Misc/NEWS

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

(editors: check NEWS.help for information about editing NEWS using ReST.)

What's New in Python 3.0 beta 5
===============================

[Note: due to the number of unresolved issues we're going back to beta
 releases for a while.]

*Release date: XX-XXX-2008*

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

- Issue #4200: Changed the atexit module to store its state in its
  PyModuleDef atexitmodule. This fixes a bug with multiple subinterpeters.

- Issue #4237: io.FileIO() was raising invalid warnings caused by
  insufficient initialization of PyFileIOObject struct members.

- Issue #4170: Pickling a collections.defaultdict object would crash the
  interpreter.

- Issue #4146: Compilation on OpenBSD has been restored.

- Issue #3574: compile() incorrectly handled source code encoded as Latin-1.

- Issues #2384 and #3975: Tracebacks were not correctly printed when the
  source file contains a ``coding:`` header: the wrong line was displayed, and
  the encoding was not respected.

- Issue #3740: Null-initialize module state.

- Issue #3946: PyObject_CheckReadBuffer crashed on a memoryview object.

- Issue #1688: On Windows, the input() prompt was not correctly displayed if it
  contains non-ascii characters.

- Bug #3951: Py_USING_MEMORY_DEBUGGER should not be enabled by default.

Library
-------

- Issue #3664: The pickle module could segfault if a subclass of Pickler fails
  to call the base __init__ method.

- Issue #3725: telnetlib now works completely in bytes.

- Issue #4072: Restore build_py_2to3.

- Issue #4014: Don't claim that Python has an Alpha release status, in addition
  to claiming it is Mature.

- Issue #3187: Add sys.setfilesystemencoding.

- Issue #3187: Better support for "undecodable" filenames.  Code by Victor
  Stinner, with small tweaks by GvR.

- Issue #3965: Allow repeated calls to turtle.Screen, by making it a
  true singleton object.

- Issue #3911: ftplib.FTP.makeport() could give invalid port numbers.

- Issue #3929: When the database cannot be opened, dbm.open() would incorrectly
  raise a TypeError: "'tuple' object is not callable" instead of the expected
  dbm.error.

- Bug #3884: Make the turtle module toplevel again.

- Issue #3547: Fixed ctypes structures bitfields of varying integer
  sizes.

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

- Issue #3659: Subclasses of str didn't work as SQL parameters.

Build
-----

- Issue #4091: Install pythonxy.dll in system32 again.

- Issue #4018: Disable "for me" installations on Vista.

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

- Issue #3717: Fix Demo/embed/demo.c.

- Issue #4072: Add a distutils demo for build_py_2to3.


What's New in Python 3.0 release candidate 1
============================================

*Release date: 17-Sep-2008*

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

- Issue #3827: memoryview lost its size attribute in favor of using len(view).

- Issue #3813: could not lanch python.exe via symbolic link on cygwin.

- Issue #3705: fix crash when given a non-ascii value on the command line for
  the "-c" and "-m" parameters. Now the behaviour is as expected under Linux,
  although under Windows it fails at a later point.

- Issue #3279: Importing site at interpreter was failing silently because the
  site module uses the open builtin which was not initialized at the time.

- Issue #3660: Corrected a reference leak in str.encode() when the encoder
  does not return a bytes object.

- Issue #3774: Added a few more checks in PyTokenizer_FindEncoding to handle
  error conditions.

- Issue #3594: Fix Parser/tokenizer.c:fp_setreadl() to open the file being
  tokenized by either a file path or file pointer for the benefit of
  PyTokenizer_FindEncoding().

- Issue #3696: Error parsing arguments on OpenBSD <= 4.4 and Cygwin. On
  these systems, the mbstowcs() function is slightly buggy and must be
  replaced with strlen() for the purpose of counting of number of wide
  characters needed to represent the multi-byte character string.

- Issue #3697: "Fatal Python error: Cannot recover from stack overflow"
  could be easily encountered under Windows in debug mode when exercising
  the recursion limit checking code, due to bogus handling of recursion
  limit when USE_STACKCHEK was enabled.

- Issue 3639: The _warnings module could segfault the interpreter when
  unexpected types were passed in as arguments.

- Issue #3712: The memoryview object had a reference leak and didn't support
  cyclic garbage collection.

- Issue #3668: Fix a memory leak with the "s*" argument parser in
  PyArg_ParseTuple and friends, which occurred when the argument for "s*" 
  was correctly parsed but parsing of subsequent arguments failed.

- Issue #3611: An exception __context__ could be cleared in a complex pattern
  involving a __del__ method re-raising an exception.

- Issue #2534: speed up isinstance() and issubclass() by 50-70%, so as to 
  match Python 2.5 speed despite the __instancecheck__ / __subclasscheck__
  mechanism. In the process, fix a bug where isinstance() and issubclass(),
  when given a tuple of classes as second argument, were looking up
  __instancecheck__ / __subclasscheck__ on the tuple rather than on each
  type object.

