很可能functools.cmp_to_key()与python的sort的底层实现密切相关。此外，cmp参数是遗留的。现代的方法是将输入项转换为支持所需的丰富比较操作的对象。

在CPython 2下。X，即使没有实现各自的富比较操作符，也可以对不同类型的对象排序。在CPython 3下。X，不同类型的对象必须显式地支持比较。参见Python如何比较字符串和int?链接到官方文件。大多数答案都依赖于这种隐含的顺序。切换到Python 3。X将需要一个新的类型来实现和统一数字和字符串之间的比较。

Python 2.7.12 (default, Sep 29 2016, 13:30:34) 
>>> (0,"foo") < ("foo",0)
True  

Python 3.5.2 (default, Oct 14 2016, 12:54:53) 
>>> (0,"foo") < ("foo",0)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  TypeError: unorderable types: int() < str()

有三种不同的方法。第一种使用嵌套类来利用Python的Iterable比较算法。第二个函数将这个嵌套展开到单个类中。第三种方法放弃了str的子类化，专注于性能。所有都是有时间的;第二辆快了一倍，第三辆快了近六倍。对str进行子类化并不是必需的，而且首先可能是一个坏主意，但它确实带来了某些便利。

排序字符被复制以强制按大小写排序，并交换大小写以强制小写字母优先排序;这就是“自然排序”的典型定义。我无法决定分组的类型;有些人可能更喜欢以下选项，这也会带来显著的性能优势:

d = lambda s: s.lower()+s.swapcase()

在使用的地方，比较运算符被设置为object的运算符，这样它们就不会被functools. total_orders忽略。

import functools
import itertools


@functools.total_ordering
class NaturalStringA(str):
    def __repr__(self):
        return "{}({})".format\
            ( type(self).__name__
            , super().__repr__()
            )
    d = lambda c, s: [ c.NaturalStringPart("".join(v))
                        for k,v in
                       itertools.groupby(s, c.isdigit)
                     ]
    d = classmethod(d)
    @functools.total_ordering
    class NaturalStringPart(str):
        d = lambda s: "".join(c.lower()+c.swapcase() for c in s)
        d = staticmethod(d)
        def __lt__(self, other):
            if not isinstance(self, type(other)):
                return NotImplemented
            try:
                return int(self) < int(other)
            except ValueError:
                if self.isdigit():
                    return True
                elif other.isdigit():
                    return False
                else:
                    return self.d(self) < self.d(other)
        def __eq__(self, other):
            if not isinstance(self, type(other)):
                return NotImplemented
            try:
                return int(self) == int(other)
            except ValueError:
                if self.isdigit() or other.isdigit():
                    return False
                else:
                    return self.d(self) == self.d(other)
        __le__ = object.__le__
        __ne__ = object.__ne__
        __gt__ = object.__gt__
        __ge__ = object.__ge__
    def __lt__(self, other):
        return self.d(self) < self.d(other)
    def __eq__(self, other):
        return self.d(self) == self.d(other)
    __le__ = object.__le__
    __ne__ = object.__ne__
    __gt__ = object.__gt__
    __ge__ = object.__ge__

import functools
import itertools


@functools.total_ordering
class NaturalStringB(str):
    def __repr__(self):
        return "{}({})".format\
            ( type(self).__name__
            , super().__repr__()
            )
    d = lambda s: "".join(c.lower()+c.swapcase() for c in s)
    d = staticmethod(d)
    def __lt__(self, other):
        if not isinstance(self, type(other)):
            return NotImplemented
        groups = map(lambda i: itertools.groupby(i, type(self).isdigit), (self, other))
        zipped = itertools.zip_longest(*groups)
        for s,o in zipped:
            if s is None:
                return True
            if o is None:
                return False
            s_k, s_v = s[0], "".join(s[1])
            o_k, o_v = o[0], "".join(o[1])
            if s_k and o_k:
                s_v, o_v = int(s_v), int(o_v)
                if s_v == o_v:
                    continue
                return s_v < o_v
            elif s_k:
                return True
            elif o_k:
                return False
            else:
                s_v, o_v = self.d(s_v), self.d(o_v)
                if s_v == o_v:
                    continue
                return s_v < o_v
        return False
    def __eq__(self, other):
        if not isinstance(self, type(other)):
            return NotImplemented
        groups = map(lambda i: itertools.groupby(i, type(self).isdigit), (self, other))
        zipped = itertools.zip_longest(*groups)
        for s,o in zipped:
            if s is None or o is None:
                return False
            s_k, s_v = s[0], "".join(s[1])
            o_k, o_v = o[0], "".join(o[1])
            if s_k and o_k:
                s_v, o_v = int(s_v), int(o_v)
                if s_v == o_v:
                    continue
                return False
            elif s_k or o_k:
                return False
            else:
                s_v, o_v = self.d(s_v), self.d(o_v)
                if s_v == o_v:
                    continue
                return False
        return True
    __le__ = object.__le__
    __ne__ = object.__ne__
    __gt__ = object.__gt__
    __ge__ = object.__ge__

