1
****************************
2
What's New In Python 3.0
3
****************************
5
.. XXX Add trademark info for Apple, Microsoft.
7
:Author: Guido van Rossum
10
Rules for maintenance:
12
* Anyone can add text to this document. Do not spend very much time
13
on the wording of your changes, because your text will probably
14
get rewritten to some degree.
16
* The maintainer will go through Misc/NEWS periodically and add
17
changes; it's therefore more important to add your changes to
18
Misc/NEWS than to this file. (Note: I didn't get to this for 3.0.
21
* This is not a complete list of every single change; completeness
22
is the purpose of Misc/NEWS. Some changes I consider too small
23
or esoteric to include. If such a change is added to the text,
24
I'll just remove it. (This is another reason you shouldn't spend
25
too much time on writing your addition.)
27
* If you want to draw your new text to the attention of the
28
maintainer, add 'XXX' to the beginning of the paragraph or
31
* It's OK to just add a fragmentary note about a change. For
32
example: "XXX Describe the transmogrify() function added to the
33
socket module." The maintainer will research the change and
34
write the necessary text.
36
* You can comment out your additions if you like, but it's not
37
necessary (especially when a final release is some months away).
39
* Credit the author of a patch or bugfix. Just the name is
40
sufficient; the e-mail address isn't necessary. (Due to time
41
constraints I haven't managed to do this for 3.0. GvR.)
43
* It's helpful to add the bug/patch number as a comment:
46
XXX Describe the transmogrify() function added to the socket
48
(Contributed by P.Y. Developer.)
50
This saves the maintainer the effort of going through the SVN log
51
when researching a change. (Again, I didn't get to this for 3.0.
54
This article explains the new features in Python 3.0, compared to 2.6.
55
Python 3.0, also known as "Python 3000" or "Py3K", is the first ever
56
*intentionally backwards incompatible* Python release. There are more
57
changes than in a typical release, and more that are important for all
58
Python users. Nevertheless, after digesting the changes, you'll find
59
that Python really hasn't changed all that much -- by and large, we're
60
mostly fixing well-known annoyances and warts, and removing a lot of
63
This article doesn't attempt to provide a complete specification of
64
all new features, but instead tries to give a convenient overview.
65
For full details, you should refer to the documentation for Python
66
3.0, and/or the many PEPs referenced in the text. If you want to
67
understand the complete implementation and design rationale for a
68
particular feature, PEPs usually have more details than the regular
69
documentation; but note that PEPs usually are not kept up-to-date once
70
a feature has been fully implemented.
72
Due to time constraints this document is not as complete as it should
73
have been. As always for a new release, the ``Misc/NEWS`` file in the
74
source distribution contains a wealth of detailed information about
75
every small thing that was changed.
77
.. Compare with previous release in 2 - 3 sentences here.
78
.. add hyperlink when the documentation becomes available online.
80
.. ======================================================================
81
.. Large, PEP-level features and changes should be described here.
82
.. Should there be a new section here for 3k migration?
83
.. Or perhaps a more general section describing module changes/deprecation?
84
.. sets module deprecated
85
.. ======================================================================
88
Common Stumbling Blocks
89
=======================
91
This section lists those few changes that are most likely to trip you
92
up if you're used to Python 2.5.
97
The ``print`` statement has been replaced with a :func:`print`
98
function, with keyword arguments to replace most of the special syntax
99
of the old ``print`` statement (:pep:`3105`). Examples::
101
Old: print "The answer is", 2*2
102
New: print("The answer is", 2*2)
104
Old: print x, # Trailing comma suppresses newline
105
New: print(x, end=" ") # Appends a space instead of a newline
107
Old: print # Prints a newline
108
New: print() # You must call the function!
110
Old: print >>sys.stderr, "fatal error"
111
New: print("fatal error", file=sys.stderr)
113
Old: print (x, y) # prints repr((x, y))
114
New: print((x, y)) # Not the same as print(x, y)!
116
You can also customize the separator between items, e.g.::
118
print("There are <", 2**32, "> possibilities!", sep="")
124
There are <4294967296> possibilities!
128
* The :func:`print` function doesn't support the "softspace" feature of
129
the old ``print`` statement. For example, in Python 2.x,
130
``print "A\n", "B"`` would write ``"A\nB\n"``; but in Python 3.0,
131
``print("A\n", "B")`` writes ``"A\n B\n"``.
133
* Initially, you'll be finding yourself typing the old ``print x``
134
a lot in interactive mode. Time to retrain your fingers to type
135
``print(x)`` instead!