- Issue #3663: Py_None was decref'd when printing SyntaxErrors.

- Issue #3651: Fix various memory leaks when using the buffer
  interface, or when the "s#" code of PyArg_ParseTuple is given a
  bytes object.

- Issue #3657: Fix uninitialized memory read when pickling longs.
  Found by valgrind.

- Apply security patches from Apple.

- Fix crashes on memory allocation failure found with failmalloc.

- Fix memory leaks found with valgrind and update suppressions file.

- Fix compiler warnings in opt mode which would lead to invalid memory reads.

- Fix problem using wrong name in decimal module reported by pychecker.

- Issue #3650: Fixed a reference leak in bytes.split('x').

- bytes(o) now tries to use o.__bytes__() before using fallbacks.

- Issue #1204: The configure script now tests for additional libraries
  that may be required when linking against readline.  This fixes issues
  with x86_64 builds on some platforms (a few Linux flavors and OpenBSD).

C API
-----

- PyObject_Bytes and PyBytes_FromObject were added.

Library
-------

- Issue #3756: make re.escape() handle bytes as well as str.

- Issue #3800: fix filter() related bug in formatter.py.

- Issue #874900: fix behaviour of threading module after a fork.

- Issue #3535: zipfile couldn't read some zip files larger than 2GB.

- Issue #3776: Deprecate the bsddb package for removal in 3.0.

- Issue #3762: platform.architecture() fails if python is lanched via
  its symbolic link.

- Issue #3660: fix a memory leak in the C accelerator of the pickle module.

- Issue #3160: the "bdist_wininst" distutils command didn't work.

- Issue #1658: tkinter changes dict size during iteration in both
    tkinter.BaseWidget and tkinter.scrolledtext.ScrolledText.

- The bsddb module (and therefore the dbm.bsd module) has been removed.
  It is now maintained outside of the standard library at
  http://www.jcea.es/programacion/pybsddb.htm.

- Issue 600362:  Relocated parse_qs() and parse_qsl(), from the cgi module
  to the urlparse one.  Added a DeprecationWarning in the old module, it
  will be deprecated in the future.

- Issue #3719: platform.architecture() fails if there are spaces in the
  path to the Python binary.

- Issue 3602: As part of the merge of r66135, make the parameters on
  warnings.catch_warnings() keyword-only. Also remove a DeprecationWarning.

- The deprecation warnings for the camelCase threading API names were removed.

- Issue #3110: multiprocessing fails to compiel on solaris 10 due to missing 
  SEM_VALUE_MAX.

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

- Issue #3782: os.write() must not accept unicode strings.

- Issue #2975: When compiling several extension modules with Visual Studio 2008
  from the same python interpreter, some environment variables would grow
  without limit.

- Issue #3643: Added a few more checks to _testcapi to prevent segfaults by
  exploitation of poor argument checking.

- bsddb code updated to version 4.7.3pre2. This code is the same than
  Python 2.6 one, since the intention is to keep an unified 2.x/3.x codebase.
  The Python code is automatically translated using "2to3". Please, do not
  update this code in Python 3.0 by hand. Update the 2.6 one and then
  do "2to3".

- The _bytesio and _stringio modules are now compiled into the python binary.

- Issue #3492 and #3790: Fixed the zlib module and zipimport module uses of
  mutable bytearray objects where they should have been using immutable bytes.

- Issue #3797: Fixed the dbm, marshal, mmap, ossaudiodev, & winreg modules to
  return bytes objects instead of bytearray objects.


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

- Fix Misc/gdbinit so it works.


Build
-----

- Issue #3812: Failed to build python if configure --without-threads.

- Issue #3791: Remove the bsddb module from the Windows installer, and the
  core bsddb library from the Windows build files.


What's new in Python 3.0b3?
===========================

*Release date: 20-Aug-2008*

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

- Issue #3653: Fix a segfault when sys.excepthook was called with invalid
  arguments.

- Issue #2394: implement more of the memoryview API, with the caveat that
  only one-dimensional contiguous buffers are supported and exercised right
  now. Slicing, slice assignment and comparison (equality and inequality)
  have been added. Also, the tolist() method has been implemented, but only
  for byte buffers. Endly, the API has been updated to return bytes objects
  wherever it used to return bytearrays.

- Issue #3560: clean up the new C PyMemoryView API so that naming is
  internally consistent; add macros PyMemoryView_GET_BASE() and
  PyMemoryView_GET_BUFFER() to access useful properties of a memory views
  without relying on a particular implementation; remove the ill-named
  PyMemoryView() function (PyMemoryView_GET_BUFFER() can be used instead).

- ctypes function pointers that are COM methods have a boolean True
  value again.

- Issue #1819: function calls with several named parameters are now on
  average 35% faster (as measured by pybench).

- The undocumented C APIs PyUnicode_AsString() and
  PyUnicode_AsStringAndSize() were made private to the interpreter, in
  order to be able to refine their interfaces for Python 3.1.

