+++++++++++

Python News

+++++++++++

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

What's New in Python 2.4 alpha 1?

=================================

*Release date: DD-MMM-YYYY*

Core and builtins

-----------------

- Implemented a newcode opcode, LIST_APPEND, that simplifies

the generated bytecode for list comprehensions and further

improves their performance (about 35%).

- Implemented rich comparisons for floats, which seems to make

comparisons involving NaNs somewhat less surprising when the

underlying C compiler actually implements C99 semantics.

- Optimized list.extend() to save memory and no longer create

intermediate sequences. Also, extend() now pre-allocates the

needed memory whenever the length of the iterable is known in

advance -- this halves the time to extend the list.

- Optimized list resize operations to make fewer calls to the system

realloc(). Significantly speeds up list appends, list pops,

list comprehensions, and the list contructor (when the input iterable

length is not known).

- Changed the internal list over-allocation scheme. For larger lists,

overallocation ranged between 3% and 25%. Now, it is a constant 12%.

For smaller lists (n<8), overallocation was upto eight elements. Now,

the overallocation is no more than three elements -- this improves space

utilization for applications that have large numbers of small lists.

- The dict.update() method now accepts all the same argument forms

as the dict() constructor. This now includes item lists and/or

keyword arguments.

- Support for arbitrary objects supporting the read-only buffer

interface as the co_code field of code objects (something that was

only possible to create from C code) has been removed.

- Made omitted callback and None equivalent for weakref.ref() and

weakref.proxy(); the None case wasn't handled correctly in all

cases.

- Fixed problem where PyWeakref_NewRef() and PyWeakref_NewProxy()

assumed that initial existing entries in an object's weakref list

would not be removed while allocating a new weakref object. Since

GC could be invoked at that time, however, that assumption was

invalid. In a truly obscure case of GC being triggered during

creation for a new weakref object for an referent which already

has a weakref without a callback which is only referenced from

cyclic trash, a memory error can occur. This consistently created a

segfault in a debug build, but provided less predictable behavior in

a release build.

- input() builtin function now respects compiler flags such as

__future__ statements. SF patch 876178.

- Removed PendingDeprecationWarning from apply(). apply() remains

deprecated, but the nuisance warning will not be issued.

- At Python shutdown time (Py_Finalize()), 2.3 called cyclic garbage

collection twice, both before and after tearing down modules. The

call after tearing down modules has been disabled, because too much

of Python has been torn down then for __del__ methods and weakref

callbacks to execute sanely. The most common symptom was a sequence

of uninformative messages on stderr when Python shut down, produced

by threads trying to raise exceptions, but unable to report the nature

of their problems because too much of the sys module had already been

destroyed.

- Removed FutureWarnings related to hex/oct literals and conversions

and left shifts. (Thanks to Kalle Svensson for SF patch 849227.)

This addresses most of the remaining semantic changes promised by

PEP 237, except for repr() of a long, which still shows the trailing

'L'. The PEP appears to promise warnings for operations that

changed semantics compared to Python 2.3, but this is not

implemented; we've suffered through enough warnings related to

hex/oct literals and I think it's best to be silent now.

- For str and unicode objects, the ljust(), center(), and rjust()

methods now accept an optional argument specifying a fill

character other than a space.

- When method objects have an attribute that can be satisfied either

by the function object or by the method object, the function

object's attribute usually wins. Christian Tismer pointed out that

that this is really a mistake, because this only happens for special

methods (like __reduce__) where the method object's version is

really more appropriate than the function's attribute. So from now

on, all method attributes will have precedence over function

attributes with the same name.

- Critical bugfix, for SF bug 839548: if a weakref with a callback,

its callback, and its weakly referenced object, all became part of

cyclic garbage during a single run of garbage collection, the order

in which they were torn down was unpredictable. It was possible for

the callback to see partially-torn-down objects, leading to immediate

segfaults, or, if the callback resurrected garbage objects, to

resurrect insane objects that caused segfaults (or other surprises)

later. In one sense this wasn't surprising, because Python's cyclic gc

had no knowledge of Python's weakref objects. It does now. When

weakrefs with callbacks become part of cyclic garbage now, those

weakrefs are cleared first. The callbacks don't trigger then,

preventing the problems. If you need callbacks to trigger, then just

as when cyclic gc is not involved, you need to write your code so

that weakref objects outlive the objects they weakly reference.

