  step. This avoids problems with counting UTF-8 bytes that ignores the effect
  of using the replace error handler in PyUnicode_DecodeUTF8().

Library
-------

- Issue #7071: byte-compilation in Distutils is now done with respect to
  sys.dont_write_bytecode.

- Issue #7066: archive_util.make_archive now restores the cwd if an error is
  raised. Initial patch by Ezio Melotti.

- Issue #6516: Added owner/group support when creating tar archives in
  Distutils.

- Issue #6954: Fixed crash when using DISTUTILS_DEBUG flag in Distutils.

- Issue #6163: Fixed HP-UX runtime library dir options in
  distutils.unixcompiler. Initial patch by Sridhar Ratnakumar and
  Michael Haubenwallner.

- Issue #6693: New functions in site.py to get user/global site packages paths.

- Issue #6511: ZipFile now raises BadZipfile (instead of an IOError) when
  opening an empty or very small file.

- Issue #6545: Removed assert statements in distutils.Extension, so the
  behavior is similar when used with -O.

- unittest has been split up into a package.  All old names should still work.

- Issue #6466: now distutils.cygwinccompiler and distutils.emxccompiler
  uses the same refactored function to get gcc/ld/dllwrap versions numbers.
  It's `distutils.util.get_compiler_versions`. Added deprecation warnings
  for the obsolete get_versions() functions.

- Issue #6433: fixed issues with multiprocessing.pool.map hanging on empty list

- Issue #6314: logging: Extra checks on the "level" argument in more places.

- Issue #2622: Fixed an ImportError when importing email.message from a
  standalone application built with py2exe or py2app.

- Issue #6455: Fixed test_build_ext under win32.

- Issue #6377: Enabled the compiler option, and deprecate its usage as an
  attribute.

- Issue #6413: Fixed the log level in distutils.dist for announce.

- Issue #6403: Fixed package path usage in build_ext.

- Issues #5155, 5313, 5331: multiprocessing.Process._bootstrap was
  unconditionally calling "os.close(sys.stdin.fileno())" resulting in file
  descriptor errors

- Issue #6365: Distutils build_ext inplace mode was copying the compiled
  extension in a subdirectory if the extension name had dots.

- Issue #6164: Added an AIX specific linker argument in Distutils
  unixcompiler. Original patch by Sridhar Ratnakumar.

- Issue #6286: Now Distutils upload command is based on urllib2 instead of
  httplib, allowing the usage of http_proxy.

- Issue #6287: Added the license field in Distutils documentation.

- Issue #6263: Fixed syntax error in distutils.cygwincompiler.

- Issue #5201: distutils.sysconfig.parse_makefile() now understands `$$`
  in Makefiles. This prevents compile errors when using syntax like:
  `LDFLAGS='-rpath=\$$LIB:/some/other/path'`. Patch by Floris Bruynooghe.

- Issue #6131: test_modulefinder leaked when run after test_distutils.
  Patch by Hirokazu Yamamoto.

- Issue #6048: Now Distutils uses the tarfile module in archive_util.

- Issue #6062: In distutils, fixed the package option of build_ext. Feedback
  and tests on pywin32 by Tim Golden.

- Issue #6053: Fixed distutils tests on win32. patch by Hirokazu Yamamoto.

- Issue #6046: Fixed the library extension when distutils build_ext is used
  inplace. Initial patch by Roumen Petrov.

- Issue #6041: Now distutils `sdist` and `register` commands use `check` as a
  subcommand.

- Issue #6022: a test file was created in the current working directory by
  test_get_outputs in Distutils.

- Issue #5977: distutils build_ext.get_outputs was not taking into account the
  inplace option. Initial patch by kxroberto.

- Issue #5984: distutils.command.build_ext.check_extensions_list checks were broken
  for old-style extensions.

- Issue #5976: Fixed Distutils test_check_environ.