  If you need to access the UTF-8 representation of a Unicode object
  as bytes string, please use PyUnicode_AsUTF8String() instead.

- Issue #3460: PyUnicode_Join() implementation is 10% to 80% faster thanks
  to Python 3.0's stricter semantics which allow to avoid successive
  reallocations of the result string (this also affects str.join()).


Library
-------

- Issue #1276: Added temporary aliases for CJK Mac encodings to resolve
  a build problem on MacOS with CJK locales.  It adds four temporary
  mappings to existing legacy codecs that are virtually compatible
  with Mac encodings.  They will be replaced by codecs correctly
  implemented in 3.1.

- Issue #3614: Corrected a typo in xmlrpc.client, leading to a NameError
  "global name 'header' is not defined".

- Issue #2834: update the regular expression library to match the unicode
  standards of py3k. In other words, mixing bytes and unicode strings
  (be it as pattern, search string or replacement string) raises a TypeError.
  Moreover, the re.UNICODE flag is enabled automatically for unicode patterns,
  and can be disabled by specifying a new re.ASCII flag; as for bytes
  patterns, ASCII matching is the only option and trying to specify re.UNICODE
  for such patterns raises a ValueError.

- Issue #3300: make urllib.parse.[un]quote() default to UTF-8.
  Code contributed by Matt Giuca.  quote() now encodes the input
  before quoting, unquote() decodes after unquoting.  There are
  new arguments to change the encoding and errors settings.
  There are also new APIs to skip the encode/decode steps.
  [un]quote_plus() are also affected.

- Issue #2235: numbers.Number now blocks inheritance of the default id()
  based hash because that hash mechanism is not correct for numeric types.
  All concrete numeric types that inherit from Number (rather than just
  registering with it) must explicitly provide a hash implementation in
  order for their instances to be hashable.

- Issue #2676: in the email package, content-type parsing was hanging on
  pathological input because of quadratic or exponential behaviour of a
  regular expression.

- Issue #3476: binary buffered reading through the new "io" library is now
  thread-safe.

- Issue #1342811: Fix leak in Tkinter.Menu.delete. Commands associated to
  menu entries were not deleted.

- Remove the TarFileCompat class from tarfile.py.

- Issue #2491: os.fdopen is now almost an alias for the built-in open(), and
  accepts the same parameters.  It just checks that its first argument is an
  integer.

- Issue #3394: zipfile.writestr sets external attributes when passed a
  file name rather than a ZipInfo instance, so files are extracted with
  mode 0600 rather than 000 under Unix.

- Issue #2523: Fix quadratic behaviour when read()ing a binary file without
  asking for a specific length.

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

- Bug #3542: Support Unicode strings in _msi module.

What's new in Python 3.0b2?
===========================

*Release date: 17-Jul-2008*

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

- Issue #3008: the float type has a new instance method 'float.hex'
  and a new class method 'float.fromhex' to convert floating-point
  numbers to and from hexadecimal strings, respectively.

- Issue #3083: Add alternate (#) formatting for bin, oct, hex output
  for str.format().  This adds the prefix 0b, 0o, or 0x, respectively.

- Issue #3280: like chr(), the "%c" format now accepts unicode code points
  beyond the Basic Multilingual Plane (above 0xffff) on all configurations. On
  "narrow Unicode" builds, the result is a string of 2 code units, forming a
  UTF-16 surrogate pair.

- Issue #3282: str.isprintable() should return False for undefined
  Unicode characters.

- Issue #3236: Return small longs from PyLong_FromString.

- Exception tracebacks now support exception chaining.

Library
-------

- Removed the sunaudio module. Use sunau instead.

- Issue #3554: ctypes.string_at and ctypes.wstring_at did call Python
  api functions without holding the GIL, which could lead to a fatal
  error when they failed.

- Issue #799428: Fix Tkinter.Misc._nametowidget to unwrap Tcl command objects.

- Removed "ast" function aliases from the parser module.

- Issue #3313: Fixed a crash when a failed dlopen() call does not set
  a valid dlerror() message.

- Issue #3258: Fixed a crash when a ctypes POINTER type to an
  incomplete structure was created.

- Issue #2683: Fix inconsistency in subprocess.Popen.communicate(): the
  argument now must be a bytes object in any case.

- Issue #3145: help("modules whatever") failed when trying to load the source
  code of every single module of the standard library, including invalid files
  used in the test suite.

- The gettext library now consistently uses Unicode strings for message ids
  and message strings, and ``ugettext()`` and the like don't exist anymore.

- The traceback module has been expanded to handle chained exceptions.

C API
-----

- Issue #3247: the function Py_FindMethod was removed. Modern types should
  use the tp_methods slot instead.

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

- The Mac/Demos directory has been removed.

- All of the Mac scripts have been removed (including BuildApplet.py).


What's new in Python 3.0b1?
===========================

*Release date: 18-Jun-2008*

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

- Issue #3211: warnings.warn_explicit() did not guard against its 'registry'
  argument being anything other than a dict or None. Also fixed a bug in error
  handling when 'message' and 'category' were both set to None, triggering a
  bus error.