- Critical bugfix, for SF bug 840829: if cyclic garbage collection

happened to occur during a weakref callback for a new-style class

instance, subtle memory corruption was the result (in a release build;

in a debug build, a segfault occurred reliably very soon after).

This has been repaired.

- Compiler flags set in PYTHONSTARTUP are now active in __main__.

- Added two builtin types, set() and frozenset().

- Added a reversed() builtin function that returns a reverse iterator

over a sequence.

- Added a sorted() builtin function that returns a new sorted list

from any iterable.

- CObjects are now mutable (on the C level) through PyCObject_SetVoidPtr.

- list.sort() now supports three keyword arguments: cmp, key, and reverse.

The key argument can be a function of one argument that extracts a

comparison key from the original record: mylist.sort(key=str.lower).

The reverse argument is a boolean value and if True will change the

sort order as if the comparison arguments were reversed. In addition,

the documentation has been amended to provide a guarantee that all sorts

starting with Py2.3 are guaranteed to be stable (the relative order of

records with equal keys is unchanged).

- Added test whether wchar_t is signed or not. A signed wchar_t is not

usable as internal unicode type base for Py_UNICODE since the

unicode implementation assumes an unsigned type.

- Fixed a bug in the cache of length-one Unicode strings that could

lead to a seg fault. The specific problem occurred when an earlier,

non-fatal error left an uninitialized Unicode object in the

freelist.

- The % formatting operator now supports '%F' which is equivalent to

'%f'. This has always been documented but never implemented.

- complex(obj) could leak a little memory if obj wasn't a string or

number.

- zip() with no arguments now returns an empty list instead of raising

a TypeError exception.

- obj.__contains__() now returns True/False instead of 1/0. SF patch

820195.

- Python no longer tries to be smart about recursive comparisons.

When comparing containers with cyclic references to themselves it

will now just hit the recursion limit. See SF patch 825639.

- str and unicode builtin types now have rsplit() method that is

same as split() except that it scans the string from the end

working towards the beginning. See SF feature request 801847.

- Fixed a bug in object.__reduce_ex__ when using protocol 2. Failure

to clear the error when attempts to get the __getstate__ attribute

fail caused intermittent errors and odd behavior.

- buffer objects based on other objects no longer cache a pointer to

the data and the data length. Instead, the appropriate tp_as_buffer

method is called as necessary.

Extension modules

-----------------

- array objects now support the copy module. Also, their resizing

scheme has been updated the same as for list objects. The improves

the performance (speed and memory usage) of append() operations.

Also, array.extend() now accepts any iterable argument for repeated

appends without needing to create another temporary array.

- cStringIO.writelines() now accepts any iterable argument and writes

the lines one at a time rather than joining them and writing once.

Made a parallel change to StringIO.writelines(). Saves memory and

makes suitable for use with generator expressions.

- time.strftime() now checks that the values in its time tuple argument

are within the proper boundaries to prevent possible crashes from the

platform's C library implementation of strftime(). Can possibly

break code that uses values outside the range that didn't cause

problems previously (such as sitting day of year to 0). Fixes bug

#897625.

- The socket module now supports Bluetooth sockets, if the

system has <bluetooth/bluetooth.h>

- Added a collections module containing a new datatype, deque(),

offering high-performance, thread-safe, memory friendly appends

and pops on either side of the deque.

- Several modules now take advantage of collections.deque() for

improved performance: Queue, mutex, shlex, threading, and pydoc.

- The operator module has two new functions, attrgetter() and

itemgetter() which are useful for creating fast data extractor

functions for map(), list.sort(), itertools.groupby(), and

other functions that expect a function argument.

- socket.SHUT_{RD,WR,RDWR} was added.

- os.getsid was added.

- The pwd module incorrectly advertised its struct type as

struct_pwent; this has been renamed to struct_passwd. (The old name

is still supported for backwards compatibility.)

- The xml.parsers.expat module now provides Expat 1.95.7.