- Issue #5941: Distutils build_clib command was not working anymore because
  of an incomplete costumization of the archiver command. Added ARFLAGS in the
  Makefile besides AR and make Distutils use it. Original patch by David
  Cournapeau.

- Issue #2245: aifc now skips chunk types it doesn't recognize, per spec.

- Issue #5874: distutils.tests.test_config_cmd is not locale-sensitive
  anymore.

- Issue #5810: Fixed Distutils test_build_scripts so it uses
  sysconfig.get_config_vars.

- Issue #4951: Fixed failure in test_httpservers.

- Issue #5795: Fixed test_distutils failure on Debian ppc.

- Issue #5607: fixed Distutils test_get_platform for Mac OS X fat binaries.

- Issue #5741: don't disallow "%%" (which is an escape for "%") when setting
  a value in SafeConfigParser.

- Issue #5732: added a new command in Distutils: check.

- Issue #5731: Distutils bdist_wininst no longer worked on non-Windows
  platforms. Initial patch by Paul Moore.

- Issue #5095: Added bdist_msi to the list of bdist supported formats.
  Initial fix by Steven Bethard.

- Issue #1491431: Fixed distutils.filelist.glob_to_re for edge cases.
  Initial fix by Wayne Davison.

- Issue #5694: removed spurious test output in Distutils (test_clean).

- Issue #1326077: fix the formatting of SyntaxErrors by the traceback module.

- Issue #1665206 (partially): Move imports in cgitb to the top of the module
  instead of performing them in functions. Helps prevent import deadlocking in
  threads.

- Issue #2522: locale.format now checks its first argument to ensure it has
  been passed only one pattern, avoiding mysterious errors where it appeared
  that it was failing to do localization.

- Issue #5583: Added optional Extensions in Distutils. Initial patch by Georg
  Brandl.

- Issue #1222: locale.format() bug when the thousands separator is a space
  character.

- Issue #5472: Fixed distutils.test_util tear down. Original patch by
  Tim Golden.

- collections.deque() objects now have a read-only attribute called maxlen.

- Issue #2638: Show a window constructed with tkSimpleDialog.Dialog only after
  it is has been populated and properly configured in order to prevent
  window flashing.

- Issue #4792: Prevent a segfault in _tkinter by using the
  guaranteed to be safe interp argument given to the PythonCmd in place of
  the Tcl interpreter taken from a PythonCmd_ClientData.

- Issue #5193: Guarantee that tkinter.Text.search returns a string.

- Issue #5394: removed > 2.3 syntax from distutils.msvc9compiler.
  Original patch by Akira Kitada.

- Issue #5334: array.fromfile() failed to insert values when EOFError was raised.

- Issue #5385: Fixed mmap crash after resize failure on windows.

- Issue #5179: Fixed subprocess handle leak on failure on windows.

- PEP 372:  Added collections.OrderedDict().

- The _asdict() for method for namedtuples now returns an OrderedDict().

- configparser now defaults to using an ordered dictionary.

- Issue #5401: Fixed a performance problem in mimetypes when ``from mimetypes
  import guess_extension`` was used.

- Issue #1733986: Fixed mmap crash in accessing elements of second map object
  with same tagname but larger size than first map. (Windows)

- Issue #5386: mmap.write_byte didn't check map size, so it could cause buffer
  overrun.

- Issue #1533164: Installed but not listed ``*.pyo`` was breaking Distutils
  bdist_rpm command.

- Issue #5378: added --quiet option to Distutils bdist_rpm command.

- Issue #5052: make Distutils compatible with 2.3 again.

- Issue #5316: Fixed buildbot failures introduced by multiple inheritance
  in Distutils tests.

- Issue #5287: Add exception handling around findCaller() call to help out
  IronPython.

- Issue #5282: Fixed mmap resize on 32bit windows and unix. When offset > 0,
  The file was resized to wrong size.

- Issue #5292: Fixed mmap crash on its boundary access m[len(m)].