137
* When using the ``2to3`` source-to-source conversion tool, all
138
``print`` statements are automatically converted to
139
:func:`print` function calls, so this is mostly a non-issue for
142
Views And Iterators Instead Of Lists
143
-------------------------------------
145
Some well-known APIs no longer return lists:
147
* :class:`dict` methods :meth:`dict.keys`, :meth:`dict.items` and
148
:meth:`dict.values` return "views" instead of lists. For example,
149
this no longer works: ``k = d.keys(); k.sort()``. Use ``k =
150
sorted(d)`` instead (this works in Python 2.5 too and is just
153
* Also, the :meth:`dict.iterkeys`, :meth:`dict.iteritems` and
154
:meth:`dict.itervalues` methods are no longer supported.
156
* :func:`map` and :func:`filter` return iterators. If you really need
157
a list and the input sequences are all of equal length, a quick
158
fix is to wrap :func:`map` in :func:`list`, e.g. ``list(map(...))``,
160
often to use a list comprehension (especially when the original code
161
uses :keyword:`lambda`), or rewriting the code so it doesn't need a
162
list at all. Particularly tricky is :func:`map` invoked for the
163
side effects of the function; the correct transformation is to use a
164
regular :keyword:`for` loop (since creating a list would just be
167
If the input sequences are not of equal length, :func:`map` will
168
stop at the termination of the shortest of the sequences. For full
169
compatibility with :func:`map` from Python 2.x, also wrap the sequences in
170
:func:`itertools.zip_longest`, e.g. ``map(func, *sequences)`` becomes
171
``list(map(func, itertools.zip_longest(*sequences)))``.
173
* :func:`range` now behaves like :func:`xrange` used to behave, except
174
it works with values of arbitrary size. The latter no longer
177
* :func:`zip` now returns an iterator.
182
Python 3.0 has simplified the rules for ordering comparisons:
184
* The ordering comparison operators (``<``, ``<=``, ``>=``, ``>``)
185
raise a TypeError exception when the operands don't have a
186
meaningful natural ordering. Thus, expressions like ``1 < ''``, ``0
187
> None`` or ``len <= len`` are no longer valid, and e.g. ``None <
188
None`` raises :exc:`TypeError` instead of returning
189
``False``. A corollary is that sorting a heterogeneous list
190
no longer makes sense -- all the elements must be comparable to each
191
other. Note that this does not apply to the ``==`` and ``!=``
192
operators: objects of different incomparable types always compare
193
unequal to each other.
195
* :meth:`builtin.sorted` and :meth:`list.sort` no longer accept the
196
*cmp* argument providing a comparison function. Use the *key*
197
argument instead. N.B. the *key* and *reverse* arguments are now
200
* The :func:`cmp` function should be treated as gone, and the :meth:`__cmp__`
201
special method is no longer supported. Use :meth:`__lt__` for sorting,
202
:meth:`__eq__` with :meth:`__hash__`, and other rich comparisons as needed.
203
(If you really need the :func:`cmp` functionality, you could use the
204
expression ``(a > b) - (a < b)`` as the equivalent for ``cmp(a, b)``.)
209
* :pep:`237`: Essentially, :class:`long` renamed to :class:`int`.
210
That is, there is only one built-in integral type, named
211
:class:`int`; but it behaves mostly like the old :class:`long` type.
213
* :pep:`238`: An expression like ``1/2`` returns a float. Use
214
``1//2`` to get the truncating behavior. (The latter syntax has
215
existed for years, at least since Python 2.2.)
217
* The :data:`sys.maxint` constant was removed, since there is no
218
longer a limit to the value of integers. However, :data:`sys.maxsize`
219
can be used as an integer larger than any practical list or string
220
index. It conforms to the implementation's "natural" integer size
221
and is typically the same as :data:`sys.maxint` in previous releases
222
on the same platform (assuming the same build options).
224
* The :func:`repr` of a long integer doesn't include the trailing ``L``
225
anymore, so code that unconditionally strips that character will
226
chop off the last digit instead. (Use :func:`str` instead.)
228
* Octal literals are no longer of the form ``0720``; use ``0o720``
231
Text Vs. Data Instead Of Unicode Vs. 8-bit
232
------------------------------------------
234
Everything you thought you knew about binary data and Unicode has
237
* Python 3.0 uses the concepts of *text* and (binary) *data* instead
238
of Unicode strings and 8-bit strings. All text is Unicode; however
239
*encoded* Unicode is represented as binary data. The type used to
240
hold text is :class:`str`, the type used to hold data is
241
:class:`bytes`. The biggest difference with the 2.x situation is
242
that any attempt to mix text and data in Python 3.0 raises
243
:exc:`TypeError`, whereas if you were to mix Unicode and 8-bit
244
strings in Python 2.x, it would work if the 8-bit string happened to
245
contain only 7-bit (ASCII) bytes, but you would get
246
:exc:`UnicodeDecodeError` if it contained non-ASCII values. This
247
value-specific behavior has caused numerous sad faces over the
250
* As a consequence of this change in philosophy, pretty much all code
251
that uses Unicode, encodings or binary data most likely has to
252
change. The change is for the better, as in the 2.x world there
253
were numerous bugs having to do with mixing encoded and unencoded
254
text. To be prepared in Python 2.x, start using :class:`unicode`
255
for all unencoded text, and :class:`str` for binary or encoded data
256
only. Then the ``2to3`` tool will do most of the work for you.