- socket.IPPROTO_IPV6 was added.

- readline.clear_history was added.

- select.select() now accepts sequences for its first three arguments.

- cStringIO now supports the f.closed attribute.

- The signal module now exposes SIGRTMIN and SIGRTMAX (if available).

- curses module now supports use_default_colors(). [patch #739124]

- Bug #811028: ncurses.h breakage on FreeBSD/MacOS X

- Bug #814613: INET_ADDRSTRLEN fix needed for all compilers on SGI

- Implemented non-recursive SRE matching scheme (#757624).

- Implemented (?(id/name)yes|no) support in SRE (#572936).

- random.seed() with no arguments or None uses time.time() as a default

seed. Modified to match Py2.2 behavior and use fractional seconds so

that successive runs are more likely to produce different sequences.

- random.Random has a new method, getrandbits(k), which returns an int

with k random bits. This method is now an optional part of the API

for user defined generators. Any generator that defines genrandbits()

can now use randrange() for ranges with a length >= 2**53. Formerly,

randrange would return only even numbers for ranges that large (see

SF bug #812202). Generators that do not define genrandbits() now

issue a warning when randrange() is called with a range that large.

- itertools has a new function, groupby() for aggregating iterables

into groups sharing the same key (as determined by a key function).

It offers some of functionality of SQL's groupby keyword and of

the Unix uniq filter.

- itertools now has a new function, tee() which produces two independent

iterators from a single iterable.

- itertools.izip() with no arguments now returns an empty iterator instead

of raising a TypeError exception.

- Fixed #853061: allow BZ2Compressor.compress() to receive an empty string

as parameter.

Library

-------

- Support non-anonymous ftp URLs in urllib2.

- The encodings package will now applies codec name aliases

first before starting to try the import of the codec module.

This simplifies overriding built-in codecs with external

packages, e.g. the included CJK codecs with the JapaneseCodecs

package, by adjusting the aliases dictionary in encodings.aliases

accordingly.

- base64 now supports RFC 3548 Base16, Base32, and Base64 encoding and

decoding standards.

- urllib2 now supports processors. A processor is a handler that

implements an xxx_request or xxx_response method. These methods are

called for all requests.

- distutils compilers now compile source files in the same order as

they are passed to the compiler.

- pprint.pprint() and pprint.pformat() now have additional parameters

indent, width and depth.

- Patch #750542: pprint now will pretty print subclasses of list, tuple

and dict too, as long as they don't overwrite __repr__().

- Bug #848614: distutils' msvccompiler fails to find the MSVC6

compiler because of incomplete registry entries.

- httplib.HTTP.putrequest now offers to omit the implicit Accept-Encoding.

- Patch #841977: modulefinder didn't find extension modules in packages

- imaplib.IMAP4.thread was added.

- Plugged a minor hole in tempfile.mktemp() due to the use of

os.path.exists(), switched to using os.lstat() directly if possible.

- bisect.py and heapq.py now have underlying C implementations

for better performance

- traceback.format_exc has been added (similar to print_exc but it returns

a string).

- xmlrpclib.MultiCall has been added.

- poplib.POP3_SSL has been added.

- tmpfile.mkstemp now returns an absolute path even if dir is relative.

- urlparse is RFC 2396 compliant.

- The fieldnames argument to the csv module's DictReader constructor is now

optional. If omitted, the first row of the file will be used as the

list of fieldnames.

- encodings.bz2_codec was added for access to bz2 compression

using "a long string".encode('bz2')

- Various improvements to unittest.py, realigned with PyUnit CVS.

- dircache now passes exceptions to the caller, instead of returning

empty lists.

- The bsddb module and dbhash module now support the iterator and

mapping protocols which make them more substitutable for dictionaries

and shelves.

- The csv module's DictReader and DictWriter classes now accept keyword

arguments. This was an omission in the initial implementation.

- The email package handles some RFC 2231 parameters with missing

CHARSET fields better. It also includes a patch to parameter

parsing when semicolons appear inside quotes.

- sets.py now runs under Py2.2. In addition, the argument restrictions

for most set methods (but not the operators) have been relaxed to

allow any iterable.