- Issue #2279: distutils.sdist.add_defaults now add files
  from the package_data and the data_files metadata.

- Issue #5257: refactored all tests in distutils, so they use
  support.TempdirManager, to avoid writing in the tests directory.

- Issue #4524: distutils build_script command failed with --with-suffix=3.
  Initial patch by Amaury Forgeot d'Arc.

- Issue #2461: added tests for distutils.util

- Issue #4998: The memory saving effect of __slots__ had been lost on Fractions
  which inherited from numbers.py which did not have __slots__ defined.  The
  numbers hierarchy now has its own __slots__ declarations.

- Issue #4631: Fix urlopen() result when an HTTP response uses chunked
  encoding.

- Issue #5203: Fixed ctypes segfaults when passing a unicode string to a
  function without argtypes (only occurs if HAVE_USABLE_WCHAR_T is false).

- Issue #3386: distutils.sysconfig.get_python_lib prefix argument was ignored
  under NT and OS2. Patch by Philip Jenvey.

- Issue #5128: Make compileall properly inspect bytecode to determine if needs
  to be recreated. This avoids a timing hole thanks to the old reliance on the
  ctime of the files involved.

- Issue #5122: Synchronize tk load failure check to prevent a potential
  deadlock.

- Issue #1818: collections.namedtuple() now supports a keyword argument
  'rename' which lets invalid fieldnames be automatically converted to
  positional names in the form, _1, _2, ...

- Issue #4890: Handle empty text search pattern in Tkinter.Text.search.

- Issue #4512 (part 2): Promote ``ZipImporter._get_filename()`` to be a
  public documented method ``ZipImporter.get_filename()``.

- Issue #4195: The ``runpy`` module (and the ``-m`` switch) now support
  the execution of packages by looking for and executing a ``__main__``
  submodule when a package name is supplied. Initial patch by Andi
  Vajda.

- Issue #1731706: Call Tcl_ConditionFinalize for Tcl_Conditions that will
  not be used again (this requires Tcl/Tk 8.3.1), also fix a memory leak in
  Tkapp_Call when calling from a thread different than the one that created
  the Tcl interpreter. Patch by Robert Hancock.

- Issue #4285: Change sys.version_info to be a named tuple. Patch by
  Ross Light.

- Issue #1520877: Now distutils.sysconfig reads $AR from the
  environment/Makefile. Patch by Douglas Greiman.

- Issue #1276768: The verbose option was not used in the code of
  distutils.file_util and distutils.dir_util.

- Issue #5132: Fixed trouble building extensions under Solaris with
  --enabled-shared activated. Initial patch by Dave Peterson.

- Issue #1581476: Always use the Tcl global namespace when calling into Tcl.

- The shelve module now defaults to pickle protocol 3.

- Fix a bug in the trace module where a bytes object from co_lnotab had its
  items being passed through ord().

- Issue #2047: shutil.move() could believe that its destination path was
  inside its source path if it began with the same letters (e.g. "src" vs.
  "src.new").

- Added the ttk module. See issue #2983: Ttk support for Tkinter.

- Removed isSequenceType(), isMappingType, and isNumberType() from the
  operator module; use the abstract base classes instead.  Also removed
  the repeat() function; use mul() instead.

- Issue #5021:  doctest.testfile() did not create __name__ and
  collections.namedtuple() relied on __name__ being defined.

- Backport importlib from Python 3.1. Only the import_module() function has
  been backported to help facilitate transitions from 2.7 to 3.1.

- Issue #1885: distutils. When running sdist with --formats=tar,gztar
  the tar file was overriden by the gztar one.

- Issue #4863: distutils.mwerkscompiler has been removed.

- Added a new itertools functions:  combinations_with_replacement()
  and compress().

- Issue #5032:  added a step argument to itertools.count() and
  allowed non-integer arguments.