258
* You can no longer use ``u"..."`` literals for Unicode text.
259
However, you must use ``b"..."`` literals for binary data.
261
* As the :class:`str` and :class:`bytes` types cannot be mixed, you
262
must always explicitly convert between them. Use :meth:`str.encode`
263
to go from :class:`str` to :class:`bytes`, and :meth:`bytes.decode`
264
to go from :class:`bytes` to :class:`str`. You can also use
265
``bytes(s, encoding=...)`` and ``str(b, encoding=...)``,
268
* Like :class:`str`, the :class:`bytes` type is immutable. There is a
269
separate *mutable* type to hold buffered binary data,
270
:class:`bytearray`. Nearly all APIs that accept :class:`bytes` also
271
accept :class:`bytearray`. The mutable API is based on
272
:class:`collections.MutableSequence`.
274
* All backslashes in raw string literals are interpreted literally.
275
This means that ``'\U'`` and ``'\u'`` escapes in raw strings are not
276
treated specially. For example, ``r'\u20ac'`` is a string of 6
277
characters in Python 3.0, whereas in 2.6, ``ur'\u20ac'`` was the
278
single "euro" character. (Of course, this change only affects raw
279
string literals; the euro character is ``'\u20ac'`` in Python 3.0.)
281
* The built-in :class:`basestring` abstract type was removed. Use
282
:class:`str` instead. The :class:`str` and :class:`bytes` types
283
don't have functionality enough in common to warrant a shared base
284
class. The ``2to3`` tool (see below) replaces every occurrence of
285
:class:`basestring` with :class:`str`.
287
* Files opened as text files (still the default mode for :func:`open`)
288
always use an encoding to map between strings (in memory) and bytes
289
(on disk). Binary files (opened with a ``b`` in the mode argument)
290
always use bytes in memory. This means that if a file is opened
291
using an incorrect mode or encoding, I/O will likely fail loudly,
292
instead of silently producing incorrect data. It also means that
293
even Unix users will have to specify the correct mode (text or
294
binary) when opening a file. There is a platform-dependent default
295
encoding, which on Unixy platforms can be set with the ``LANG``
296
environment variable (and sometimes also with some other
297
platform-specific locale-related environment variables). In many
298
cases, but not all, the system default is UTF-8; you should never
299
count on this default. Any application reading or writing more than
300
pure ASCII text should probably have a way to override the encoding.
301
There is no longer any need for using the encoding-aware streams
302
in the :mod:`codecs` module.
304
* The initial values of :data:`sys.stdin`, :data:`sys.stdout` and
305
:data:`sys.stderr` are now unicode-only text files (i.e., they are
306
instances of :class:`io.TextIOBase`). To read and write bytes data
307
with these streams, you need to use their :data:`io.TextIOBase.buffer`
310
* Filenames are passed to and returned from APIs as (Unicode) strings.
311
This can present platform-specific problems because on some
312
platforms filenames are arbitrary byte strings. (On the other hand,
313
on Windows filenames are natively stored as Unicode.) As a
314
work-around, most APIs (e.g. :func:`open` and many functions in the
315
:mod:`os` module) that take filenames accept :class:`bytes` objects
316
as well as strings, and a few APIs have a way to ask for a
317
:class:`bytes` return value. Thus, :func:`os.listdir` returns a
318
list of :class:`bytes` instances if the argument is a :class:`bytes`
319
instance, and :func:`os.getcwdb` returns the current working
320
directory as a :class:`bytes` instance. Note that when
321
:func:`os.listdir` returns a list of strings, filenames that
322
cannot be decoded properly are omitted rather than raising
325
* Some system APIs like :data:`os.environ` and :data:`sys.argv` can
326
also present problems when the bytes made available by the system is
327
not interpretable using the default encoding. Setting the ``LANG``
328
variable and rerunning the program is probably the best approach.
330
* :pep:`3138`: The :func:`repr` of a string no longer escapes
331
non-ASCII characters. It still escapes control characters and code
332
points with non-printable status in the Unicode standard, however.
334
* :pep:`3120`: The default source encoding is now UTF-8.
336
* :pep:`3131`: Non-ASCII letters are now allowed in identifiers.
337
(However, the standard library remains ASCII-only with the exception
338
of contributor names in comments.)
