Python 3.0 有什么新变化

作者

Guido van Rossum

This article explains the new features in Python 3.0, compared to 2.6. Python 3.0, also known as "Python 3000" or "Py3K", is the first ever intentionally backwards incompatible Python release. There are more changes than in a typical release, and more that are important for all Python users. Nevertheless, after digesting the changes, you'll find that Python really hasn't changed all that much -- by and large, we're mostly fixing well-known annoyances and warts, and removing a lot of old cruft.

本文并不试图提供所有新特性的完整规范，而是试图提供一个方便的概述。 要了解完整的细节，您应该参考 Python 3.0 的文档和/或文中引用的许多 PEP。 如果您想了解某个特性的完整实现和设计原理，PEP 通常比常规文档有更多的细节；但要注意的是，一旦某个特性被完全实现，PEP 通常不会保持更新。

由于时间有限，本文档不够完整。 对于新发布的版本，源代码发行版中的 Misc/NEWS 文件总是包含大量关于每一个细小改动的详细信息。

常见的绊脚石

本节列出了在你已习惯了 Python 2.5 的情况下最有可能让您感到困惑的几处更改。

用视图和迭代器取代列表

某些知名的 API 将不再返回列表:

  • dict 方法 dict.keys()、dict.items() 和 dict.values() 返回 “视图” 而不是列表。 例如，这个写法不再有效: k = d.keys(); k.sort()。 请使用 k = sorted(d) 代替（这在 Python 2.5 中也有效，而且同样高效）。

  • 此外，不再支持 dict.iterkeys()、dict.iteritems() 和 dict.itervalues() 方法。

  • map() 和 filter() 均返回迭代器。 如果你确实需要一个列表并且所有输入序列的长度相等，简单的解决办法是将 map() 包装在 list() 中，例如 list(map(...))，但更好的办法通常是使用列表推导式（特别是当原始代码使用了 lambda 的时候），或是重写代码使得它完全不需要列表。 还有一种特殊技巧是将 map() 作为函数的附带影响发起调用；正确的转换方式是使用一个常规的 for 循环（因为创建列表会浪费资源）。

    如果输入序列的长度不相等，map() 将在最短序列的终点停止。 为了与 Python 2.x 中的 map() 完全兼容，也可将序列包装在 itertools.zip_longest() 中，例如将 map(func, *sequences) 变成 list(map(func, itertools.zip_longest(*sequences)))。

  • 现在 range() 的行为与过去 xrange() 的行为类似，区别在于它可以处理任意大小的值。 后者已不复存在。

  • zip() 现在将返回一个迭代器。

排序比较

Python 3.0 简化了排序比较的规则：

  • 当操作数不存在有意义的自然排序时，排序比较操作符 (<, <=, >=, >) 会引发 TypeError 异常。 因此，像 1 < '', 0 > None 或 len <= len 这样的表达式不再有效，例如 None < None 会引发 TypeError 而不是返回 False。 由此推论，对异构列表进行排序不再有意义 —— 所有元素必须相互可比。 请注意，这不适用于 == 和 != 操作符：不同的不可比类型的对象总是互不相等的。

  • builtin.sorted() 和 list.sort() 不再接受提供比较函数的 cmp 参数。 请使用 key 参数。 注意 key 和 reverse 参数现在是“仅限关键字”参数。

  • cmp() 函数应视为已不复存在，而 __cmp__() 特殊方法也不再支持。 请使用 __lt__() 进行排序，使用 __eq__() 与 __hash__() 进行比较，并根据需要使用其他的丰富比较方法。 （如果确实需要 cmp() 功能，可以使用表达式 (a > b) - (a < b) 以实现 cmp(a, b)。）

整数

  • PEP 237: 在实质上，long 已被重命名为 int。 也就是说，现在只有一种内置整数类型，叫做 int；但其行为更像是旧的 long 类型。

  • PEP 238: 像 1/2 这样的表达式将返回一个浮点数。 请使用 1//2 来得到取整的行为。 （后面这种语法已存在多年，至少从 Python 2.2 起就有了。）

  • sys.maxint 常量已被删除，因为整数值不再有限制。 不过，sys.maxsize 仍可用作大于任何实际列表或字符串索引的整数。 它符合实现的“自然”整数大小，通常与同一平台上以前版本中的 sys.maxint 相同（假设使用相同的构建选项）。

