| Author: | A.M. Kuchling (amk at amk.ca) |
|---|---|
| Release: | 2.7a3 |
| Date: | February 08, 2010 |
This article explains the new features in Python 2.7. The final release of 2.7 is currently scheduled for June 2010; the detailed schedule is described in PEP 373.
Python 2.7 is planned to be the last major release in the 2.x series. Though more major releases have not been absolutely ruled out, it’s likely that the 2.7 release will have an extended period of maintenance compared to earlier 2.x versions.
Much as Python 2.6 incorporated features from Python 3.0, version 2.7 incorporates some of the new features in Python 3.1. The 2.x series continues to provide tools for migrating to the 3.x series.
A partial list of 3.1 features that were backported to 2.7:
One porting change: the -3 switch now automatically enables the -Qwarn switch that causes warnings about using classic division with integers and long integers.
Other new Python3-mode warnings include:
Regular Python dictionaries iterate over key/value pairs in arbitrary order. Over the years, a number of authors have written alternative implementations that remember the order that the keys were originally inserted. Based on the experiences from those implementations, a new collections.OrderedDict class has been introduced.
The OrderedDict API is substantially the same as regular dictionaries but will iterate over keys and values in a guaranteed order depending on when a key was first inserted:
>>> from collections import OrderedDict
>>> d = OrderedDict([('first', 1), ('second', 2),
... ('third', 3)])
>>> d.items()
[('first', 1), ('second', 2), ('third', 3)]
If a new entry overwrites an existing entry, the original insertion position is left unchanged:
>>> d['second'] = 4
>>> d.items()
[('first', 1), ('second', 4), ('third', 3)]
Deleting an entry and reinserting it will move it to the end:
>>> del d['second']
>>> d['second'] = 5
>>> d.items()
[('first', 1), ('third', 3), ('second', 5)]
The popitem() method has an optional last argument that defaults to True. If last is True, the most recently added key is returned and removed; if it’s False, the oldest key is selected:
>>> od = OrderedDict([(x,0) for x in range(20)])
>>> od.popitem()
(19, 0)
>>> od.popitem()
(18, 0)
>>> od.popitem(False)
(0, 0)
>>> od.popitem(False)
(1, 0)
Comparing two ordered dictionaries checks both the keys and values, and requires that the insertion order was the same:
>>> od1 = OrderedDict([('first', 1), ('second', 2),
... ('third', 3)])
>>> od2 = OrderedDict([('third', 3), ('first', 1),
... ('second', 2)])
>>> od1==od2
False
>>> # Move 'third' key to the end
>>> del od2['third'] ; od2['third'] = 3
>>> od1==od2
True
Comparing an OrderedDict with a regular dictionary ignores the insertion order and just compares the keys and values.
How does the OrderedDict work? It maintains a doubly-linked list of keys, appending new keys to the list as they’re inserted. A secondary dictionary maps keys to their corresponding list node, so deletion doesn’t have to traverse the entire linked list and therefore remains O(1).
The standard library now supports use of ordered dictionaries in several modules. The configparser module uses them by default. This lets configuration files be read, modified, and then written back in their original order. The _asdict() method for collections.namedtuple() now returns an ordered dictionary with the values appearing in the same order as the underlying tuple indicies. The json module is being built-out with an object_pairs_hook to allow OrderedDicts to be built by the decoder. Support was also added for third-party tools like PyYAML.
See also
To make program output more readable, it can be useful to add separators to large numbers and render them as 18,446,744,073,709,551,616 instead of 18446744073709551616.
The fully general solution for doing this is the locale module, which can use different separators (“,” in North America, “.” in Europe) and different grouping sizes, but locale is complicated to use and unsuitable for multi-threaded applications where different threads are producing output for different locales.
Therefore, a simple comma-grouping mechanism has been added to the mini-language used by the string format() method. When formatting a floating-point number, simply include a comma between the width and the precision:
>>> '{:20,.2}'.format(f)
'18,446,744,073,709,551,616.00'
This mechanism is not adaptable at all; commas are always used as the separator and the grouping is always into three-digit groups. The comma-formatting mechanism isn’t as general as the locale module, but it’s easier to use.
See also
XXX write this section.
See also
XXX write this section.
See also
Some smaller changes made to the core Python language are:
The syntax for set literals has been backported from Python 3.x. Curly brackets are used to surround the contents of the resulting mutable set; set literals are distinguished from dictionaries by not containing colons and values. {} continues to represent an empty dictionary; use set() for an empty set.
>>> {1,2,3,4,5}
set([1, 2, 3, 4, 5])
>>> set()
set([])
>>> {}
{}
Backported by Alexandre Vassalotti; issue 2335.