340
* The :mod:`StringIO` and :mod:`cStringIO` modules are gone. Instead,
341
import the :mod:`io` module and use :class:`io.StringIO` or
342
:class:`io.BytesIO` for text and data respectively.
344
* See also the :ref:`unicode-howto`, which was updated for Python 3.0.
347
Overview Of Syntax Changes
348
==========================
350
This section gives a brief overview of every *syntactic* change in
356
* :pep:`3107`: Function argument and return value annotations. This
357
provides a standardized way of annotating a function's parameters
358
and return value. There are no semantics attached to such
359
annotations except that they can be introspected at runtime using
360
the :attr:`__annotations__` attribute. The intent is to encourage
361
experimentation through metaclasses, decorators or frameworks.
363
* :pep:`3102`: Keyword-only arguments. Named parameters occurring
364
after ``*args`` in the parameter list *must* be specified using
365
keyword syntax in the call. You can also use a bare ``*`` in the
366
parameter list to indicate that you don't accept a variable-length
367
argument list, but you do have keyword-only arguments.
369
* Keyword arguments are allowed after the list of base classes in a
370
class definition. This is used by the new convention for specifying
371
a metaclass (see next section), but can be used for other purposes
372
as well, as long as the metaclass supports it.
374
* :pep:`3104`: :keyword:`nonlocal` statement. Using ``nonlocal x``
375
you can now assign directly to a variable in an outer (but
376
non-global) scope. :keyword:`nonlocal` is a new reserved word.
378
* :pep:`3132`: Extended Iterable Unpacking. You can now write things
379
like ``a, b, *rest = some_sequence``. And even ``*rest, a =
380
stuff``. The ``rest`` object is always a (possibly empty) list; the
381
right-hand side may be any iterable. Example::
383
(a, *rest, b) = range(5)
385
This sets *a* to ``0``, *b* to ``4``, and *rest* to ``[1, 2, 3]``.
387
* Dictionary comprehensions: ``{k: v for k, v in stuff}`` means the
388
same thing as ``dict(stuff)`` but is more flexible. (This is
389
:pep:`274` vindicated. :-)
391
* Set literals, e.g. ``{1, 2}``. Note that ``{}`` is an empty
392
dictionary; use ``set()`` for an empty set. Set comprehensions are
393
also supported; e.g., ``{x for x in stuff}`` means the same thing as
394
``set(stuff)`` but is more flexible.
396
* New octal literals, e.g. ``0o720`` (already in 2.6). The old octal
397
literals (``0720``) are gone.
399
* New binary literals, e.g. ``0b1010`` (already in 2.6), and
400
there is a new corresponding built-in function, :func:`bin`.
402
* Bytes literals are introduced with a leading ``b`` or ``B``, and
403
there is a new corresponding built-in function, :func:`bytes`.
408
* :pep:`3109` and :pep:`3134`: new :keyword:`raise` statement syntax:
409
:samp:`raise [{expr} [from {expr}]]`. See below.
411
* :keyword:`as` and :keyword:`with` are now reserved words. (Since
414
* ``True``, ``False``, and ``None`` are reserved words. (2.6 partially enforced
415
the restrictions on ``None`` already.)
417
* Change from :keyword:`except` *exc*, *var* to
418
:keyword:`except` *exc* :keyword:`as` *var*. See :pep:`3110`.
420
* :pep:`3115`: New Metaclass Syntax. Instead of::
428
class C(metaclass=M):
431
The module-global :data:`__metaclass__` variable is no longer
432
supported. (It was a crutch to make it easier to default to
433
new-style classes without deriving every class from
436
* List comprehensions no longer support the syntactic form
437
:samp:`[... for {var} in {item1}, {item2}, ...]`. Use
438
:samp:`[... for {var} in ({item1}, {item2}, ...)]` instead.
439
Also note that list comprehensions have different semantics: they
440
are closer to syntactic sugar for a generator expression inside a
441
:func:`list` constructor, and in particular the loop control
442
variables are no longer leaked into the surrounding scope.
444
* The *ellipsis* (``...``) can be used as an atomic expression
445
anywhere. (Previously it was only allowed in slices.) Also, it
446
*must* now be spelled as ``...``. (Previously it could also be
447
spelled as ``. . .``, by a mere accident of the grammar.)
452
* :pep:`3113`: Tuple parameter unpacking removed. You can no longer
453
write ``def foo(a, (b, c)): ...``.
454
Use ``def foo(a, b_c): b, c = b_c`` instead.
456
* Removed backticks (use :func:`repr` instead).
458
* Removed ``<>`` (use ``!=`` instead).
460
* Removed keyword: :func:`exec` is no longer a keyword; it remains as
461
a function. (Fortunately the function syntax was also accepted in
462
2.x.) Also note that :func:`exec` no longer takes a stream argument;
463
instead of ``exec(f)`` you can use ``exec(f.read())``.