- Fix and properly document the multiprocessing module's logging
  support, expose the internal levels and provide proper usage
  examples.

- Issue #1672332: fix unpickling of subnormal floats, which was
  producing a ValueError on some platforms.

- Issue #3881: Help Tcl to load even when started through the
  unreadable local symlink to "Program Files" on Vista.

- Issue #4710: Extract directories properly in the zipfile module;
  allow adding directories to a zipfile.

- Issue #3807: _multiprocessing build fails when configure is passed
  --without-threads argument. When this occurs, _multiprocessing will
  be disabled, and not compiled.

- Issue #5008: When a file is opened in append mode with the new IO library,
  do an explicit seek to the end of file (so that e.g. tell() returns the
  file size rather than 0). This is consistent with the behaviour of the
  traditional 2.x file object.

- Issue #5013: Fixed a bug in FileHandler which occurred when the delay
  parameter was set.

- Issue #4842: Always append a trailing 'L' when pickling longs using
  pickle protocol 0.  When reading, the 'L' is optional.

- Add the importlib package.

- Issue #4301: Patch the logging module to add processName support, remove
  _check_logger_class from multiprocessing.

- Issue #3325: Remove python2.x try: except: imports for old cPickle from
  multiprocessing.

- Issue #4959: inspect.formatargspec now works for keyword only arguments
  without defaults.

- Issue #3321: ``_multiprocessing.Connection()`` doesn't check handle; added checks
  for Unix machines for negative handles and large int handles. Without this check
  it is possible to segfault the interpreter.

- Issue #4449: AssertionError in mp_benchmarks.py, caused by an underlying issue
  in sharedctypes.py.

- Issue #1225107: inspect.isclass() returned True for instances with a custom
  __getattr__.

- Issue #3826 and #4791: The socket module now closes the underlying socket
  appropriately when it is being used via socket.makefile() objects
  rather than delaying the close by waiting for garbage collection to do it.

- Issue #1696199:  Add collections.Counter() for rapid and convenient
  counting.

- Issue #3860: GzipFile and BZ2File now support the context manager protocol.

- Issue #4867: Fixed a crash in ctypes when passing a string to a
  function without defining argtypes.

- Issue #4272: Add an optional argument to the GzipFile constructor to override
  the timestamp in the gzip stream. The default value remains the current time.
  The information can be used by e.g. gunzip when decompressing. Patch by
  Jacques Frechet.

- Restore Python 2.3 compatibility for decimal.py.

- Issue #3638: Remove functions from _tkinter module level that depend on
  TkappObject to work with multiple threads.

- Issue #4718: Adapt the wsgiref package so that it actually works with
  Python 3.x, in accordance with the `official amendments of the spec
  <http://www.wsgi.org/wsgi/Amendments_1.0>`_.

- Issue #4796: Added Decimal.from_float() and Context.create_decimal_from_float()
  to the decimal module.

- Fractions.from_float() no longer loses precision for integers too big to
  cast as floats.

- Issue #4812: add missing underscore prefix to some internal-use-only
  constants in the decimal module.  (Dec_0 becomes _Dec_0, etc.)

- Issue #4790: The nsmallest() and nlargest() functions in the heapq module
  did unnecessary work in the common case where no key function was specified.

- Issue #4795: inspect.isgeneratorfunction() returns False instead of None when
  the function is not a generator.

- Issue #4702: Throwing a DistutilsPlatformError instead of IOError in case
  no MSVC compiler is found under Windows. Original patch by Philip Jenvey.

- Issue #4646: distutils was choking on empty options arg in the setup
  function. Original patch by Thomas Heller.

- Issue #3767: Convert Tk object to string in tkColorChooser.

- Issue #3248: Allow placing ScrolledText in a PanedWindow.

- Issue #4444: Allow assertRaises() to be used as a context handler, so that
  the code under test can be written inline if more practical.