Dictionary and set comprehensions are another feature backported from 3.x, generalizing list/generator comprehensions to use the literal syntax for sets and dictionaries.
>>> {x:x*x for x in range(6)}
{0: 0, 1: 1, 2: 4, 3: 9, 4: 16, 5: 25}
>>> {'a'*x for x in range(6)}
set(['', 'a', 'aa', 'aaa', 'aaaa', 'aaaaa'])
Backported by Alexandre Vassalotti; issue 2333.
The with statement can now use multiple context managers in one statement. Context managers are processed from left to right and each one is treated as beginning a new with statement. This means that:
with A() as a, B() as b:
... suite of statements ...
is equivalent to:
with A() as a:
with B() as b:
... suite of statements ...
The contextlib.nested() function provides a very similar function, so it’s no longer necessary and has been deprecated.
(Proposed in http://codereview.appspot.com/53094; implemented by Georg Brandl.)
Conversions between floating-point numbers and strings are now correctly rounded on most platforms. These conversions occur in many different places: str() on floats and complex numbers; the float and complex constructors; numeric formatting; serialization and deserialization of floats and complex numbers using the marshal, pickle and json modules; parsing of float and imaginary literals in Python code; and Decimal-to-float conversion.
Related to this, the repr() of a floating-point number x now returns a result based on the shortest decimal string that’s guaranteed to round back to x under correct rounding (with round-half-to-even rounding mode). Previously it gave a string based on rounding x to 17 decimal digits.
The rounding library responsible for this improvement works on Windows, and on Unix platforms using the gcc, icc, or suncc compilers. There may be a small number of platforms where correct operation of this code cannot be guaranteed, so the code is not used on such systems. You can find out which code is being used by checking sys.float_repr_style, which will be short if the new code is in use and legacy if it isn’t.
Implemented by Eric Smith and Mark Dickinson, using David Gay’s dtoa.c library; issue 7117.
The str.format() method now supports automatic numbering of the replacement fields. This makes using str.format() more closely resemble using %s formatting:
>>> '{}:{}:{}'.format(2009, 04, 'Sunday')
'2009:4:Sunday'
>>> '{}:{}:{day}'.format(2009, 4, day='Sunday')
'2009:4:Sunday'
The auto-numbering takes the fields from left to right, so the first {...} specifier will use the first argument to str.format(), the next specifier will use the next argument, and so on. You can’t mix auto-numbering and explicit numbering – either number all of your specifier fields or none of them – but you can mix auto-numbering and named fields, as in the second example above. (Contributed by Eric Smith; issue 5237.)
Complex numbers now correctly support usage with format(). Specifying a precision or comma-separation applies to both the real and imaginary parts of the number, but a specified field width and alignment is applied to the whole of the resulting 1.5+3j output. (Contributed by Eric Smith; issue 1588.)
The ‘F’ format code now always formats its output using uppercase characters, so it will now produce ‘INF’ and ‘NAN’. (Contributed by Eric Smith; issue 3382.)
The int() and long() types gained a bit_length method that returns the number of bits necessary to represent its argument in binary:
>>> n = 37
>>> bin(37)
'0b100101'
>>> n.bit_length()
6
>>> n = 2**123-1
>>> n.bit_length()
123
>>> (n+1).bit_length()
124
(Contributed by Fredrik Johansson and Victor Stinner; issue 3439.)
Conversions from long integers and regular integers to floating point now round differently, returning the floating-point number closest to the number. This doesn’t matter for small integers that can be converted exactly, but for large numbers that will unavoidably lose precision, Python 2.7 now approximates more closely. For example, Python 2.6 computed the following:
>>> n = 295147905179352891391
>>> float(n)
2.9514790517935283e+20
>>> n - long(float(n))
65535L
Python 2.7’s floating-point result is larger, but much closer to the true value:
>>> n = 295147905179352891391
>>> float(n)
2.9514790517935289e+20
>>> n-long(float(n)
... )
-1L
(Implemented by Mark Dickinson; issue 3166.)
Integer division is also more accurate in its rounding behaviours. (Also implemented by Mark Dickinson; issue 1811.)
The bytearray type’s translate() method now accepts None as its first argument. (Fixed by Georg Brandl; issue 4759.)
When using @classmethod and @staticmethod to wrap methods as class or static methods, the wrapper object now exposes the wrapped function as their __func__ attribute. (Contributed by Amaury Forgeot d’Arc, after a suggestion by George Sakkis; issue 5982.)
A new encoding named “cp720”, used primarily for Arabic text, is now supported. (Contributed by Alexander Belchenko and Amaury Forgeot d’Arc; issue 1616979.)