- Issue #3100: Corrected a crash on deallocation of a subclassed weakref which
  holds the last (strong) reference to its referent.

- Issue #2630: implement PEP 3138. repr() now returns printable
  Unicode characters unescaped, to get an ASCII-only representation
  of an object use ascii().

- Issue #1342: On windows, Python could not start when installed in a
  directory with non-ascii characters.

- Implement PEP 3121: new module initialization and finalization API.

- Removed the already-defunct ``-t`` option.

- Issue #2957: Corrected a ValueError "recursion limit exceeded", when
  unmarshalling many code objects, which happens when importing a
  large .pyc file (~1000 functions).

- Issue #2963: fix merging oversight that disabled method cache for
  all types.

- Issue #2964: fix a missing INCREF in instancemethod_descr_get.

- Issue #2895: Don't crash when given bytes objects as keyword names.

- Issue #2798: When parsing arguments with PyArg_ParseTuple, the "s"
  code now allows any unicode string and returns a utf-8 encoded
  buffer, just like the "s#" code already does.  The "z" code was
  corrected as well.

- Issue #2863: generators now have a ``gen.__name__`` attribute that
  equals ``gen.gi_code.co_name``, like ``func.__name___`` that equals
  ``func.func_code.co_name``.  The repr() of a generator now also
  contains this name.

- Issue #2831: enumerate() now has a ``start`` argument.

- Issue #2801: fix bug in the float.is_integer method where a
  ValueError was sometimes incorrectly raised.

- The ``--with-toolbox-glue`` option (and the associated
  pymactoolbox.h) have been removed.

- Issue #2196: hasattr() now lets exceptions which do not inherit
  Exception (KeyboardInterrupt, and SystemExit) propagate instead of
  ignoring them.

- #3021 Exception reraising sematics have been significantly improved.  However,
  f_exc_type, f_exc_value, and f_exc_traceback cannot be accessed from Python
  code anymore.

- Three of PyNumberMethods' members, nb_coerce, nb_hex, and nb_oct, have been
  removed.

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

- Renamed ``_winreg`` module to ``winreg``.

- Support os.O_ASYNC and fcntl.FASYNC if the constants exist on the
  platform.

- Support for Windows 9x has been removed from the winsound module.

- Issue #2870: cmathmodule.c compile error.

Library
-------

- The methods ``is_in_tuple()``, ``is_vararg()``, and ``is_keywordarg()`` of
  symtable.Symbol have been removed.

- Patch #3133: http.server.CGIHTTPRequestHandler did not work on windows.

- a new ``urllib`` package was created.  It consists of code from
  ``urllib``, ``urllib2``, ``urlparse``, and ``robotparser``.  The old
  modules have all been removed.  The new package has five submodules:
  ``urllib.parse``, ``urllib.request``, ``urllib.response``,
  ``urllib.error``, and ``urllib.robotparser``.  The
  ``urllib.request.urlopen()`` function uses the url opener from
  ``urllib2``.  (Note that the unittests have not been renamed for the
  beta, but they will be renamed in the future.)

- rfc822 has been removed in favor of the email package.

- mimetools has been removed in favor of the email package.

- Patch #2849: Remove use of rfc822 module from standard library.

- Added C optimized implementation of io.StringIO.

- The ``pickle`` module is now automatically use an optimized C
  implementation of Pickler and Unpickler when available. The
  ``cPickle`` module is no longer needed.

- Removed the ``htmllib`` and ``sgmllib`` modules.

- The deprecated ``SmartCookie`` and ``SimpleCookie`` classes have
  been removed from ``http.cookies``.

- The ``commands`` module has been removed. Its getoutput() and
  getstatusoutput() functions have been moved to the ``subprocess`` module.

- The ``http`` package was created; it contains the old ``httplib``
  as ``http.client``, ``Cookie`` as ``http.cookies``, ``cookielib``
  as ``http.cookiejar``, and the content of the three ``HTTPServer``
  modules as ``http.server``.

- The ``xmlrpc`` package was created; it contains the old
  ``xmlrpclib`` module as ``xmlrpc.client`` and the content of
  the old ``SimpleXMLRPCServer`` and ``DocXMLRPCServer`` modules
  as ``xmlrpc.server``.

- The ``dbm`` package was created, containing the old modules
  ``anydbm`` and ``whichdb`` in its ``__init__.py``, and having
  ``dbm.gnu`` (was ``gdbm``), ``dbm.bsd`` (was ``dbhash``),
  ``dbm.ndbm`` (was ``dbm``) and ``dbm.dumb`` (was ``dumbdbm``)
  as submodules.

- The ``repr`` module has been renamed to ``reprlib``.

- The ``statvfs`` module has been removed.

- Issue #1713041: fix pprint's handling of maximum depth.

- Issue #2250: Exceptions raised during evaluation of names in
  rlcompleter's ``Completer.complete()`` method are now caught and
  ignored.