  • 长整数的 repr() 不再包括尾部的 L，因此无条件地删除该字符的代码会删除最后一位数字。 （请使用 str() 代替。）

  • 八进制数字面值不再是 0720 的形式；而是改用 0o720 的形式。

文本与数据而不是 Unicode 与 8 比特位

你对二进制数据和 Unicode 的所有认知都已改变。

  • Python 3.0 uses the concepts of text and (binary) data instead of Unicode strings and 8-bit strings. All text is Unicode; however encoded Unicode is represented as binary data. The type used to hold text is str, the type used to hold data is bytes. The biggest difference with the 2.x situation is that any attempt to mix text and data in Python 3.0 raises TypeError, whereas if you were to mix Unicode and 8-bit strings in Python 2.x, it would work if the 8-bit string happened to contain only 7-bit (ASCII) bytes, but you would get UnicodeDecodeError if it contained non-ASCII values. This value-specific behavior has caused numerous sad faces over the years.

  • As a consequence of this change in philosophy, pretty much all code that uses Unicode, encodings or binary data most likely has to change. The change is for the better, as in the 2.x world there were numerous bugs having to do with mixing encoded and unencoded text. To be prepared in Python 2.x, start using unicode for all unencoded text, and str for binary or encoded data only. Then the 2to3 tool will do most of the work for you.

  • 你不能再使用 u"..." 字面值来表示 Unicode 文本。 不过，你必须使用 b"..." 字面值来表示二进制数据。

  • As the str and bytes types cannot be mixed, you must always explicitly convert between them. Use str.encode() to go from str to bytes, and bytes.decode() to go from bytes to str. You can also use bytes(s, encoding=...) and str(b, encoding=...), respectively.

  • 与 str 一样，bytes 类型是不可变的。 还有一个单独的 mutable 类型用于保存缓冲二进制数据，即 bytearray。 几乎所有接受 bytes 的应用程序接口也接受 bytearray。 可变 API 基于 collections.MutableSequence。

  • 原始字符串字面中的所有反斜线均按字面解释。 这意味着原始字符串中的 '\U' 和 '\u' 转义符不会被特殊处理。 例如，在 Python 3.0 中，r'\u20ac' 是一个包含 6 个字符的字符串，而在 2.6 中，ur'\u20ac' 是一个 “欧元” 字符。 （当然，这种变化只影响原始字符串的字面意义；在 Python 3.0 中，欧元字符是 '\u20ac'。）

  • 内置的 basestring 抽象类型已被移除，请使用 str 代替。 str 和 bytes 类型在功能上没有足够的共通性，因此不需要共享基类。 2to3 工具（见下文）用 str 替换了每一个 basestring。

  • Files opened as text files (still the default mode for open()) always use an encoding to map between strings (in memory) and bytes (on disk). Binary files (opened with a b in the mode argument) always use bytes in memory. This means that if a file is opened using an incorrect mode or encoding, I/O will likely fail loudly, instead of silently producing incorrect data. It also means that even Unix users will have to specify the correct mode (text or binary) when opening a file. There is a platform-dependent default encoding, which on Unixy platforms can be set with the LANG environment variable (and sometimes also with some other platform-specific locale-related environment variables). In many cases, but not all, the system default is UTF-8; you should never count on this default. Any application reading or writing more than pure ASCII text should probably have a way to override the encoding. There is no longer any need for using the encoding-aware streams in the codecs module.

  • sys.stdin、sys.stdout 和 sys.stderr 的初始值现在是仅 Unicode 的文本文件（即它们是 io.TextIOBase 的实例）。 要使用这些数据流读写字节数据，需要使用它们的 io.TextIOBase.buffer 属性。

  • Filenames are passed to and returned from APIs as (Unicode) strings. This can present platform-specific problems because on some platforms filenames are arbitrary byte strings. (On the other hand, on Windows filenames are natively stored as Unicode.) As a work-around, most APIs (e.g. open() and many functions in the os module) that take filenames accept bytes objects as well as strings, and a few APIs have a way to ask for a bytes return value. Thus, os.listdir() returns a list of bytes instances if the argument is a bytes instance, and os.getcwdb() returns the current working directory as a bytes instance. Note that when os.listdir() returns a list of strings, filenames that cannot be decoded properly are omitted rather than raising UnicodeError.