- _strptime.py now has a behind-the-scenes caching mechanism for the most

recent TimeRE instance used along with the last five unique directive

patterns. The overall module was also made more thread-safe.

- random.cunifvariate() and random.stdgamma() were deprecated in Py2.3

and removed in Py2.4.

- Bug #823328: urllib2.py's HTTP Digest Auth support works again.

- Patch #873597: CJK codecs are imported into rank of default codecs.

Tools/Demos

-----------

- A hotshotmain script was added to the Tools/scripts directory that

makes it easy to run a script under control of the hotshot profiler.

- The db2pickle and pickle2db scripts can now dump/load gdbm files.

- The file order on the command line of the pickle2db script was reversed.

It is now [ picklefile ] dbfile. This provides better symmetry with

db2pickle. The file arguments to both scripts are now source followed by

destination in situations where both files are given.

- The pydoc script will display a link to the module documentation for

modules determined to be part of the core distribution. The documentation

base directory defaults to http://www.python.org/doc/current/lib/ but can

be changed by setting the PYTHONDOCS environment variable.

- texcheck.py now detects double word errors.

- md5sum.py mistakenly opened input files in text mode by default, a

silent and dangerous change from previous releases. It once again

opens input files in binary mode by default. The -t and -b flags

remain for compatibility with the 2.3 release, but -b is the default

now.

- py-electric-colon now works when pending-delete/delete-selection mode is

in effect

- py-help-at-point is no longer bound to the F1 key - it's still bound to

C-c C-h

- Pynche was fixed to not crash when there is no ~/.pynche file and no

-d option was given.

Build

-----

- Support for missing ANSI C header files (limits.h, stddef.h, etc) was

removed.

- Systems requiring the D4, D6 or D7 variants of pthreads are no longer

supported (see PEP 11).

- Universal newline support can no longer be disabled (see PEP 11).

- Support for DGUX, SunOS 4, IRIX 4 and Minix was removed (see PEP 11).

- Support for systems requiring --with-dl-dld or --with-sgi-dl was removed

(see PEP 11).

- Tests for sizeof(char) were removed since ANSI C mandates that

sizeof(char) must be 1.

C API

-----

- Added a new macro, PySequence_Fast_ITEMS, which retrieves a fast sequence's

underlying array of PyObject pointers. Useful for high speed looping.

- Created a new method flag, METH_COEXIST, which causes a method to be loaded

even if already defined by a slot wrapper. This allows a __contains__

method, for example, to co-exist with a defined sq_contains slot. This

is helpful because the PyCFunction can take advantage of optimized calls

whenever METH_O or METH_NOARGS flags are defined.

- Added a new function, PyDict_Contains(d, k) which is like

PySequence_Contains() but is specific to dictionaries and executes

about 10% faster.

- Added three new macros: Py_RETURN_NONE, Py_RETURN_TRUE, and Py_RETURN_FALSE.

Each return the singleton they mention after Py_INCREF()ing them.

- Added a new function, PyTuple_Pack(n, ...) for constructing tuples from a

variable length argument list of Python objects without having to invoke

the more complex machinery of Py_BuildValue(). PyTuple_Pack(3, a, b, c)

is equivalent to Py_BuildValue("(OOO)", a, b, c).

New platforms

-------------

Tests

-----

Windows

-------

- The _winreg module could segfault when reading very large registry

values, due to unchecked alloca() calls (SF bug 851056). The fix is

uses either PyMem_Malloc(n) or PyString_FromStringAndSize(NULL, n),

as appropriate, followed by a size check.

- file.truncate() could misbehave if the file was open for update

(modes r+, rb+, w+, wb+), and the most recent file operation before

the truncate() call was an input operation. SF bug 801631.

Mac

----

What's New in Python 2.3 final?

===============================

*Release date: 29-Jul-2003*

IDLE

----

- Bug 778400: IDLE hangs when selecting "Edit with IDLE" from explorer.

This was unique to Windows, and was fixed by adding an -n switch to

the command the Windows installer creates to execute "Edit with IDLE"

context-menu actions.