- Patch #2659: Added ``break_on_hyphens`` option to textwrap's
  ``TextWrapper`` class.

- Issue #2487: change the semantics of math.ldexp(x, n) when n is too
  large to fit in a C long.  ldexp(x, n) now returns a zero (with
  suitable sign) if n is large and negative; previously, it raised
  OverflowError.

- The ``ConfigParser`` module has been renamed to ``configparser``.

- Issue #2865: webbrowser.open() works again in a KDE environment.

- The ``multifile`` module has been removed.

- The ``SocketServer`` module has been renamed to ``socketserver``.

- Fixed the ``__all__`` setting on ``collections`` to include
  ``UserList`` and ``UserString``.

- The sre module has been removed.

- The Queue module has been renamed to queue.

- The copy_reg module has been renamed to copyreg.

- The mhlib module has been removed.

- The ihooks module has been removed.

- The fpformat module has been removed.

- The dircache module has been removed.

- The Canvas module has been removed.

- The Decimal module gained the magic methods __round__, __ceil__,
  __floor__ and __trunc__, to give support for round, math.ceil,
  math.floor and math.trunc.

- The user module has been removed.

- The mutex module has been removed.

- The imputil module has been removed.

- os.path.walk has been removed in favor of os.walk.

- pdb gained the "until" command.

- The test.test_support module has been renamed to test.support.

- The threading module API was renamed to be PEP 8 compliant.  The
  old names are still present, but will be removed in the near future.

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

- The bgen tool has been removed.

Build
-----


What's New in Python 3.0a5?
===========================

*Release date: 08-May-2008*

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

- Fixed misbehaviour of PyLong_FromSsize_t on systems where
  sizeof(size_t) > sizeof(long).

- Issue #2221: Corrected a SystemError "error return without exception
  set", when the code executed by exec() raises an exception, and
  sys.stdout.flush() also raises an error.

- Bug #2565: The repr() of type objects now calls them 'class', not
  'type' - whether they are builtin types or not.

- The command line processing was converted to pass Unicode strings
  through as unmodified as possible; as a consequence, the C API
  related to command line arguments was changed to use wchar_t.

- All backslashes in raw strings are interpreted literally.  This
  means that '\u' and '\U' escapes are not treated specially.

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

Library
-------

- ctypes objects now support the PEP3118 buffer interface.

- Issue #2682: ctypes callback functions now longer contain a cyclic
  reference to themselves.

- Issue #2058: Remove the buf attribute and add __slots__ to the
  TarInfo class in order to reduce tarfile's memory usage.

- Bug #2606: Avoid calling .sort() on a dict_keys object.

- The bundled libffi copy is now in sync with the recently released
  libffi3.0.5 version, apart from some small changes to
  Modules/_ctypes/libffi/configure.ac.

Build
-----

- Issue #1496032: On alpha, use -mieee when gcc is the compiler.

- "make install" is now an alias for "make altinstall", to prevent
  accidentally overwriting a Python 2.x installation.  Use "make
  fullinstall" to force Python 3.0 to be installed as "python".

- Issue #2544: On HP-UX systems, use 'gcc -shared' for linking when
  gcc is used as compiler.


What's New in Python 3.0a4?
===========================

*Release date: 02-Apr-2008*

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

- Bug #2301: Don't try decoding the source code into the original
  encoding for syntax errors.

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

- The dl module was removed, use the ctypes module instead.

- Use wchar_t functions in _locale module.

Library
-------

- The class distutils.commands.build_py.build_py_2to3 can be used as a
  build_py replacement to automatically run 2to3 on modules that are
  going to be installed.

- A new pickle protocol (protocol 3) is added with explicit support
  for bytes.  This is the default protocol.  It intentionally cannot
  be unpickled by Python 2.x.

- When a pickle written by Python 2.x contains an (8-bit) str
  instance, this is now decoded to a (Unicode) str instance.  The
  encoding used to do this defaults to ASCII, but can be overridden
  via two new keyword arguments to the Unpickler class.  Previously
  this would create bytes instances, which is usually wrong: str
  instances are often used to pickle attribute names etc., and text is
  more common than binary data anyway.

- Default to ASCII as the locale.getpreferredencoding, if the POSIX
  system doesn't support CODESET and LANG isn't set or doesn't allow
  deduction of an encoding.

- Issue #1202: zlib.crc32 and zlib.adler32 now return an unsigned
  value.

- Issue #719888: Updated tokenize to use a bytes API.  generate_tokens
  has been renamed tokenize and now works with bytes rather than
  strings.  A new detect_encoding function has been added for
  determining source file encoding according to PEP-0263.  Token
  sequences returned by tokenize always start with an ENCODING token
  which specifies the encoding used to decode the file.  This token is
  used to encode the output of untokenize back to bytes.


What's New in Python 3.0a3?
===========================

*Release date: 29-Feb-2008*

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

- Issue #2282: io.TextIOWrapper was not overriding seekable() from
  io.IOBase.