- Issue #4739: Add pydoc help topics for symbols, so that e.g. help('@')
  works as expected in the interactive environment.

- Issue #4756: zipfile.is_zipfile() now supports file-like objects. Patch by
  Gabriel Genellina.

- Issue #4574: reading an UTF16-encoded text file crashes if \r on 64-char
  boundary.

- Issue #4223: inspect.getsource() will now correctly display source code
  for packages loaded via zipimport (or any other conformant PEP 302
  loader). Original patch by Alexander Belopolsky.

- Issue #4201: pdb can now access and display source code loaded via
  zipimport (or any other conformant PEP 302 loader). Original patch by
  Alexander Belopolsky.

- Issue #4197: doctests in modules loaded via zipimport (or any other PEP
  302 conformant loader) will now work correctly in most cases (they
  are still subject to the constraints that exist for all code running
  from inside a module loaded via a PEP 302 loader and attempting to
  perform IO operations based on __file__). Original patch by
  Alexander Belopolsky.

- Issues #4082 and #4512: Add runpy support to zipimport in a manner that
  allows backporting to maintenance branches. Original patch by
  Alexander Belopolsky.

- Issue #4163: textwrap module: allow word splitting on a hyphen preceded by
  a non-ASCII letter.

- Issue #4616: TarFile.utime(): Restore directory times on Windows.

- Issue #4021: tokenize.detect_encoding() now raises a SyntaxError when the
  codec cannot be found.  This is for compatibility with the builtin behavior.

- Issue #4084: Fix max, min, max_mag and min_mag Decimal methods to
  give correct results in the case where one argument is a quiet NaN
  and the other is a finite number that requires rounding.

- Issue #4483: _dbm module now builds on systems with gdbm & gdbm_compat
  libs.

- Added the subprocess.check_call_output() convenience function to get output
  from a subprocess on success or raise an exception on error.

- Issue #1055234: cgi.parse_header(): Fixed parsing of header parameters to
  support unusual filenames (such as those containing semi-colons) in
  Content-Disposition headers.

- Issue #4384: Added logging integration with warnings module using
  captureWarnings(). This change includes a NullHandler which does nothing;
  it will be of use to library developers who want to avoid the "No handlers
  could be found for logger XXX" message which can appear if the library user
  doesn't configure logging.

- Issue #3741: DISTUTILS_USE_SDK set causes msvc9compiler.py to raise an
  exception.

- Issue #4529: fix the parser module's validation of try-except-finally
  statements.

- Issue #4458: getopt.gnu_getopt() now recognizes a single "-" as an argument,
  not a malformed option.

- Added the subprocess.check_output() convenience function to get output
  from a subprocess on success or raise an exception on error.

- Issue #4542: On Windows, binascii.crc32 still accepted str as binary input;
  the corresponding tests now pass.

- Issue #4537: webbrowser.UnixBrowser would fail to open the browser because
  it was calling the wrong open() function.

- Issue #1055234: cgi.parse_header(): Fixed parsing of header parameters to
  support unusual filenames (such as those containing semi-colons) in
  Content-Disposition headers.

- Issue #4861: ctypes.util.find_library(): Robustify. Fix library detection on
  biarch systems. Try to rely on ldconfig only, without using objdump and gcc.

- Issue #5104: The socket module now raises OverflowError when 16-bit port and
  protocol numbers are supplied outside the allowed 0-65536 range on bind()
  and getservbyport().

- Windows locale mapping updated to Vista.

Tools/Demos
-----------

- Issue #4704: remove use of cmp() in pybench, bump its version number to 2.1,
  and make it 2.6-compatible.

- Ttk demos added in Demo/tkinter/ttk/

- Issue #4677: add two list comprehension tests to pybench.


Build
-----

- Issue #6094: Build correctly with Subversion 1.7.

- Issue #5847: Remove -n switch on "Edit with IDLE" menu item.