- IDLE displays a new message upon startup: some "personal firewall"

kinds of programs (for example, ZoneAlarm) open a dialog of their

own when any program opens a socket. IDLE does use sockets, talking

on the computer's internal loopback interface. This connection is not

visible on any external interface and no data is sent to or received

from the Internet. So, if you get such a dialog when opening IDLE,

asking whether to let pythonw.exe talk to address 127.0.0.1, say yes,

and rest assured no communication external to your machine is taking

place. If you don't allow it, IDLE won't be able to start.

What's New in Python 2.3 release candidate 2?

=============================================

*Release date: 24-Jul-2003*

Core and builtins

-----------------

- It is now possible to import from zipfiles containing additional

data bytes before the zip compatible archive. Zipfiles containing a

comment at the end are still unsupported.

Extension modules

-----------------

- A longstanding bug in the parser module's initialization could cause

fatal internal refcount confusion when the module got initialized more

than once. This has been fixed.

- Fixed memory leak in pyexpat; using the parser's ParseFile() method

with open files that aren't instances of the standard file type

caused an instance of the bound .read() method to be leaked on every

call.

- Fixed some leaks in the locale module.

Library

-------

- Lib/encodings/rot_13.py when used as a script, now more properly

uses the first Python interpreter on your path.

- Removed caching of TimeRE (and thus LocaleTime) in _strptime.py to

fix a locale related bug in the test suite. Although another patch

was needed to actually fix the problem, the cache code was not

restored.

IDLE

----

- Calltips patches.

Build

-----

- For MacOSX, added -mno-fused-madd to BASECFLAGS to fix test_coercion

on Panther (OSX 10.3).

C API

-----

Windows

-------

- The tempfile module could do insane imports on Windows if PYTHONCASEOK

was set, making temp file creation impossible. Repaired.

- Add a patch to workaround pthread_sigmask() bugs in Cygwin.

Mac

---

- Various fixes to pimp.

- Scripts runs with pythonw no longer had full window manager access.

- Don't force boot-disk-only install, for reasons unknown it causes

more problems than it solves.

What's New in Python 2.3 release candidate 1?

=============================================

*Release date: 18-Jul-2003*

Core and builtins

-----------------

- The new function sys.getcheckinterval() returns the last value set

by sys.setcheckinterval().

- Several bugs in the symbol table phase of the compiler have been

fixed. Errors could be lost and compilation could fail without

reporting an error. SF patch 763201.

- The interpreter is now more robust about importing the warnings

module. In an executable generated by freeze or similar programs,

earlier versions of 2.3 would fail if the warnings module could

not be found on the file system. Fixes SF bug 771097.

- A warning about assignments to module attributes that shadow

builtins, present in earlier releases of 2.3, has been removed.

- It is not possible to create subclasses of builtin types like str

and tuple that define an itemsize. Earlier releases of Python 2.3

allowed this by mistake, leading to crashes and other problems.

- The thread_id is now initialized to 0 in a non-thread build. SF bug

770247.

- SF bug 762891: "del p[key]" on proxy object no longer raises SystemError.

Extension modules

-----------------

- weakref.proxy() can now handle "del obj[i]" for proxy objects

defining __delitem__. Formerly, it generated a SystemError.

- SSL no longer crashes the interpreter when the remote side disconnects.

- On Unix the mmap module can again be used to map device files.

- time.strptime now exclusively uses the Python implementation

contained within the _strptime module.

- The print slot of weakref proxy objects was removed, because it was

not consistent with the object's repr slot.

- The mmap module only checks file size for regular files, not

character or block devices. SF patch 708374.

- The cPickle Pickler garbage collection support was fixed to traverse

the find_class attribute, if present.

- There are several fixes for the bsddb3 wrapper module.

bsddb3 no longer crashes if an environment is closed before a cursor

(SF bug 763298).

The DB and DBEnv set_get_returns_none function was extended to take

a level instead of a boolean flag. The new level 2 means that in

addition, cursor.set()/.get() methods return None instead of raising

an exception.

A typo was fixed in DBCursor.join_item(), preventing a crash.