- Issue #2115: Important speedup in setting __slot__ attributes.  Also
  prevent a possible crash: an Abstract Base Class would try to access
  a slot on a registered virtual subclass.

- Fixed repr() and str() of complex numbers with infinity or nan as
  real or imaginary part.

- Clear all free list during a gc.collect() of the highest generation
  in order to allow pymalloc to free more arenas. Python may give back
  memory to the OS earlier.

- Issue #2045: Fix an infinite recursion triggered when printing a
  subclass of collections.defaultdict, if its default_factory is set
  to a bound method.

- Fixed a minor memory leak in dictobject.c. The content of the free
  list was not freed on interpreter shutdown.

- Limit free list of method and builtin function objects to 256
  entries each.

- Patch #1953: Added ``sys._compact_freelists()`` and the C API
  functions ``PyInt_CompactFreeList`` and ``PyFloat_CompactFreeList``
  to compact the internal free lists of pre-allocted ints and floats.

- Bug #1983: Fixed return type of fork(), fork1() and forkpty() calls.
  Python expected the return type int but the fork familie returns
  pi_t.

- Issue #1678380: Fix a bug that identifies 0j and -0j when they
  appear in the same code unit.

- Issue #2025: Added tuple.count() and tuple.index() methods to comply
  with the collections.Sequence API.

- Fixed multiple reinitialization of the Python interpreter. The small
  int list in longobject.c has caused a seg fault during the third
  finalization.

- Issue #1973: bytes.fromhex('') raised SystemError.

- Issue #1771: remove cmp parameter from sorted() and list.sort().

- Issue #1969: split and rsplit in bytearray are inconsistent.

- map() no longer accepts None for the first argument.  Use zip()
  instead.

- Issue #1769: Now int("- 1") is not allowed any more.

- Object/longobject.c: long(float('nan')) raises an OverflowError
  instead of returning 0.

- Issue #1762972: __file__ points to the source file instead of the
  pyc/pyo file if the py file exists.

- Issue #1393: object_richcompare() returns NotImplemented instead of
  False if the objects aren't equal, to give the other side a chance.

- Issue #1692: Interpreter was not displaying location of SyntaxError.

- Improve some exception messages when Windows fails to load an
  extension module.  Now we get for example '%1 is not a valid Win32
  application' instead of 'error code 193'.  Also use Unicode strings
  to deal with non-English locales.

- Issue #1587: Added instancemethod wrapper for PyCFunctions. The
  Python C API has gained a new type *PyInstanceMethod_Type* and the
  functions *PyInstanceMethod_Check(o)*, *PyInstanceMethod_New(func)*
  and *PyInstanceMethod_Function(im)*.

- Constants gc.DEBUG_OBJECT and gc.DEBUG_INSTANCE have been removed
  from the gc module; gc.DEBUG_COLLECTABLE or gc.DEBUG_UNCOLLECTABLE
  are now enough to print the corresponding list of objects considered
  by the garbage collector.

- Issue #1573: Improper use of the keyword-only syntax makes the
  parser crash.

- Issue #1564: The set implementation should special-case PyUnicode
  instead of PyString.

- Patch #1031213: Decode source line in SyntaxErrors back to its
  original source encoding.

- inspect.getsource() includes the decorators again.

- Bug #1713: posixpath.ismount() claims symlink to a mountpoint is a
  mountpoint.

- Fix utf-8-sig incremental decoder, which didn't recognise a BOM when
  the first chunk fed to the decoder started with a BOM, but was
  longer than 3 bytes.

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

- Code for itertools ifilter(), imap(), and izip() moved to bultins
  and renamed to filter(), map(), and zip().  Also, renamed
  izip_longest() to zip_longest() and ifilterfalse() to filterfalse().

- Issue #1762972: Readded the reload() function as imp.reload().

- Bug #2111: mmap segfaults when trying to write a block opened with
  PROT_READ.

- Issue #2063: correct order of utime and stime in os.times() result
  on Windows.

Library
-------

- Weakref dictionaries now inherit from MutableMapping.

- Created new UserDict class in collections module.  This one inherits
  from and complies with the MutableMapping ABC.  Also, moved
  UserString and UserList to the collections module.  The
  MutableUserString class was removed.

- Removed UserDict.DictMixin.  Replaced all its uses with
  collections.MutableMapping.

- Issue #1703: getpass() should flush after writing prompt.

- Issue #1585: IDLE uses non-existent xrange() function.

- Issue #1578: Problems in win_getpass.

Build
-----

- Renamed --enable-unicode configure flag to --with-wide-unicode,
  since Unicode strings can't be disabled anymore.

C API
-----

- Issue #1629: Renamed Py_Size, Py_Type and Py_Refcnt to Py_SIZE,
  Py_TYPE and Py_REFCNT.

- New API PyImport_ImportModuleNoBlock(), works like
  PyImport_ImportModule() but won't block on the import lock
  (returning an error instead).