The file object will now set the filename attribute on the IOError exception when trying to open a directory on POSIX platforms (noted by Jan Kaliszewski; issue 4764), and now explicitly checks for and forbids writing to read-only file objects instead of trusting the C library to catch and report the error (fixed by Stefan Krah; issue 5677).
The Python tokenizer now translates line endings itself, so the compile() built-in function can now accept code using any line-ending convention. Additionally, it no longer requires that the code end in a newline.
Extra parentheses in function definitions are illegal in Python 3.x, meaning that you get a syntax error from def f((x)): pass. In Python3-warning mode, Python 2.7 will now warn about this odd usage. (Noted by James Lingard; issue 7362.)
When a module object is garbage-collected, the module’s dictionary is now only cleared if no one else is holding a reference to the dictionary (issue 7140).
Several performance enhancements have been added:
A new opcode was added to perform the initial setup for with statements, looking up the __enter__() and __exit__() methods. (Contributed by Benjamin Peterson.)
The garbage collector now performs better for one common usage pattern: when many objects are being allocated without deallocating any of them. This would previously take quadratic time for garbage collection, but now the number of full garbage collections is reduced as the number of objects on the heap grows. The new logic is to only perform a full garbage collection pass when the middle generation has been collected 10 times and when the number of survivor objects from the middle generation exceeds 10% of the number of objects in the oldest generation. (Suggested by Martin von Loewis and implemented by Antoine Pitrou; issue 4074.)
The garbage collector tries to avoid tracking simple containers which can’t be part of a cycle. In Python 2.7, this is now true for tuples and dicts containing atomic types (such as ints, strings, etc.). Transitively, a dict containing tuples of atomic types won’t be tracked either. This helps reduce the cost of each garbage collection by decreasing the number of objects to be considered and traversed by the collector. (Contributed by Antoine Pitrou; issue 4688.)
Long integers are now stored internally either in base 2**15 or in base 2**30, the base being determined at build time. Previously, they were always stored in base 2**15. Using base 2**30 gives significant performance improvements on 64-bit machines, but benchmark results on 32-bit machines have been mixed. Therefore, the default is to use base 2**30 on 64-bit machines and base 2**15 on 32-bit machines; on Unix, there’s a new configure option --enable-big-digits that can be used to override this default.
Apart from the performance improvements this change should be invisible to end users, with one exception: for testing and debugging purposes there’s a new structseq sys.long_info that provides information about the internal format, giving the number of bits per digit and the size in bytes of the C type used to store each digit:
>>> import sys
>>> sys.long_info
sys.long_info(bits_per_digit=30, sizeof_digit=4)
(Contributed by Mark Dickinson; issue 4258.)
Another set of changes made long objects a few bytes smaller: 2 bytes smaller on 32-bit systems and 6 bytes on 64-bit. (Contributed by Mark Dickinson; issue 5260.)
The division algorithm for long integers has been made faster by tightening the inner loop, doing shifts instead of multiplications, and fixing an unnecessary extra iteration. Various benchmarks show speedups of between 50% and 150% for long integer divisions and modulo operations. (Contributed by Mark Dickinson; issue 5512.) Bitwise operations are also significantly faster (initial patch by Gregory Smith; issue 1087418).
The implementation of % checks for the left-side operand being a Python string and special-cases it; this results in a 1-3% performance increase for applications that frequently use % with strings, such as templating libraries. (Implemented by Collin Winter; issue 5176.)
List comprehensions with an if condition are compiled into faster bytecode. (Patch by Antoine Pitrou, back-ported to 2.7 by Jeffrey Yasskin; issue 4715.)
Converting an integer or long integer to a decimal string was made faster by special-casing base 10 instead of using a generalized conversion function that supports arbitrary bases. (Patch by Gawain Bolton; issue 6713.)
The split(), replace(), rindex(), rpartition(), and rsplit() methods of string-like types (strings, Unicode strings, and bytearray objects) now use a fast reverse-search algorithm instead of a character-by-character scan. This is sometimes faster by a factor of 10. (Added by Florent Xicluna; issue 7462 and issue 7622.)
The pickle and cPickle modules now automatically intern the strings used for attribute names, reducing memory usage of the objects resulting from unpickling. (Contributed by Jake McGuire; issue 5084.)
The cPickle module now special-cases dictionaries, nearly halving the time required to pickle them. (Contributed by Collin Winter; issue 5670.)
As in every release, Python’s standard library received a number of enhancements and bug fixes. Here’s a partial list of the most notable changes, sorted alphabetically by module name. Consult the Misc/NEWS file in the source tree for a more complete list of changes, or look through the Subversion logs for all the details.
The bdb module’s base debugging class Bdb gained a feature for skipping modules. The constructor now takes an iterable containing glob-style patterns such as django.*; the debugger will not step into stack frames from a module that matches one of these patterns. (Contributed by Maru Newby after a suggestion by Senthil Kumaran; issue 5142.)