  • 当系统提供的字节无法使用默认编码进行解释时，一些系统 API，如 os.environ 和 sys.argv，也会出现问题。 最好的办法可能是设置 LANG 变量并重新运行程序。

  • PEP 3138: 字符串的 repr() 将不再转义非 ASCII 字符。 不过，它仍然会转义控制字符和在 Unicode 标准中具有不可打印状态的码位。

  • PEP 3120：现在默认的源码编码格式是UTF-8。

  • PEP 3131: 现在允许在标识符中使用非 ASCII 字符（不过，标准库中的异常和注释中的贡献者名字仍然只使用 ASCII 字符）。

  • The StringIO and cStringIO modules are gone. Instead, import the io module and use io.StringIO or io.BytesIO for text and data respectively.

  • 另请参阅 Unicode 指南，其内容已针对 Python 3.0 进行更新。

语法变化概述

本节提供了 Python 3.0 中每个 语法 变化的简要说明。

新语法

  • PEP 3107: Function argument and return value annotations. This provides a standardized way of annotating a function's parameters and return value. There are no semantics attached to such annotations except that they can be introspected at runtime using the __annotations__ attribute. The intent is to encourage experimentation through metaclasses, decorators or frameworks.

  • PEP 3102：仅限关键字参数。在参数列表``*args`` 之后出现的命名参数 必须 在调用中使用关键字语法指定。也可以在参数列表中使用``*``来表示不接受长度可变的参数列表，但可以使用只包含关键字的参数。

  • 类定义中的基类列表后允许使用关键字参数。 这是用于指定元类的新约定（见下一节），但也可用于其他目的，只要元类支持它。

  • PEP 3104: nonlocal statement. Using nonlocal x you can now assign directly to a variable in an outer (but non-global) scope. nonlocal is a new reserved word.

  • PEP 3132: Extended Iterable Unpacking. You can now write things like a, b, *rest = some_sequence. And even *rest, a = stuff. The rest object is always a (possibly empty) list; the right-hand side may be any iterable. Example:

    (a, *rest, b) = range(5)

    This sets a to 0, b to 4, and rest to [1, 2, 3].

  • Dictionary comprehensions: {k: v for k, v in stuff} means the same thing as dict(stuff) but is more flexible. (This is PEP 274 vindicated. :-)

  • Set literals, e.g. {1, 2}. Note that {} is an empty dictionary; use set() for an empty set. Set comprehensions are also supported; e.g., {x for x in stuff} means the same thing as set(stuff) but is more flexible.

  • New octal literals, e.g. 0o720 (already in 2.6). The old octal literals (0720) are gone.

  • New binary literals, e.g. 0b1010 (already in 2.6), and there is a new corresponding built-in function, bin().

  • Bytes literals are introduced with a leading b or B, and there is a new corresponding built-in function, bytes().

语法变化

  • PEP 3109 和 PEP 3134: 新增 raise 语句的语法: raise [expr [from expr]]。 见下文。

  • 现在 as 和 with 是保留关键字。 （实际是从 2.6 开始。）

  • True, False 和 None 已成为保留关键字。 （2.6 已经对 None 部分强制应用限制。）

  • 将 except exc, var 改为 except exc as var。 参见 PEP 3110。

  • PEP 3115: 新的元类语法。 替换:

    class C:
    __metaclass__ = M
    ...

    你现在需要使用:

    class C(metaclass=M):
    ...

    不再支持全局模块变量 __metaclass__。 (它是一个“拐杖”，可以使默认使用新风格类变得更容易，而无需从 object 派生每个类。）

  • List comprehensions no longer support the syntactic form [... for var in item1, item2, ...]. Use [... for var in (item1, item2, ...)] instead. Also note that list comprehensions have different semantics: they are closer to syntactic sugar for a generator expression inside a list() constructor, and in particular the loop control variables are no longer leaked into the surrounding scope.