What's New in Python 3.0a2?
===========================

*Release date: 07-Dec-2007*

(Note: this list is incomplete.)

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

- str8 now has the same construction signature as bytes.

- Comparisons between str and str8 now return False/True for ==/!=.
  sqlite3 returns str8 when recreating on object from it's __conform__
  value.  The struct module returns str8 for all string-related
  formats.  This was true before this change, but becomes more
  apparent thanks to string comparisons always being False.

- Replaced `PyFile_FromFile()` with `PyFile_FromFd(fd, name. mode,
  buffer, encoding, newline)`.

- Fixed `imp.find_module()` to obey the -*- coding: -*- header.

- Changed `__file__` and `co_filename` to unicode. The path names are decoded
  with `Py_FileSystemDefaultEncoding` and a new API method
  `PyUnicode_DecodeFSDefault(char*)` was added.

- io.open() and _fileio.FileIO have grown a new argument closefd. A
  false value disables the closing of the file descriptor.

- Added a new option -b to issues warnings (-bb for errors) about
  certain operations between bytes/buffer and str like str(b'') and
  comparison.

- The standards streams sys.stdin, stdout and stderr may be None when
  the when the C runtime library returns an invalid file descriptor
  for the streams (fileno(stdin) < 0). For now this happens only for
  Windows GUI apps and scripts started with `pythonw.exe`.

- Added PCbuild9 directory for VS 2008.

- Renamed structmember.h WRITE_RESTRICTED to PY_WRITE_RESTRICTED to
  work around a name clash with VS 2008 on Windows.

- Unbound methods are gone for good. ClassObject.method returns an
  ordinary function object, instance.method still returns a bound
  method object.  The API of bound methods is cleaned up, too. The
  im_class attribute is removed and im_func + im_self are renamed to
  __func__ and __self__. The factory PyMethod_New takes only func and
  instance as argument.

- intobject.h is no longer included by Python.h. The remains were
  moved to longobject.h. It still exists to define several aliases
  from PyInt_ to PyLong_ functions.

- Removed sys.maxint, use sys.maxsize instead.

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

- The `hotshot` profiler has been removed; use `cProfile` instead.

Library
-------

- When loading an external file using testfile(), the passed-in
  encoding argument was being ignored if __loader__ is defined and
  forcing the source to be UTF-8.

- The methods `os.tmpnam()`, `os.tempnam()` and `os.tmpfile()` have
  been removed in favor of the tempfile module.

- Removed the 'new' module.

- Removed all types from the 'types' module that are easily accessable
  through builtins.


What's New in Python 3.0a1?
==========================

*Release date: 31-Aug-2007*

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

- PEP 3131: Support non-ASCII identifiers.

- PEP 3120: Change default encoding to UTF-8.

- PEP 3123: Use proper C inheritance for PyObject.

- Removed the __oct__ and __hex__ special methods and added a bin()
  builtin function.

- PEP 3127: octal literals now start with "0o". Old-style octal
  literals are invalid. There are binary literals with a prefix of
  "0b".  This also affects int(x, 0).

- None, True, False are now keywords.

- PEP 3119: isinstance() and issubclass() can be overridden.

- Remove BaseException.message.

- Remove tuple parameter unpacking (PEP 3113).

- Remove the f_restricted attribute from frames.  This naturally leads
  to the removal of PyEval_GetRestricted() and PyFrame_IsRestricted().

- PEP 3132 was accepted. That means that you can do ``a, *b =
  range(5)`` to assign 0 to a and [1, 2, 3, 4] to b.

- range() now returns an iterator rather than a list.  Floats are not
  allowed.  xrange() is no longer defined.

- Patch #1660500: hide iteration variable in list comps, add set comps
  and use common code to handle compilation of iterative expressions.

- By default, != returns the opposite of ==, unless the latter returns
  NotImplemented.

- Patch #1680961: sys.exitfunc has been removed and replaced with a
  private C-level API.

- PEP 3115: new metaclasses: the metaclass is now specified as a
  keyword arg in the class statement, which can now use the full
  syntax of a parameter list. Also, the metaclass can implement a
  __prepare__ function which will be called to create the dictionary
  for the new class namespace.

- The long-deprecated argument "pend" of PyFloat_FromString() has been
  removed.

- The dir() function has been extended to call the __dir__() method on
  its argument, if it exists. If not, it will work like before. This
  allows customizing the output of dir() in the presence of a
  __getattr__().

- Removed support for __members__ and __methods__.

- Removed indexing/slicing on BaseException.

- input() became raw_input(): the name input() now implements the
  functionality formerly known as raw_input(); the name raw_input() is
  no longer defined.

- Classes listed in an 'except' clause must inherit from
  BaseException.

- PEP 3106: dict.iterkeys(), .iteritems(), .itervalues() are now gone;
  and .keys(), .items(), .values() return dict views, which behave
  like sets.