The bz2 module’s BZ2File now supports the context management protocol, so you can write with bz2.BZ2File(...) as f: .... (Contributed by Hagen Fuerstenau; issue 3860.)
New class: the Counter class in the collections module is useful for tallying data. Counter instances behave mostly like dictionaries but return zero for missing keys instead of raising a KeyError:
>>> from collections import Counter
>>> c = Counter()
>>> for letter in 'here is a sample of english text':
... c[letter] += 1
...
>>> c
Counter({' ': 6, 'e': 5, 's': 3, 'a': 2, 'i': 2, 'h': 2,
'l': 2, 't': 2, 'g': 1, 'f': 1, 'm': 1, 'o': 1, 'n': 1,
'p': 1, 'r': 1, 'x': 1})
>>> c['e']
5
>>> c['z']
0
There are two additional Counter methods: most_common() returns the N most common elements and their counts, and elements() returns an iterator over the contained element, repeating each element as many times as its count:
>>> c.most_common(5)
[(' ', 6), ('e', 5), ('s', 3), ('a', 2), ('i', 2)]
>>> c.elements() ->
'a', 'a', ' ', ' ', ' ', ' ', ' ', ' ',
'e', 'e', 'e', 'e', 'e', 'g', 'f', 'i', 'i',
'h', 'h', 'm', 'l', 'l', 'o', 'n', 'p', 's',
's', 's', 'r', 't', 't', 'x'
Contributed by Raymond Hettinger; issue 1696199.
The new OrderedDict class is described in the earlier section PEP 372: Adding an ordered dictionary to collections.
The namedtuple class now has an optional rename parameter. If rename is true, field names that are invalid because they’ve been repeated or that aren’t legal Python identifiers will be renamed to legal names that are derived from the field’s position within the list of fields:
>>> from collections import namedtuple
>>> T = namedtuple('T', ['field1', '$illegal', 'for', 'field2'], rename=True)
>>> T._fields
('field1', '_1', '_2', 'field2')
(Added by Raymond Hettinger; issue 1818.)
The deque data type now exposes its maximum length as the read-only maxlen attribute, and has a reverse() method that reverses the elements of the deque in-place. (Added by Raymond Hettinger.)
The copy module’s deepcopy() function will now correctly copy bound instance methods. (Implemented by Robert Collins; issue 1515.)
The ctypes module now always converts None to a C NULL pointer for arguments declared as pointers. (Changed by Thomas Heller; issue 4606.)
New method: the datetime module’s timedelta class gained a total_seconds() method that returns the number of seconds in the duration. (Contributed by Brian Quinlan; issue 5788.)
New method: the Decimal class gained a from_float() class method that performs an exact conversion of a floating-point number to a Decimal. Note that this is an exact conversion that strives for the closest decimal approximation to the floating-point representation’s value; the resulting decimal value will therefore still include the inaccuracy, if any. For example, Decimal.from_float(0.1) returns Decimal('0.1000000000000000055511151231257827021181583404541015625'). (Implemented by Raymond Hettinger; issue 4796.)
The constructor for Decimal now accepts non-European Unicode characters, such as Arabic-Indic digits. (Contributed by Mark Dickinson; issue 6595.)
When using Decimal instances with a string’s format() method, the default alignment was previously left-alignment. This has been changed to right-alignment, which seems more sensible for numeric types. (Changed by Mark Dickinson; issue 6857.)
The Fraction class now accepts two rational numbers as arguments to its constructor. (Implemented by Mark Dickinson; issue 5812.)
The ftplib module gained the ability to establish secure FTP connections using TLS encapsulation of authentication as well as subsequent control and data transfers. This is provided by the new ftplib.FTP_TLS class. (Contributed by Giampaolo Rodola’, issue 2054.) The storbinary() method for binary uploads can now restart uploads thanks to an added rest parameter (patch by Pablo Mouzo; issue 6845.)
New function: the gc module’s is_tracked() returns true if a given instance is tracked by the garbage collector, false otherwise. (Contributed by Antoine Pitrou; issue 4688.)
The gzip module’s GzipFile now supports the context management protocol, so you can write with gzip.GzipFile(...) as f: ... (contributed by Hagen Fuerstenau; issue 3860), and it now implements the io.BufferedIOBase ABC, so you can wrap it with io.BufferedReader for faster processing (contributed by Nir Aides; issue 7471). It’s also now possible to override the modification time recorded in a gzipped file by providing an optional timestamp to the constructor. (Contributed by Jacques Frechet; issue 4272.)