  • ellipsis (...) 可以在任何地方作为原子表达式使用。（以前只允许在片段中使用。）另外，现在 必须 拼写为``...`` 。（以前也可以拼写为``. . .`` ，这只是一个偶然的语法。）

移除的语法

  • PEP 3113: Tuple parameter unpacking removed. You can no longer write def foo(a, (b, c)): .... Use def foo(a, b_c): b, c = b_c instead.

  • Removed backticks (use repr() instead).

  • Removed <> (use != instead).

  • Removed keyword: exec() is no longer a keyword; it remains as a function. (Fortunately the function syntax was also accepted in 2.x.) Also note that exec() no longer takes a stream argument; instead of exec(f) you can use exec(f.read()).

  • Integer literals no longer support a trailing l or L.

  • String literals no longer support a leading u or U.

  • The from module import * syntax is only allowed at the module level, no longer inside functions.

  • The only acceptable syntax for relative imports is from .[module] import name. All import forms not starting with . are interpreted as absolute imports. (PEP 328)

  • Classic classes are gone.

Changes Already Present In Python 2.6

Since many users presumably make the jump straight from Python 2.5 to Python 3.0, this section reminds the reader of new features that were originally designed for Python 3.0 but that were back-ported to Python 2.6. The corresponding sections in Python 2.6 有什么新变化 should be consulted for longer descriptions.

库的修改

Due to time constraints, this document does not exhaustively cover the very extensive changes to the standard library. PEP 3108 is the reference for the major changes to the library. Here's a capsule review:

  • Many old modules were removed. Some, like gopherlib (no longer used) and md5 (replaced by hashlib), were already deprecated by PEP 4. Others were removed as a result of the removal of support for various platforms such as Irix, BeOS and Mac OS 9 (see PEP 11). Some modules were also selected for removal in Python 3.0 due to lack of use or because a better replacement exists. See PEP 3108 for an exhaustive list.

  • The bsddb3 package was removed because its presence in the core standard library has proved over time to be a particular burden for the core developers due to testing instability and Berkeley DB's release schedule. However, the package is alive and well, externally maintained at https://www.jcea.es/programacion/pybsddb.htm.

  • Some modules were renamed because their old name disobeyed PEP 8, or for various other reasons. Here's the list:

    旧名称

    新名称

    _winreg

    winreg

    ConfigParser

    configparser

    copy_reg

    copyreg

    Queue

    queue

    SocketServer

    socketserver

    markupbase

    _markupbase

    repr

    reprlib

    test.test_support

    test.support

  • A common pattern in Python 2.x is to have one version of a module implemented in pure Python, with an optional accelerated version implemented as a C extension; for example, pickle and cPickle. This places the burden of importing the accelerated version and falling back on the pure Python version on each user of these modules. In Python 3.0, the accelerated versions are considered implementation details of the pure Python versions. Users should always import the standard version, which attempts to import the accelerated version and falls back to the pure Python version. The pickle / cPickle pair received this treatment. The profile module is on the list for 3.1. The StringIO module has been turned into a class in the io module.

  • Some related modules have been grouped into packages, and usually the submodule names have been simplified. The resulting new packages are:

    • dbm (anydbm, dbhash, dbm, dumbdbm, gdbm, whichdb).

    • html (HTMLParser, htmlentitydefs).

    • http (httplib, BaseHTTPServer, CGIHTTPServer, SimpleHTTPServer, Cookie, cookielib).

    • tkinter (all Tkinter-related modules except turtle). The target audience of turtle doesn't really care about tkinter. Also note that as of Python 2.6, the functionality of turtle has been greatly enhanced.

    • urllib (urllib, urllib2, urlparse, robotparse).

    • xmlrpc (xmlrpclib, DocXMLRPCServer, SimpleXMLRPCServer).

Some other changes to standard library modules, not covered by PEP 3108:

  • Killed sets. Use the built-in set() class.

  • Cleanup of the sys module: removed sys.exitfunc(), sys.exc_clear(), sys.exc_type, sys.exc_value, sys.exc_traceback. (Note that sys.last_type etc. remain.)

  • Cleanup of the array.array type: the read() and write() methods are gone; use fromfile() and tofile() instead. Also, the 'c' typecode for array is gone -- use either 'b' for bytes or 'u' for Unicode characters.