465
* Integer literals no longer support a trailing ``l`` or ``L``.
467
* String literals no longer support a leading ``u`` or ``U``.
469
* The :keyword:`from` *module* :keyword:`import` ``*`` syntax is only
470
allowed at the module level, no longer inside functions.
472
* The only acceptable syntax for relative imports is :samp:`from .[{module}]
473
import {name}`. All :keyword:`import` forms not starting with ``.`` are
474
interpreted as absolute imports. (:pep:`328`)
476
* Classic classes are gone.
479
Changes Already Present In Python 2.6
480
=====================================
482
Since many users presumably make the jump straight from Python 2.5 to
483
Python 3.0, this section reminds the reader of new features that were
484
originally designed for Python 3.0 but that were back-ported to Python
485
2.6. The corresponding sections in :ref:`whats-new-in-2.6` should be
486
consulted for longer descriptions.
488
* :ref:`pep-0343`. The :keyword:`with` statement is now a standard
489
feature and no longer needs to be imported from the :mod:`__future__`.
490
Also check out :ref:`new-26-context-managers` and
491
:ref:`new-module-contextlib`.
493
* :ref:`pep-0366`. This enhances the usefulness of the :option:`-m`
494
option when the referenced module lives in a package.
500
* :ref:`pep-3101`. Note: the 2.6 description mentions the
501
:meth:`format` method for both 8-bit and Unicode strings. In 3.0,
502
only the :class:`str` type (text strings with Unicode support)
503
supports this method; the :class:`bytes` type does not. The plan is
504
to eventually make this the only API for string formatting, and to
505
start deprecating the ``%`` operator in Python 3.1.
507
* :ref:`pep-3105`. This is now a standard feature and no longer needs
508
to be imported from :mod:`__future__`. More details were given above.
510
* :ref:`pep-3110`. The :keyword:`except` *exc* :keyword:`as` *var*
511
syntax is now standard and :keyword:`except` *exc*, *var* is no
512
longer supported. (Of course, the :keyword:`as` *var* part is still
515
* :ref:`pep-3112`. The ``b"..."`` string literal notation (and its
516
variants like ``b'...'``, ``b"""..."""``, and ``br"..."``) now
517
produces a literal of type :class:`bytes`.
519
* :ref:`pep-3116`. The :mod:`io` module is now the standard way of
520
doing file I/O. The built-in :func:`open` function is now an
521
alias for :func:`io.open` and has additional keyword arguments
522
*encoding*, *errors*, *newline* and *closefd*. Also note that an
523
invalid *mode* argument now raises :exc:`ValueError`, not
524
:exc:`IOError`. The binary file object underlying a text file
525
object can be accessed as :attr:`f.buffer` (but beware that the
526
text object maintains a buffer of itself in order to speed up
527
the encoding and decoding operations).
529
* :ref:`pep-3118`. The old builtin :func:`buffer` is now really gone;
530
the new builtin :func:`memoryview` provides (mostly) similar
533
* :ref:`pep-3119`. The :mod:`abc` module and the ABCs defined in the
534
:mod:`collections` module plays a somewhat more prominent role in
535
the language now, and built-in collection types like :class:`dict`
536
and :class:`list` conform to the :class:`collections.MutableMapping`
537
and :class:`collections.MutableSequence` ABCs, respectively.
539
* :ref:`pep-3127`. As mentioned above, the new octal literal
540
notation is the only one supported, and binary literals have been
545
* :ref:`pep-3141`. The :mod:`numbers` module is another new use of
546
ABCs, defining Python's "numeric tower". Also note the new
547
:mod:`fractions` module which implements :class:`numbers.Rational`.
553
Due to time constraints, this document does not exhaustively cover the
554
very extensive changes to the standard library. :pep:`3108` is the
555
reference for the major changes to the library. Here's a capsule
558
* Many old modules were removed. Some, like :mod:`gopherlib` (no
559
longer used) and :mod:`md5` (replaced by :mod:`hashlib`), were
560
already deprecated by :pep:`4`. Others were removed as a result
561
of the removal of support for various platforms such as Irix, BeOS
562
and Mac OS 9 (see :pep:`11`). Some modules were also selected for
563
removal in Python 3.0 due to lack of use or because a better
564
replacement exists. See :pep:`3108` for an exhaustive list.
566
* The :mod:`bsddb3` package was removed because its presence in the
567
core standard library has proved over time to be a particular burden
568
for the core developers due to testing instability and Berkeley DB's
569
release schedule. However, the package is alive and well,
570
externally maintained at https://www.jcea.es/programacion/pybsddb.htm.