Files in gzip format can be padded with trailing zero bytes; the gzip module will now consume these trailing bytes. (Fixed by Tadek Pietraszek and Brian Curtin; issue 2846.)
The default HTTPResponse class used by the httplib module now supports buffering, resulting in much faster reading of HTTP responses. (Contributed by Kristjan Valur Jonsson; issue 4879.)
The imaplib module now supports IPv6 addresses. (Contributed by Derek Morr; issue 1655.)
The io library has been upgraded to the version shipped with Python 3.1. For 3.1, the I/O library was entirely rewritten in C and is 2 to 20 times faster depending on the task at hand. The original Python version was renamed to the _pyio module.
One minor resulting change: the io.TextIOBase class now has an errors attribute giving the error setting used for encoding and decoding errors (one of 'strict', 'replace', 'ignore').
The io.FileIO class now raises an OSError when passed an invalid file descriptor. (Implemented by Benjamin Peterson; issue 4991.) The truncate() method now preserves the file position; previously it would change the file position to the end of the new file. (Fixed by Pascal Chambon; issue 6939.)
New function: itertools.compress(data, selectors) takes two iterators. Elements of data are returned if the corresponding value in selectors is true:
itertools.compress('ABCDEF', [1,0,1,0,1,1]) =>
A, C, E, F
New function: itertools.combinations_with_replacement(iter, r) returns all the possible r-length combinations of elements from the iterable iter. Unlike combinations(), individual elements can be repeated in the generated combinations:
itertools.combinations_with_replacement('abc', 2) =>
('a', 'a'), ('a', 'b'), ('a', 'c'),
('b', 'b'), ('b', 'c'), ('c', 'c')
Note that elements are treated as unique depending on their position in the input, not their actual values.
The itertools.count function now has a step argument that allows incrementing by values other than 1. count() also now allows keyword arguments, and using non-integer values such as floats or Decimal instances. (Implemented by Raymond Hettinger; issue 5032.)
itertools.combinations() and itertools.product() were previously raising ValueError for values of r larger than the input iterable. This was deemed a specification error, so they now return an empty iterator. (Fixed by Raymond Hettinger; issue 4816.)
The json module was upgraded to version 2.0.9 of the simplejson package, which includes a C extension that makes encoding and decoding faster. (Contributed by Bob Ippolito; issue 4136.)
To support the new OrderedDict type, json.load() now has an optional object_pairs_hook parameter that will be called with any object literal that decodes to a list of pairs. (Contributed by Raymond Hettinger; issue 5381.)
New functions: the math module gained erf() and erfc() for the error function and the complementary error function, expm1() which computes e**x - 1 with more precision than using exp() and subtracting 1, gamma() for the Gamma function, and lgamma() for the natural log of the Gamma function. (Contributed by Mark Dickinson and nirinA raseliarison; issue 3366.)
The multiprocessing module’s Manager* classes can now be passed a callable that will be called whenever a subprocess is started, along with a set of arguments that will be passed to the callable. (Contributed by lekma; issue 5585.)
The Pool class, which controls a pool of worker processes, now has an optional maxtasksperchild parameter. Worker processes will perform the specified number of tasks and then exit, causing the Pool to start a new worker. This is useful if tasks may leak memory or other resources, or if some tasks will cause the worker to become very large. (Contributed by Charles Cazabon; issue 6963.)
The nntplib module now supports IPv6 addresses. (Contributed by Derek Morr; issue 1664.)
New functions: the os module wraps the following POSIX system calls: getresgid() and getresuid(), which return the real, effective, and saved GIDs and UIDs; setresgid() and setresuid(), which set real, effective, and saved GIDs and UIDs to new values; initgroups(). (GID/UID functions contributed by Travis H.; issue 6508. Support for initgroups added by Jean-Paul Calderone; issue 7333.)
The normpath() function now preserves Unicode; if its input path is a Unicode string, the return value is also a Unicode string. (Fixed by Matt Giuca; issue 5827.)
The pydoc module now has help for the various symbols that Python uses. You can now do help('<<') or help('@'), for example. (Contributed by David Laban; issue 4739.)
The re module’s split(), sub(), and subn() now accept an optional flags argument, for consistency with the other functions in the module. (Added by Gregory P. Smith.)
The shutil module’s copyfile() and copytree() functions now raises a SpecialFileError exception when asked to copy a named pipe. Previously the code would treat named pipes like a regular file by opening them for reading, and this would block indefinitely. (Fixed by Antoine Pitrou; issue 3002.)
New functions: in the site module, three new functions return various site- and user-specific paths. getsitepackages() returns a list containing all global site-packages directories, and getusersitepackages() returns the path of the user’s site-packages directory. getuserbase() returns the value of the USER_BASE environment variable, giving the path to a directory that can be used to store data. (Contributed by Tarek Ziadé; issue 6693.)