- PEP 3105: print is now a function.  Also (not in the PEP) the
  'softspace' attribute of files is now gone (since print() doesn't
  use it).  A side effect of this change is that you can get
  incomplete output lines in interactive sessions:

  >>> print(42, end="")
  42>>>

  We may be able to fix this after the I/O library rewrite.

- PEP 3102: keyword-only arguments.

- Int/Long unification is complete.  The 'long' built-in type and
  literals with trailing 'L' or 'l' have been removed.  Performance
  may be sub-optimal (haven't really benchmarked).

- 'except E, V' must now be spelled as 'except E as V' and deletes V
  at the end of the except clause; V must be a simple name.

- Added function annotations per PEP 3107.

- Added nonlocal declaration from PEP 3104:

  >>> def f(x):
  ...     def inc():
  ...         nonlocal x
  ...         x += 1
  ...         return x
  ...     return inc
  ...
  >>> inc = f(0)
  >>> inc()
  1
  >>> inc()
  2

- Moved intern() to sys.intern().

- exec is now a function.

- Renamed nb_nonzero to nb_bool and __nonzero__ to __bool__.

- Classic classes are a thing of the past.  All classes are new style.

- Exceptions *must* derive from BaseException.

- Integer division always returns a float.  The -Q option is no more.
  All the following are gone:
   * PyNumber_Divide and PyNumber_InPlaceDivide
   * __div__, __rdiv__, and __idiv__
   * nb_divide, nb_inplace_divide
   * operator.div, operator.idiv, operator.__div__, operator.__idiv__
  (Only __truediv__ and __floordiv__ remain, not sure how to handle them
   if we want to re-use __div__ and friends.  If we do, it will make
   it harder to write code for both 2.x and 3.x.)

- 'as' and 'with' are keywords.

- Absolute import is the default behavior for 'import foo' etc.

- Removed support for syntax: backticks (ie, `x`), <>.

- Removed these Python builtins: apply(), callable(), coerce(),
  execfile(), file(), reduce(), reload().

- Removed these Python methods: {}.has_key.

- Removed these opcodes: BINARY_DIVIDE, INPLACE_DIVIDE, UNARY_CONVERT.

- Remove C API support for restricted execution.

- zip(), map() and filter() now return iterators, behaving like their
  itertools counterparts.  This also affect map()'s behavior on
  sequences of unequal length -- it now stops after the shortest one
  is exhausted.

- Additions: set literals, set comprehensions, ellipsis literal.

- Added class decorators per PEP 3129.


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

- Removed the imageop module.  Obsolete long with its unit tests
  becoming useless from the removal of rgbimg and imgfile.

- Removed these attributes from the operator module: div, idiv,
  __div__, __idiv__, isCallable, sequenceIncludes.

- Removed these attributes from the sys module: exc_clear(), exc_type,
  exc_value, exc_traceback.


Library
-------

- Removed the compiler package.  Use of the _ast module and (an
  eventual) AST -> bytecode mechanism.

- Removed these modules: audiodev, Bastion, bsddb185, exceptions,
  linuxaudiodev, md5, MimeWriter, mimify, popen2, rexec, sets, sha,
  stringold, strop, sunaudiodev, timing, xmllib.

- Moved the toaiff module to Tools/Demos.

- Removed obsolete IRIX modules: al/AL, cd/CD, cddb, cdplayer, cl/CL,
  DEVICE, ERRNO, FILE, fl/FL, flp, fm, GET, gl/GL, GLWS, IN, imgfile,
  IOCTL, jpeg, panel, panelparser, readcd, sgi, sv/SV, torgb, WAIT.

- Removed obsolete functions: commands.getstatus(), os.popen*().

- Removed functions in the string module that are also string methods;
  Remove string.{letters, lowercase, uppercase}.

- Removed support for long obsolete platforms: plat-aix3, plat-irix5.

- Removed xmlrpclib.SlowParser.  It was based on xmllib.

- Patch #1680961: atexit has been reimplemented in C.

- Add new codecs for UTF-32, UTF-32-LE and UTF-32-BE.

Build
-----

C API
-----

- Removed these Python slots: __coerce__, __div__, __idiv__, __rdiv__.

- Removed these C APIs: PyNumber_Coerce(), PyNumber_CoerceEx(),
  PyMember_Get, PyMember_Set.

- Removed these C slots/fields: nb_divide, nb_inplace_divide.

- Removed these macros: staticforward, statichere, PyArg_GetInt,
  PyArg_NoArgs, _PyObject_Del.

- Removed these typedefs: intargfunc, intintargfunc, intobjargproc,
  intintobjargproc, getreadbufferproc, getwritebufferproc,
  getsegcountproc, getcharbufferproc, memberlist.

Tests
-----

- Removed test.testall as test.regrtest replaces it.

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

Mac
---

- The cfmfile module was removed.

Platforms
---------

- Support for BeOS and AtheOS was removed (according to PEP 11).

- Support for RiscOS, Irix, Tru64 was removed (alledgedly).

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


**(For information about older versions, consult the HISTORY file.)**