Library

-------

- distutils now supports MSVC 7.1

- doctest now examines all docstrings by default. Previously, it would

skip over functions with private names (as indicated by the underscore

naming convention). The old default created too much of a risk that

user tests were being skipped inadvertently. Note, this change could

break code in the unlikely case that someone had intentionally put

failing tests in the docstrings of private functions. The breakage

is easily fixable by specifying the old behavior when calling testmod()

or Tester().

- There were several fixes to the way dumbdbms are closed. It's vital

that a dumbdbm database be closed properly, else the on-disk data

and directory files can be left in mutually inconsistent states.

dumbdbm.py's _Database.__del__() method attempted to close the

database properly, but a shutdown race in _Database._commit() could

prevent this from working, so that a program trusting __del__() to

get the on-disk files in synch could be badly surprised. The race

has been repaired. A sync() method was also added so that shelve

can guarantee data is written to disk.

The close() method can now be called more than once without complaint.

- The classes in threading.py are now new-style classes. That they

weren't before was an oversight.

- The urllib2 digest authentication handlers now define the correct

auth_header. The earlier versions would fail at runtime.

- SF bug 763023: fix uncaught ZeroDivisionError in difflib ratio methods

when there are no lines.

- SF bug 763637: fix exception in Tkinter with after_cancel

which could occur with Tk 8.4

- SF bug 770601: CGIHTTPServer.py now passes the entire environment

to child processes.

- SF bug 765238: add filter to fnmatch's __all__.

- SF bug 748201: make time.strptime() error messages more helpful.

- SF patch 764470: Do not dump the args attribute of a Fault object in

xmlrpclib.

- SF patch 549151: urllib and urllib2 now redirect POSTs on 301

responses.

- SF patch 766650: The whichdb module was fixed to recognize dbm files

generated by gdbm on OS/2 EMX.

- SF bugs 763047 and 763052: fixes bug of timezone value being left as

-1 when ``time.tzname[0] == time.tzname[1] and not time.daylight``

is true when it should only when time.daylight is true.

- SF bug 764548: re now allows subclasses of str and unicode to be

used as patterns.

- SF bug 763637: In Tkinter, change after_cancel() to handle tuples

of varying sizes. Tk 8.4 returns a different number of values

than Tk 8.3.

- SF bug 763023: difflib.ratio() did not catch zero division.

- The Queue module now has an __all__ attribute.

Tools/Demos

-----------

- See Lib/idlelib/NEWS.txt for IDLE news.

- SF bug 753592: webchecker/wsgui now handles user supplied directories.

- The trace.py script has been removed. It is now in the standard library.

Build

-----

- Python now compiles with -fno-strict-aliasing if possible (SF bug 766696).

- The socket module compiles on IRIX 6.5.10.

- An irix64 system is treated the same way as an irix6 system (SF

patch 764560).

- Several definitions were missing on FreeBSD 5.x unless the

__BSD_VISIBLE symbol was defined. configure now defines it as

needed.

C API

-----

- Unicode objects now support mbcs as a built-in encoding, so the C

API can use it without deferring to the encodings package.

Windows

-------

- The Windows implementation of PyThread_start_new_thread() never

checked error returns from Windows functions correctly. As a result,

it could claim to start a new thread even when the Microsoft

_beginthread() function failed (due to "too many threads" -- this is

on the order of thousands when it happens). In these cases, the

Python exception ::

thread.error: can't start new thread

is raised now.

- SF bug 766669: Prevent a GPF on interpreter exit when sockets are in

use. The interpreter now calls WSACleanup() from Py_Finalize()

instead of from DLL teardown.

Mac

---

- Bundlebuilder now inherits default values in the right way. It was

previously possible for app bundles to get a type of "BNDL" instead

of "APPL." Other improvements include, a --build-id option to

specify the CFBundleIdentifier and using the --python option to set

the executable in the bundle.

- Fixed two bugs in MacOSX framework handling.

- pythonw did not allow user interaction in 2.3rc1, this has been fixed.

- Python is now compiled with -mno-fused-madd, making all tests pass

on Panther.

What's New in Python 2.3 beta 2?

================================

*Release date: 29-Jun-2003*

Core and builtins

-----------------

- A program can now set the environment variable PYTHONINSPECT to some

string value in Python, and cause the interpreter to enter the

interactive prompt at program exit, as if Python had been invoked

with the -i option.