572
* Some modules were renamed because their old name disobeyed
573
:pep:`8`, or for various other reasons. Here's the list:
575
======================= =======================
577
======================= =======================
579
ConfigParser configparser
582
SocketServer socketserver
583
markupbase _markupbase
585
test.test_support test.support
586
======================= =======================
588
* A common pattern in Python 2.x is to have one version of a module
589
implemented in pure Python, with an optional accelerated version
590
implemented as a C extension; for example, :mod:`pickle` and
591
:mod:`cPickle`. This places the burden of importing the accelerated
592
version and falling back on the pure Python version on each user of
593
these modules. In Python 3.0, the accelerated versions are
594
considered implementation details of the pure Python versions.
595
Users should always import the standard version, which attempts to
596
import the accelerated version and falls back to the pure Python
597
version. The :mod:`pickle` / :mod:`cPickle` pair received this
598
treatment. The :mod:`profile` module is on the list for 3.1. The
599
:mod:`StringIO` module has been turned into a class in the :mod:`io`
602
* Some related modules have been grouped into packages, and usually
603
the submodule names have been simplified. The resulting new
606
* :mod:`dbm` (:mod:`anydbm`, :mod:`dbhash`, :mod:`dbm`,
607
:mod:`dumbdbm`, :mod:`gdbm`, :mod:`whichdb`).
609
* :mod:`html` (:mod:`HTMLParser`, :mod:`htmlentitydefs`).
611
* :mod:`http` (:mod:`httplib`, :mod:`BaseHTTPServer`,
612
:mod:`CGIHTTPServer`, :mod:`SimpleHTTPServer`, :mod:`Cookie`,
615
* :mod:`tkinter` (all :mod:`Tkinter`-related modules except
616
:mod:`turtle`). The target audience of :mod:`turtle` doesn't
617
really care about :mod:`tkinter`. Also note that as of Python
618
2.6, the functionality of :mod:`turtle` has been greatly enhanced.
620
* :mod:`urllib` (:mod:`urllib`, :mod:`urllib2`, :mod:`urlparse`,
623
* :mod:`xmlrpc` (:mod:`xmlrpclib`, :mod:`DocXMLRPCServer`,
624
:mod:`SimpleXMLRPCServer`).
626
Some other changes to standard library modules, not covered by
629
* Killed :mod:`sets`. Use the built-in :func:`set` class.
631
* Cleanup of the :mod:`sys` module: removed :func:`sys.exitfunc`,
632
:func:`sys.exc_clear`, :data:`sys.exc_type`, :data:`sys.exc_value`,
633
:data:`sys.exc_traceback`. (Note that :data:`sys.last_type`
636
* Cleanup of the :class:`array.array` type: the :meth:`read` and
637
:meth:`write` methods are gone; use :meth:`fromfile` and
638
:meth:`tofile` instead. Also, the ``'c'`` typecode for array is
639
gone -- use either ``'b'`` for bytes or ``'u'`` for Unicode
642
* Cleanup of the :mod:`operator` module: removed
643
:func:`sequenceIncludes` and :func:`isCallable`.
645
* Cleanup of the :mod:`thread` module: :func:`acquire_lock` and
646
:func:`release_lock` are gone; use :func:`acquire` and
647
:func:`release` instead.
649
* Cleanup of the :mod:`random` module: removed the :func:`jumpahead` API.
651
* The :mod:`new` module is gone.
653
* The functions :func:`os.tmpnam`, :func:`os.tempnam` and
654
:func:`os.tmpfile` have been removed in favor of the :mod:`tempfile`
657
* The :mod:`tokenize` module has been changed to work with bytes. The
658
main entry point is now :func:`tokenize.tokenize`, instead of
661
* :data:`string.letters` and its friends (:data:`string.lowercase` and
662
:data:`string.uppercase`) are gone. Use
663
:data:`string.ascii_letters` etc. instead. (The reason for the
664
removal is that :data:`string.letters` and friends had
665
locale-specific behavior, which is a bad idea for such
666
attractively-named global "constants".)
668
* Renamed module :mod:`__builtin__` to :mod:`builtins` (removing the
669
underscores, adding an 's'). The :data:`__builtins__` variable
670
found in most global namespaces is unchanged. To modify a builtin,
671
you should use :mod:`builtins`, not :data:`__builtins__`!
674
:pep:`3101`: A New Approach To String Formatting
675
================================================
677
* A new system for built-in string formatting operations replaces the
678
``%`` string formatting operator. (However, the ``%`` operator is
679
still supported; it will be deprecated in Python 3.1 and removed
680
from the language at some later time.) Read :pep:`3101` for the full
684
Changes To Exceptions
685
=====================
687
The APIs for raising and catching exception have been cleaned up and
688
new powerful features added:
690
* :pep:`352`: All exceptions must be derived (directly or indirectly)
691
from :exc:`BaseException`. This is the root of the exception
692
hierarchy. This is not new as a recommendation, but the
693
*requirement* to inherit from :exc:`BaseException` is new. (Python
694
2.6 still allowed classic classes to be raised, and placed no
695
restriction on what you can catch.) As a consequence, string
696
exceptions are finally truly and utterly dead.