  • Cleanup of the operator module: removed sequenceIncludes() and isCallable().

  • Cleanup of the thread module: acquire_lock() and release_lock() are gone; use acquire() and release() instead.

  • Cleanup of the random module: removed the jumpahead() API.

  • The new module is gone.

  • The functions os.tmpnam(), os.tempnam() and os.tmpfile() have been removed in favor of the tempfile module.

  • The tokenize module has been changed to work with bytes. The main entry point is now tokenize.tokenize(), instead of generate_tokens.

  • string.letters and its friends (string.lowercase and string.uppercase) are gone. Use string.ascii_letters etc. instead. (The reason for the removal is that string.letters and friends had locale-specific behavior, which is a bad idea for such attractively-named global "constants".)

  • Renamed module __builtin__ to builtins (removing the underscores, adding an 's'). The __builtins__ variable found in most global namespaces is unchanged. To modify a builtin, you should use builtins, not __builtins__!

PEP 3101: A New Approach To String Formatting

  • A new system for built-in string formatting operations replaces the % string formatting operator. (However, the % operator is still supported; it will be deprecated in Python 3.1 and removed from the language at some later time.) Read PEP 3101 for the full scoop.

Changes To Exceptions

The APIs for raising and catching exception have been cleaned up and new powerful features added:

  • PEP 352: All exceptions must be derived (directly or indirectly) from BaseException. This is the root of the exception hierarchy. This is not new as a recommendation, but the requirement to inherit from BaseException is new. (Python 2.6 still allowed classic classes to be raised, and placed no restriction on what you can catch.) As a consequence, string exceptions are finally truly and utterly dead.

  • Almost all exceptions should actually derive from Exception; BaseException should only be used as a base class for exceptions that should only be handled at the top level, such as SystemExit or KeyboardInterrupt. The recommended idiom for handling all exceptions except for this latter category is to use except Exception.

  • StandardError was removed.

  • Exceptions no longer behave as sequences. Use the args attribute instead.

  • PEP 3109: Raising exceptions. You must now use raise Exception(args) instead of raise Exception, args. Additionally, you can no longer explicitly specify a traceback; instead, if you have to do this, you can assign directly to the __traceback__ attribute (see below).

  • PEP 3110: Catching exceptions. You must now use except SomeException as variable instead of except SomeException, variable. Moreover, the variable is explicitly deleted when the except block is left.

  • PEP 3134: Exception chaining. There are two cases: implicit chaining and explicit chaining. Implicit chaining happens when an exception is raised in an except or finally handler block. This usually happens due to a bug in the handler block; we call this a secondary exception. In this case, the original exception (that was being handled) is saved as the __context__ attribute of the secondary exception. Explicit chaining is invoked with this syntax:

    raise SecondaryException() from primary_exception

    (where primary_exception is any expression that produces an exception object, probably an exception that was previously caught). In this case, the primary exception is stored on the __cause__ attribute of the secondary exception. The traceback printed when an unhandled exception occurs walks the chain of __cause__ and __context__ attributes and prints a separate traceback for each component of the chain, with the primary exception at the top. (Java users may recognize this behavior.)

  • PEP 3134: Exception objects now store their traceback as the __traceback__ attribute. This means that an exception object now contains all the information pertaining to an exception, and there are fewer reasons to use sys.exc_info() (though the latter is not removed).

  • A few exception messages are improved when Windows fails to load an extension module. For example, error code 193 is now %1 is not a valid Win32 application. Strings now deal with non-English locales.

Miscellaneous Other Changes

Operators And Special Methods

  • != now returns the opposite of ==, unless == returns NotImplemented.

  • The concept of "unbound methods" has been removed from the language. When referencing a method as a class attribute, you now get a plain function object.

  • __getslice__(), __setslice__() and __delslice__() were killed. The syntax a[i:j] now translates to a.__getitem__(slice(i, j)) (or __setitem__() or __delitem__(), when used as an assignment or deletion target, respectively).

  • PEP 3114: the standard next() method has been renamed to __next__().

  • The __oct__() and __hex__() special methods are removed -- oct() and hex() use __index__() now to convert the argument to an integer.

  • Removed support for __members__ and __methods__.