The socket module’s SSL objects now support the buffer API, which fixed a test suite failure. (Fixed by Antoine Pitrou; issue 7133.) The create_connection() function gained a source_address parameter, a (host, port) 2-tuple giving the source address that will be used for the connection. (Contributed by Eldon Ziegler; issue 3972.)
The SocketServer module’s TCPServer class now has a disable_nagle_algorithm class attribute. The default value is False; if overridden to be True, new request connections will have the TCP_NODELAY option set to prevent buffering many small sends into a single TCP packet. (Contributed by Kristjan Valur Jonsson; issue 6192.)
The struct module will no longer silently ignore overflow errors when a value is too large for a particular integer format code (one of bBhHiIlLqQ); it now always raises a struct.error exception. (Changed by Mark Dickinson; issue 1523.)
New function: the subprocess module’s check_output() runs a command with a specified set of arguments and returns the command’s output as a string when the command runs without error, or raises a CalledProcessError exception otherwise.
>>> subprocess.check_output(['df', '-h', '.'])
'Filesystem Size Used Avail Capacity Mounted on\n
/dev/disk0s2 52G 49G 3.0G 94% /\n'
>>> subprocess.check_output(['df', '-h', '/bogus'])
...
subprocess.CalledProcessError: Command '['df', '-h', '/bogus']' returned non-zero exit status 1
(Contributed by Gregory P. Smith.)
New function: is_declared_global() in the symtable module returns true for variables that are explicitly declared to be global, false for ones that are implicitly global. (Contributed by Jeremy Hylton.)
The sys.version_info value is now a named tuple, with attributes named major, minor, micro, releaselevel, and serial. (Contributed by Ross Light; issue 4285.)
sys.getwindowsversion() also returns a named tuple, with attributes named major, minor, build, platform:, service_pack, service_pack_major, service_pack_minor, suite_mask, and product_type. (Contributed by Brian Curtin; issue 7766.)
The tarfile module’s default error handling has changed, to no longer suppress fatal errors. The default error level was previously 0, which meant that errors would only result in a message being written to the debug log, but because the debug log is not activated by default, these errors go unnoticed. The default error level is now 1, which raises an exception if there’s an error. (Changed by Lars Gustäbel; issue 7357.)
tarfile now supports filtering the TarInfo objects being added to a tar file. When you call TarFile.add(), instance, you may supply an optional filter argument that’s a callable. The filter callable will be passed the TarInfo for every file being added, and can modify and return it. If the callable returns None, the file will be excluded from the resulting archive. This is more powerful than the existing exclude argument, which has therefore been deprecated. (Added by Lars Gustäbel; issue 6856.)
The threading module’s Event.wait() method now returns the internal flag on exit. This means the method will usually return true because wait() is supposed to block until the internal flag becomes true. The return value will only be false if a timeout was provided and the operation timed out. (Contributed by Tim Lesher; issue 1674032.)
The UserDict class is now a new-style class. (Changed by Benjamin Peterson.)
The zipfile module’s ZipFile now supports the context management protocol, so you can write with zipfile.ZipFile(...) as f: .... (Contributed by Brian Curtin; issue 5511.)
zipfile now supports archiving empty directories and extracts them correctly. (Fixed by Kuba Wieczorek; issue 4710.) Reading files out of an archive is now faster, and interleaving read() and readline() now works correctly. (Contributed by Nir Aides; issue 7610.)
The is_zipfile() function in the module now accepts a file object, in addition to the path names accepted in earlier versions. (Contributed by Gabriel Genellina; issue 4756.)
XXX A new sysconfig module has been extracted from distutils and put in the standard library.
The sysconfig module provides access to Python’s configuration information like the list of installation paths and the configuration variables relevant for the current platform.
Distutils is being more actively developed, thanks to Tarek Ziadé who has taken over maintenance of the package, so there are a number of fixes and improvements.
A new setup.py subcommand, check, will check that the arguments being passed to the setup() function are complete and correct (issue 5732).
Byte-compilation by the install_lib subcommand is now only done if the sys.dont_write_bytecode setting allows it (issue 7071).
distutils.sdist.add_defaults() now uses package_dir and data_files to create the MANIFEST file. distutils.sysconfig now reads the AR and ARFLAGS environment variables.
It is no longer mandatory to store clear-text passwords in the .pypirc file when registering and uploading packages to PyPI. As long as the username is present in that file, the distutils package will prompt for the password if not present. (Added by Tarek Ziadé, based on an initial contribution by Nathan Van Gheem; issue 4394.)