698
* Almost all exceptions should actually derive from :exc:`Exception`;
699
:exc:`BaseException` should only be used as a base class for
700
exceptions that should only be handled at the top level, such as
701
:exc:`SystemExit` or :exc:`KeyboardInterrupt`. The recommended
702
idiom for handling all exceptions except for this latter category is
703
to use :keyword:`except` :exc:`Exception`.
705
* :exc:`StandardError` was removed.
707
* Exceptions no longer behave as sequences. Use the :attr:`args`
710
* :pep:`3109`: Raising exceptions. You must now use :samp:`raise
711
{Exception}({args})` instead of :samp:`raise {Exception}, {args}`.
712
Additionally, you can no longer explicitly specify a traceback;
713
instead, if you *have* to do this, you can assign directly to the
714
:attr:`__traceback__` attribute (see below).
716
* :pep:`3110`: Catching exceptions. You must now use
717
:samp:`except {SomeException} as {variable}` instead
718
of :samp:`except {SomeException}, {variable}`. Moreover, the
719
*variable* is explicitly deleted when the :keyword:`except` block
722
* :pep:`3134`: Exception chaining. There are two cases: implicit
723
chaining and explicit chaining. Implicit chaining happens when an
724
exception is raised in an :keyword:`except` or :keyword:`finally`
725
handler block. This usually happens due to a bug in the handler
726
block; we call this a *secondary* exception. In this case, the
727
original exception (that was being handled) is saved as the
728
:attr:`__context__` attribute of the secondary exception.
729
Explicit chaining is invoked with this syntax::
731
raise SecondaryException() from primary_exception
733
(where *primary_exception* is any expression that produces an
734
exception object, probably an exception that was previously caught).
735
In this case, the primary exception is stored on the
736
:attr:`__cause__` attribute of the secondary exception. The
737
traceback printed when an unhandled exception occurs walks the chain
738
of :attr:`__cause__` and :attr:`__context__` attributes and prints a
739
separate traceback for each component of the chain, with the primary
740
exception at the top. (Java users may recognize this behavior.)
742
* :pep:`3134`: Exception objects now store their traceback as the
743
:attr:`__traceback__` attribute. This means that an exception
744
object now contains all the information pertaining to an exception,
745
and there are fewer reasons to use :func:`sys.exc_info` (though the
746
latter is not removed).
748
* A few exception messages are improved when Windows fails to load an
749
extension module. For example, ``error code 193`` is now ``%1 is
750
not a valid Win32 application``. Strings now deal with non-English
754
Miscellaneous Other Changes
755
===========================
757
Operators And Special Methods
758
-----------------------------
760
* ``!=`` now returns the opposite of ``==``, unless ``==`` returns
761
:data:`NotImplemented`.
763
* The concept of "unbound methods" has been removed from the language.
764
When referencing a method as a class attribute, you now get a plain
767
* :meth:`__getslice__`, :meth:`__setslice__` and :meth:`__delslice__`
768
were killed. The syntax ``a[i:j]`` now translates to
769
``a.__getitem__(slice(i, j))`` (or :meth:`__setitem__` or
770
:meth:`__delitem__`, when used as an assignment or deletion target,
773
* :pep:`3114`: the standard :meth:`next` method has been renamed to
774
:meth:`~iterator.__next__`.
776
* The :meth:`__oct__` and :meth:`__hex__` special methods are removed
777
-- :func:`oct` and :func:`hex` use :meth:`__index__` now to convert
778
the argument to an integer.
780
* Removed support for :attr:`__members__` and :attr:`__methods__`.
782
* The function attributes named :attr:`func_X` have been renamed to
783
use the :data:`__X__` form, freeing up these names in the function
784
attribute namespace for user-defined attributes. To wit,
785
:attr:`func_closure`, :attr:`func_code`, :attr:`func_defaults`,
786
:attr:`func_dict`, :attr:`func_doc`, :attr:`func_globals`,
787
:attr:`func_name` were renamed to :attr:`__closure__`,
788
:attr:`__code__`, :attr:`__defaults__`, :attr:`~object.__dict__`,
789
:attr:`__doc__`, :attr:`__globals__`, :attr:`~definition.__name__`,
792
* :meth:`__nonzero__` is now :meth:`__bool__`.
797
* :pep:`3135`: New :func:`super`. You can now invoke :func:`super`
798
without arguments and (assuming this is in a regular instance method
799
defined inside a :keyword:`class` statement) the right class and
800
instance will automatically be chosen. With arguments, the behavior
801
of :func:`super` is unchanged.