- list.index() now accepts optional start and stop arguments. Similar

changes were made to UserList.index(). SF feature request 754014.

- SF patch 751998 fixes an unwanted side effect of the previous fix

for SF bug 742860 (the next item).

- SF bug 742860: "WeakKeyDictionary __delitem__ uses iterkeys". This

wasn't threadsafe, was very inefficient (expected time O(len(dict))

instead of O(1)), and could raise a spurious RuntimeError if another

thread mutated the dict during __delitem__, or if a comparison function

mutated it. It also neglected to raise KeyError when the key wasn't

present; didn't raise TypeError when the key wasn't of a weakly

referencable type; and broke various more-or-less obscure dict

invariants by using a sequence of equality comparisons over the whole

set of dict keys instead of computing the key's hash code to narrow

the search to those keys with the same hash code. All of these are

considered to be bugs. A new implementation of __delitem__ repairs all

that, but note that fixing these bugs may change visible behavior in

code relying (whether intentionally or accidentally) on old behavior.

- SF bug 734869: Fixed a compiler bug that caused a fatal error when

compiling a list comprehension that contained another list comprehension

embedded in a lambda expression.

- SF bug 705231: builtin pow() no longer lets the platform C pow()

raise -1.0 to integer powers, because (at least) glibc gets it wrong

in some cases. The result should be -1.0 if the power is odd and 1.0

if the power is even, and any float with a sufficiently large exponent

is (mathematically) an exact even integer.

- SF bug 759227: A new-style class that implements __nonzero__() must

return a bool or int (but not an int subclass) from that method. This

matches the restriction on classic classes.

- The encoding attribute has been added for file objects, and set to

the terminal encoding on Unix and Windows.

- The softspace attribute of file objects became read-only by oversight.

It's writable again.

- Reverted a 2.3 beta 1 change to iterators for subclasses of list and

tuple. By default, the iterators now access data elements directly

instead of going through __getitem__. If __getitem__ access is

preferred, then __iter__ can be overridden.

- SF bug 735247: The staticmethod and super types participate in

garbage collection. Before this change, it was possible for leaks to

occur in functions with non-global free variables that used these types.

Extension modules

-----------------

- the socket module has a new exception, socket.timeout, to allow

timeouts to be handled separately from other socket errors.

- SF bug 751276: cPickle has fixed to propagate exceptions raised in

user code. In earlier versions, cPickle caught and ignored any

exception when it performed operations that it expected to raise

specific exceptions like AttributeError.

- cPickle Pickler and Unpickler objects now participate in garbage

collection.

- mimetools.choose_boundary() could return duplicate strings at times,

especially likely on Windows. The strings returned are now guaranteed

unique within a single program run.

- thread.interrupt_main() raises KeyboardInterrupt in the main thread.

dummy_thread has also been modified to try to simulate the behavior.

- array.array.insert() now treats negative indices as being relative

to the end of the array, just like list.insert() does. (SF bug #739313)

- The datetime module classes datetime, time, and timedelta are now

properly subclassable.

- _tkinter.{get|set}busywaitinterval was added.

- itertools.islice() now accepts stop=None as documented.

Fixes SF bug #730685.

- the bsddb185 module is built in one restricted instance -

/usr/include/db.h exists and defines HASHVERSION to be 2. This is true

for many BSD-derived systems.

Library

-------

- Some happy doctest extensions from Jim Fulton have been added to

doctest.py. These are already being used in Zope3. The two

primary ones:

doctest.debug(module, name) extracts the doctests from the named object

in the given module, puts them in a temp file, and starts pdb running

on that file. This is great when a doctest fails.

doctest.DocTestSuite(module=None) returns a synthesized unittest

TestSuite instance, to be run by the unittest framework, which

runs all the doctests in the module. This allows writing tests in

doctest style (which can be clearer and shorter than writing tests

in unittest style), without losing unittest's powerful testing

framework features (which doctest lacks).

- For compatibility with doctests created before 2.3, if an expected

output block consists solely of "1" and the actual output block

consists solely of "True", it's accepted as a match; similarly

for "0" and "False". This is quite un-doctest-like, but is practical.