  • The function attributes named func_X have been renamed to use the __X__ form, freeing up these names in the function attribute namespace for user-defined attributes. To wit, func_closure, func_code, func_defaults, func_dict, func_doc, func_globals, func_name were renamed to __closure__, __code__, __defaults__, __dict__, __doc__, __globals__, __name__, respectively.

  • __nonzero__() is now __bool__().

Builtins

  • PEP 3135: New super(). You can now invoke super() without arguments and (assuming this is in a regular instance method defined inside a class statement) the right class and instance will automatically be chosen. With arguments, the behavior of super() is unchanged.

  • PEP 3111: raw_input() was renamed to input(). That is, the new input() function reads a line from sys.stdin and returns it with the trailing newline stripped. It raises EOFError if the input is terminated prematurely. To get the old behavior of input(), use eval(input()).

  • A new built-in function next() was added to call the __next__() method on an object.

  • The round() function rounding strategy and return type have changed. Exact halfway cases are now rounded to the nearest even result instead of away from zero. (For example, round(2.5) now returns 2 rather than 3.) round(x[, n]) now delegates to x.__round__([n]) instead of always returning a float. It generally returns an integer when called with a single argument and a value of the same type as x when called with two arguments.

  • Moved intern() to sys.intern().

  • Removed: apply(). Instead of apply(f, args) use f(*args).

  • Removed callable(). Instead of callable(f) you can use isinstance(f, collections.Callable). The operator.isCallable() function is also gone.

  • Removed coerce(). This function no longer serves a purpose now that classic classes are gone.

  • Removed execfile(). Instead of execfile(fn) use exec(open(fn).read()).

  • Removed the file type. Use open(). There are now several different kinds of streams that open can return in the io module.

  • Removed reduce(). Use functools.reduce() if you really need it; however, 99 percent of the time an explicit for loop is more readable.

  • Removed reload(). Use imp.reload().

  • Removed. dict.has_key() -- use the in operator instead.

构建和 C API 的改变

Due to time constraints, here is a very incomplete list of changes to the C API.

  • Support for several platforms was dropped, including but not limited to Mac OS 9, BeOS, RISCOS, Irix, and Tru64.

  • PEP 3118: 新的缓冲区 API。

  • PEP 3121: Extension Module Initialization & Finalization.

  • PEP 3123: Making PyObject_HEAD conform to standard C.

  • No more C API support for restricted execution.

  • PyNumber_Coerce(), PyNumber_CoerceEx(), PyMember_Get(), and PyMember_Set() C API 已被移除。

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

  • Renamed the boolean conversion C-level slot and method: nb_nonzero is now nb_bool.

  • 从 C API 中移除 METH_OLDARGS 和 WITH_CYCLE_GC。

性能

The net result of the 3.0 generalizations is that Python 3.0 runs the pystone benchmark around 10% slower than Python 2.5. Most likely the biggest cause is the removal of special-casing for small integers. There's room for improvement, but it will happen after 3.0 is released!

移植到 Python 3.0

For porting existing Python 2.5 or 2.6 source code to Python 3.0, the best strategy is the following:

  1. (Prerequisite:) Start with excellent test coverage.

  2. Port to Python 2.6. This should be no more work than the average port from Python 2.x to Python 2.(x+1). Make sure all your tests pass.

  3. (Still using 2.6:) Turn on the -3 command line switch. This enables warnings about features that will be removed (or change) in 3.0. Run your test suite again, and fix code that you get warnings about until there are no warnings left, and all your tests still pass.

  4. Run the 2to3 source-to-source translator over your source code tree. (See 2to3 - 自动将 Python 2 代码转为 Python 3 代码 for more on this tool.) Run the result of the translation under Python 3.0. Manually fix up any remaining issues, fixing problems until all tests pass again.

It is not recommended to try to write source code that runs unchanged under both Python 2.6 and 3.0; you'd have to use a very contorted coding style, e.g. avoiding print statements, metaclasses, and much more. If you are maintaining a library that needs to support both Python 2.6 and Python 3.0, the best approach is to modify step 3 above by editing the 2.6 version of the source code and running the 2to3 translator again, rather than editing the 3.0 version of the source code.

For porting C extensions to Python 3.0, please see 将扩展模块移植到 Python 3.