A Distutils setup can now specify that a C extension is optional by setting the optional option setting to true. If this optional is supplied, failure to build the extension will not abort the build process, but instead simply not install the failing extension. (Contributed by Georg Brandl; issue 5583.)
The distutils.dist.DistributionMetadata class’ read_pkg_file() method will read the contents of a package’s PKG-INFO metadata file. For an example of its use, see Reading the metadata. (Contributed by Tarek Ziadé; issue 7457.)
setup.py files will now accept a --no-user-cfg switch to skip reading the ~/.pydistutils.cfg file. (Suggested by by Michael Hoffman, and implemented by Paul Winkler; issue 1180.)
When creating a tar-format archive, the sdist subcommand now allows specifying the user id and group that will own the files in the archives using the --owner and --group switches (issue 6516).
The unittest module was enhanced in several ways. The progress messages now shows ‘x’ for expected failures and ‘u’ for unexpected successes when run in verbose mode. (Contributed by Benjamin Peterson.) Test cases can raise the SkipTest exception to skip a test. (issue 1034053.)
The error messages for assertEqual(), assertTrue(), and assertFalse() failures now provide more information. If you set the longMessage attribute of your TestCase classes to true, both the standard error message and any additional message you provide will be printed for failures. (Added by Michael Foord; issue 5663.)
The assertRaises() and failUnlessRaises() methods now return a context handler when called without providing a callable object to run. For example, you can write this:
with self.assertRaises(KeyError):
raise ValueError
(Implemented by Antoine Pitrou; issue 4444.)
The methods addCleanup() and doCleanups() were added. addCleanup() allows you to add cleanup functions that will be called unconditionally (after setUp() if setUp() fails, otherwise after tearDown()). This allows for much simpler resource allocation and deallocation during tests. issue 5679
A number of new methods were added that provide more specialized tests. Many of these methods were written by Google engineers for use in their test suites; Gregory P. Smith, Michael Foord, and GvR worked on merging them into Python’s version of unittest.
unittest.main() now takes an optional exit argument. If False main doesn’t call sys.exit() allowing it to be used from the interactive interpreter. issue 3379.
TestResult has new startTestRun() and stopTestRun() methods; called immediately before and after a test run. issue 5728 by Robert Collins.
With all these changes, the unittest.py was becoming awkwardly large, so the module was turned into a package and the code split into several files (by Benjamin Peterson). This doesn’t affect how the module is imported.
Python 3.1 includes the importlib package, a re-implementation of the logic underlying Python’s import statement. importlib is useful for implementors of Python interpreters and to users who wish to write new importers that can participate in the import process. Python 2.7 doesn’t contain the complete importlib package, but instead has a tiny subset that contains a single function, import_module().
import_module(name, package=None) imports a module. name is a string containing the module or package’s name. It’s possible to do relative imports by providing a string that begins with a . character, such as ..utils.errors. For relative imports, the package argument must be provided and is the name of the package that will be used as the anchor for the relative import. import_module() both inserts the imported module into sys.modules and returns the module object.
Here are some examples:
>>> from importlib import import_module
>>> anydbm = import_module('anydbm') # Standard absolute import
>>> anydbm
<module 'anydbm' from '/p/python/Lib/anydbm.py'>
>>> # Relative import
>>> sysconfig = import_module('..sysconfig', 'distutils.command')
>>> sysconfig
<module 'distutils.sysconfig' from '/p/python/Lib/distutils/sysconfig.pyc'>
importlib was implemented by Brett Cannon and introduced in Python 3.1.
Tcl/Tk 8.5 includes a set of themed widgets that re-implement basic Tk widgets but have a more customizable appearance and can therefore more closely resemble the native platform’s widgets. This widget set was originally called Tile, but was renamed to Ttk (for “themed Tk”) on being added to Tcl/Tck release 8.5.
XXX write a brief discussion and an example here.
The ttk module was written by Guilherme Polo and added in issue 2983. An alternate version called Tile.py, written by Martin Franklin and maintained by Kevin Walzer, was proposed for inclusion in issue 2618, but the authors argued that Guilherme Polo’s work was more comprehensive.
Changes to Python’s build process and to the C API include:
If you use the .gdbinit file provided with Python, the “pyo” macro in the 2.7 version now works correctly when the thread being debugged doesn’t hold the GIL; the macro now acquires it before printing. (Contributed by Victor Stinner; issue 3632.)
Py_AddPendingCall() is now thread-safe, letting any worker thread submit notifications to the main Python thread. This is particularly useful for asynchronous IO operations. (Contributed by Kristjan Valur Jonsson; issue 4293.)
New function: PyCode_NewEmpty() creates an empty code object; only the filename, function name, and first line number are required. This is useful to extension modules that are attempting to construct a more useful traceback stack. Previously such extensions needed to call PyCode_New(), which had many more arguments. (Added by Jeffrey Yasskin.)