The behavior can be disabled by passing the new doctest module

constant DONT_ACCEPT_TRUE_FOR_1 to the new optionflags optional

argument.

- ZipFile.testzip() now only traps BadZipfile exceptions. Previously,

a bare except caught to much and reported all errors as a problem

in the archive.

- The logging module now has a new function, makeLogRecord() making

LogHandler easier to interact with DatagramHandler and SocketHandler.

- The cgitb module has been extended to support plain text display (SF patch

569574).

- A brand new version of IDLE (from the IDLEfork project at

SourceForge) is now included as Lib/idlelib. The old Tools/idle is

no more.

- Added a new module: trace (documentation missing). This module used

to be distributed in Tools/scripts. It uses sys.settrace() to trace

code execution -- either function calls or individual lines. It can

generate tracing output during execution or a post-mortem report of

code coverage.

- The threading module has new functions settrace() and setprofile()

that cooperate with the functions of the same name in the sys

module. A function registered with the threading module will

be used for all threads it creates. The new trace module uses this

to provide tracing for code running in threads.

- copy.py: applied SF patch 707900, fixing bug 702858, by Steven

Taschuk. Copying a new-style class that had a reference to itself

didn't work. (The same thing worked fine for old-style classes.)

Builtin functions are now treated as atomic, fixing bug #746304.

- difflib.py has two new functions: context_diff() and unified_diff().

- More fixes to urllib (SF 549151): (a) When redirecting, always use

GET. This is common practice and more-or-less sanctioned by the

HTTP standard. (b) Add a handler for 307 redirection, which becomes

an error for POST, but a regular redirect for GET and HEAD

- Added optional 'onerror' argument to os.walk(), to control error

handling.

- inspect.is{method|data}descriptor was added, to allow pydoc display

__doc__ of data descriptors.

- Fixed socket speed loss caused by use of the _socketobject wrapper class

in socket.py.

- timeit.py now checks the current directory for imports.

- urllib2.py now knows how to order proxy classes, so the user doesn't

have to insert it in front of other classes, nor do dirty tricks like

inserting a "dummy" HTTPHandler after a ProxyHandler when building an

opener with proxy support.

- Iterators have been added for dbm keys.

- random.Random objects can now be pickled.

Tools/Demos

-----------

- pydoc now offers help on keywords and topics.

- Tools/idle is gone; long live Lib/idlelib.

- diff.py prints file diffs in context, unified, or ndiff formats,

providing a command line interface to difflib.py.

- texcheck.py is a new script for making a rough validation of Python LaTeX

files.

Build

-----

- Setting DESTDIR during 'make install' now allows specifying a

different root directory.

C API

-----

- PyType_Ready(): If a type declares that it participates in gc

(Py_TPFLAGS_HAVE_GC), and its base class does not, and its base class's

tp_free slot is the default _PyObject_Del, and type does not define

a tp_free slot itself, _PyObject_GC_Del is assigned to type->tp_free.

Previously _PyObject_Del was inherited, which could at best lead to a

segfault. In addition, if even after this magic the type's tp_free

slot is _PyObject_Del or NULL, and the type is a base type

(Py_TPFLAGS_BASETYPE), TypeError is raised: since the type is a base

type, its dealloc function must call type->tp_free, and since the type

is gc'able, tp_free must not be NULL or _PyObject_Del.

- PyThreadState_SetAsyncExc(): A new API (deliberately accessible only

from C) to interrupt a thread by sending it an exception. It is

intentional that you have to write your own C extension to call it

from Python.

New platforms

-------------

None this time.

Tests

-----

- test_imp rewritten so that it doesn't raise RuntimeError if run as a

side effect of being imported ("import test.autotest").

Windows

-------

- The Windows installer ships with Tcl/Tk 8.4.3 (upgraded from 8.4.1).

- The installer always suggested that Python be installed on the C:

drive, due to a hardcoded "C:" generated by the Wise installation

wizard. People with machines where C: is not the system drive

usually want Python installed on whichever drive is their system drive

instead. We removed the hardcoded "C:", and two testers on machines

where C: is not the system drive report that the installer now

suggests their system drive. Note that you can always select the