- Issue #5726: Make Modules/ld_so_aix return the actual exit code of the
  linker, rather than always exit successfully. Patch by Floris Bruynooghe.

- Issue #4587: Add configure option --with-dbmliborder=db1:db2:... to specify
  the order that backends for the dbm extension are checked.

- Link the shared python library with $(MODLIBS).

- Issue #5134: Silence compiler warnings when compiling sqlite with VC++.

- Issue #4494: Fix build with Py_NO_ENABLE_SHARED on Windows.

- Issue #4895: Use _strdup on Windows CE.

- Issue #4472: "configure --enable-shared" now works on OSX

- Issues #4728 and #4060: WORDS_BIGEDIAN is now correct in Universal builds.

- Issue #4389: Add icon to the uninstall entry in "add-and-remove-programs".

- Issue #4289: Remove Cancel button from AdvancedDlg.

- Issue #1656675: Register a drop handler for .py* files on Windows.

- Issue #4120: Exclude manifest from extension modules in VS2008.

- Issue #4091: Install pythonxy.dll in system32 again.

- Issue #4018: Disable "for me" installations on Vista.

- Issue #3758: Add ``patchcheck`` build target to .PHONY.

- Issue #4204: Fixed module build errors on FreeBSD 4.


C-API
-----

- Issue #6624: yArg_ParseTuple with "s" format when parsing argument with
  NUL: Bogus TypeError detail string.

- Issue #5175: PyLong_AsUnsignedLongLong now raises OverflowError
  for negative arguments.  Previously, it raised TypeError.

- Issue #4720: The format for PyArg_ParseTupleAndKeywords can begin with '|'.

- Issue #3632: from the gdb debugger, the 'pyo' macro can now be called when
  the GIL is released, or owned by another thread.

- Issue #4122: On Windows, fix a compilation error when using the
  Py_UNICODE_ISSPACE macro in an extension module.


Extension Modules
-----------------

- Issue #3745: Fix hashlib to always reject unicode and non buffer-api
  supporting objects as input no matter how it was compiled (built in
  implementations or external openssl library).

- Issue #4397: Fix occasional test_socket failure on OS X.

- Issue #4279: Fix build of parsermodule under Cygwin.

- Issue #4751: hashlib now releases the GIL when hashing large buffers
  (with a hardwired threshold of 2048 bytes), allowing better parallelization
  on multi-CPU systems. Contributed by Lukas Lueg (ebfe) and Victor Stinner.

- Issue #4051: Prevent conflict of UNICODE macros in cPickle.

- Issue #4738: Each zlib object now has a separate lock, allowing to compress
  or decompress several streams at once on multi-CPU systems. Also, the GIL
  is now released when computing the CRC of a large buffer. Patch by ebfe.

- Issue #4228: Pack negative values the same way as 2.4 in struct's L format.

- Issue #1040026: Fix os.times result on systems where HZ is incorrect.

- Issues #3167, #3682: Fix test_math failures for log, log10 on Solaris,
  OpenBSD.

- Issue #4583: array.array would not always prohibit resizing when a buffer
  has been exported, resulting in an interpreter crash when accessing the
  buffer.


- Issue #5228: Make functools.partial objects can now be pickled.

Tests
-----

- Issue #6152: New option '-j'/'--multiprocess' for regrtest allows running
  regression tests in parallel, shortening the total runtime.

- Issue #5450: Moved tests involving loading tk from Lib/test/test_tcl to
  Lib/tkinter/test/test_tkinter/test_loadtk. With this, these tests demonstrate
  the same behaviour as test_ttkguionly (and now also test_tk) which is to
  skip the tests if DISPLAY is defined but can't be used.

- regrtest no longer treats ImportError as equivalent to SkipTest.  Imports
  that should cause a test to be skipped are now done using import_module
  from test support, which does the conversion.

- Issue #5083: New 'gui' resource for regrtest.


Docs
----


**(For information about older versions, consult the HISTORY file.)**