New function: PyErr_NewExceptionWithDoc() creates a new exception class, just as the existing PyErr_NewException() does, but takes an extra char * argument containing the docstring for the new exception class. (Added by the ‘lekma’ user on the Python bug tracker; issue 7033.)
New function: PyFrame_GetLineNumber() takes a frame object and returns the line number that the frame is currently executing. Previously code would need to get the index of the bytecode instruction currently executing, and then look up the line number corresponding to that address. (Added by Jeffrey Yasskin.)
New functions: PyLong_AsLongAndOverflow() and PyLong_AsLongLongAndOverflow() approximates a Python long integer as a C long or long long. If the number is too large to fit into the output type, an overflow flag is set and returned to the caller. (Contributed by Case Van Horsen; issue 7528 and issue 7767.)
New function: stemming from the rewrite of string-to-float conversion, a new PyOS_string_to_double() function was added. The old PyOS_ascii_strtod() and PyOS_ascii_atof() functions are now deprecated.
New macros: the Python header files now define the following macros: Py_ISALNUM, Py_ISALPHA, Py_ISDIGIT, Py_ISLOWER, Py_ISSPACE, Py_ISUPPER, Py_ISXDIGIT, and Py_TOLOWER, Py_TOUPPER. All of these functions are analogous to the C standard macros for classifying characters, but ignore the current locale setting, because in several places Python needs to analyze characters in a locale-independent way. (Added by Eric Smith; issue 5793.)
New format codes: the PyFormat_FromString(), PyFormat_FromStringV(), and PyErr_Format() now accepts %lld and %llu format codes for displaying values of C’s long long types. (Contributed by Mark Dickinson; issue 7228.)
The complicated interaction between threads and process forking has been changed. Previously, the child process created by os.fork() might fail because the child is created with only a single thread running, the thread performing the os.fork(). If other threads were holding a lock, such as Python’s import lock, when the fork was performed, the lock would still be marked as “held” in the new process. But in the child process nothing would ever release the lock, since the other threads weren’t replicated, and the child process would no longer be able to perform imports.
Python 2.7 now acquires the import lock before performing an os.fork(), and will also clean up any locks created using the threading module. C extension modules that have internal locks, or that call fork() themselves, will not benefit from this clean-up.
(Fixed by Thomas Wouters; issue 1590864.)
The Py_Finalize() function now calls the internal threading._shutdown() function; this prevents some exceptions from being raised when an interpreter shuts down. (Patch by Adam Olsen; issue 1722344.)
Global symbols defined by the ctypes module are now prefixed with Py, or with _ctypes. (Implemented by Thomas Heller; issue 3102.)
New configure option: the --with-system-expat switch allows building the pyexpat module to use the system Expat library. (Contributed by Arfrever Frehtes Taifersar Arahesis; issue 7609.)
New configure option: Compiling Python with the --with-valgrind option will now disable the pymalloc allocator, which is difficult for the Valgrind to analyze correctly. Valgrind will therefore be better at detecting memory leaks and overruns. (Contributed by James Henstridge; issue 2422.)
New configure option: you can now supply no arguments to --with-dbmliborder= in order to build none of the various DBM modules. (Added by Arfrever Frehtes Taifersar Arahesis; issue 6491.)
The configure script now checks for floating-point rounding bugs on certain 32-bit Intel chips and defines a X87_DOUBLE_ROUNDING preprocessor definition. No code currently uses this definition, but it’s available if anyone wishes to use it. (Added by Mark Dickinson; issue 2937.)
The build process now creates the necessary files for pkg-config support. (Contributed by Clinton Roy; issue 3585.)
The build process now supports Subversion 1.7. (Contributed by Arfrever Frehtes Taifersar Arahesis; issue 6094.)
This section lists previously described changes and other bugfixes that may require changes to your code:
When using Decimal instances with a string’s format() method, the default alignment was previously left-alignment. This has been changed to right-alignment, which might change the output of your programs. (Changed by Mark Dickinson; issue 6857.)
Another format()-related change: the default precision used for floating-point and complex numbers was changed from 6 decimal places to 12, which matches the precision used by str(). (Changed by Eric Smith; issue 5920.)
Because of an optimization for the with statement, the special methods __enter__() and __exit__() must belong to the object’s type, and cannot be directly attached to the object’s instance. This affects new-style classes (derived from object) and C extension types. (issue 6101.)
The readline() method of StringIO objects now does nothing when a negative length is requested, as other file-like objects do. (issue 7348).
For C extensions:
The author would like to thank the following people for offering suggestions, corrections and assistance with various drafts of this article: Ryan Lovett, Hugh Secker-Walker.