803
* :pep:`3111`: :func:`raw_input` was renamed to :func:`input`. That
804
is, the new :func:`input` function reads a line from
805
:data:`sys.stdin` and returns it with the trailing newline stripped.
806
It raises :exc:`EOFError` if the input is terminated prematurely.
807
To get the old behavior of :func:`input`, use ``eval(input())``.
809
* A new built-in function :func:`next` was added to call the
810
:meth:`~iterator.__next__` method on an object.
812
* The :func:`round` function rounding strategy and return type have
813
changed. Exact halfway cases are now rounded to the nearest even
814
result instead of away from zero. (For example, ``round(2.5)`` now
815
returns ``2`` rather than ``3``.) ``round(x[, n])`` now
816
delegates to ``x.__round__([n])`` instead of always returning a
817
float. It generally returns an integer when called with a single
818
argument and a value of the same type as ``x`` when called with two
821
* Moved :func:`intern` to :func:`sys.intern`.
823
* Removed: :func:`apply`. Instead of ``apply(f, args)`` use
826
* Removed :func:`callable`. Instead of ``callable(f)`` you can use
827
``isinstance(f, collections.Callable)``. The :func:`operator.isCallable`
828
function is also gone.
830
* Removed :func:`coerce`. This function no longer serves a purpose
831
now that classic classes are gone.
833
* Removed :func:`execfile`. Instead of ``execfile(fn)`` use
834
``exec(open(fn).read())``.
836
* Removed the :class:`file` type. Use :func:`open`. There are now several
837
different kinds of streams that open can return in the :mod:`io` module.
839
* Removed :func:`reduce`. Use :func:`functools.reduce` if you really
840
need it; however, 99 percent of the time an explicit :keyword:`for`
841
loop is more readable.
843
* Removed :func:`reload`. Use :func:`imp.reload`.
845
* Removed. :meth:`dict.has_key` -- use the :keyword:`in` operator
848
.. ======================================================================
851
Build and C API Changes
852
=======================
854
Due to time constraints, here is a *very* incomplete list of changes
857
* Support for several platforms was dropped, including but not limited
858
to Mac OS 9, BeOS, RISCOS, Irix, and Tru64.
860
* :pep:`3118`: New Buffer API.
862
* :pep:`3121`: Extension Module Initialization & Finalization.
864
* :pep:`3123`: Making :c:macro:`PyObject_HEAD` conform to standard C.
866
* No more C API support for restricted execution.
868
* :c:func:`PyNumber_Coerce`, :c:func:`PyNumber_CoerceEx`,
869
:c:func:`PyMember_Get`, and :c:func:`PyMember_Set` C APIs are removed.
871
* New C API :c:func:`PyImport_ImportModuleNoBlock`, works like
872
:c:func:`PyImport_ImportModule` but won't block on the import lock
873
(returning an error instead).
875
* Renamed the boolean conversion C-level slot and method:
876
``nb_nonzero`` is now ``nb_bool``.
878
* Removed :c:macro:`METH_OLDARGS` and :c:macro:`WITH_CYCLE_GC` from the C API.
880
.. ======================================================================
886
The net result of the 3.0 generalizations is that Python 3.0 runs the
887
pystone benchmark around 10% slower than Python 2.5. Most likely the
888
biggest cause is the removal of special-casing for small integers.
889
There's room for improvement, but it will happen after 3.0 is
892
.. ======================================================================
895
Porting To Python 3.0
896
=====================
898
For porting existing Python 2.5 or 2.6 source code to Python 3.0, the
899
best strategy is the following:
901
0. (Prerequisite:) Start with excellent test coverage.
903
1. Port to Python 2.6. This should be no more work than the average
904
port from Python 2.x to Python 2.(x+1). Make sure all your tests
907
2. (Still using 2.6:) Turn on the :option:`!-3` command line switch.
908
This enables warnings about features that will be removed (or
909
change) in 3.0. Run your test suite again, and fix code that you
910
get warnings about until there are no warnings left, and all your
913
3. Run the ``2to3`` source-to-source translator over your source code
914
tree. (See :ref:`2to3-reference` for more on this tool.) Run the
915
result of the translation under Python 3.0. Manually fix up any
916
remaining issues, fixing problems until all tests pass again.
918
It is not recommended to try to write source code that runs unchanged
919
under both Python 2.6 and 3.0; you'd have to use a very contorted
920
coding style, e.g. avoiding ``print`` statements, metaclasses,
921
and much more. If you are maintaining a library that needs to support
922
both Python 2.6 and Python 3.0, the best approach is to modify step 3
923
above by editing the 2.6 version of the source code and running the
924
``2to3`` translator again, rather than editing the 3.0 version of the
927
For porting C extensions to Python 3.0, please see :ref:`cporting-howto`.
929
.. ======================================================================