import functools
import itertools
import enum


class OrderingType(enum.Enum):
    PerWordSwapCase         = lambda s: s.lower()+s.swapcase()
    PerCharacterSwapCase    = lambda s: "".join(c.lower()+c.swapcase() for c in s)


class NaturalOrdering:
    @classmethod
    def by(cls, ordering):
        def wrapper(string):
            return cls(string, ordering)
        return wrapper
    def __init__(self, string, ordering=OrderingType.PerCharacterSwapCase):
        self.string = string
        self.groups = [ (k,int("".join(v)))
                            if k else
                        (k,ordering("".join(v)))
                            for k,v in
                        itertools.groupby(string, str.isdigit)
                      ]
    def __repr__(self):
        return "{}({})".format\
            ( type(self).__name__
            , self.string
            )
    def __lesser(self, other, default):
        if not isinstance(self, type(other)):
            return NotImplemented
        for s,o in itertools.zip_longest(self.groups, other.groups):
            if s is None:
                return True
            if o is None:
                return False
            s_k, s_v = s
            o_k, o_v = o
            if s_k and o_k:
                if s_v == o_v:
                    continue
                return s_v < o_v
            elif s_k:
                return True
            elif o_k:
                return False
            else:
                if s_v == o_v:
                    continue
                return s_v < o_v
        return default
    def __lt__(self, other):
        return self.__lesser(other, default=False)
    def __le__(self, other):
        return self.__lesser(other, default=True)
    def __eq__(self, other):
        if not isinstance(self, type(other)):
            return NotImplemented
        for s,o in itertools.zip_longest(self.groups, other.groups):
            if s is None or o is None:
                return False
            s_k, s_v = s
            o_k, o_v = o
            if s_k and o_k:
                if s_v == o_v:
                    continue
                return False
            elif s_k or o_k:
                return False
            else:
                if s_v == o_v:
                    continue
                return False
        return True
    # functools.total_ordering doesn't create single-call wrappers if both
    # __le__ and __lt__ exist, so do it manually.
    def __gt__(self, other):
        op_result = self.__le__(other)
        if op_result is NotImplemented:
            return op_result
        return not op_result
    def __ge__(self, other):
        op_result = self.__lt__(other)
        if op_result is NotImplemented:
            return op_result
        return not op_result
    # __ne__ is the only implied ordering relationship, it automatically
    # delegates to __eq__

>>> import natsort
>>> import timeit
>>> l1 = ['Apple', 'corn', 'apPlE', 'arbour', 'Corn', 'Banana', 'apple', 'banana']
>>> l2 = list(map(str, range(30)))
>>> l3 = ["{} {}".format(x,y) for x in l1 for y in l2]
>>> print(timeit.timeit('sorted(l3+["0"], key=NaturalStringA)', number=10000, globals=globals()))
362.4729259099986
>>> print(timeit.timeit('sorted(l3+["0"], key=NaturalStringB)', number=10000, globals=globals()))
189.7340817489967
>>> print(timeit.timeit('sorted(l3+["0"], key=NaturalOrdering.by(OrderingType.PerCharacterSwapCase))', number=10000, globals=globals()))
69.34636392899847
>>> print(timeit.timeit('natsort.natsorted(l3+["0"], alg=natsort.ns.GROUPLETTERS | natsort.ns.LOWERCASEFIRST)', number=10000, globals=globals()))
98.2531585780016

自然排序既相当复杂，又定义模糊。不要忘记事先运行unicodedata.normalize(…)，并考虑使用str.casefold()而不是str.lower()。可能有一些微妙的编码问题我还没有考虑到。因此，我尝试性地推荐natsort库。我快速浏览了一下github存储库;代码维护非常出色。

All the algorithms I've seen depend on tricks such as duplicating and lowering characters, and swapping case. While this doubles the running time, an alternative would require a total natural ordering on the input character set. I don't think this is part of the unicode specification, and since there are many more unicode digits than [0-9], creating such a sorting would be equally daunting. If you want locale-aware comparisons, prepare your strings with locale.strxfrm per Python's Sorting HOW TO.

本职位的价值

我的观点是提供一个可以普遍应用的非正则表达式解决方案。 我将创建三个函数:

find_first_digit，这是我从@AnuragUniyal借来的。它将查找字符串中第一个数字或非数字的位置。 Split_digits是一个生成器，它将字符串分成数字块和非数字块。当它是数字时，它也会产生整数。 Natural_key只是将split_digits包装成一个元组。这是我们用来排序，最大，最小的键。

功能

def find_first_digit(s, non=False):
    for i, x in enumerate(s):
        if x.isdigit() ^ non:
            return i
    return -1

def split_digits(s, case=False):
    non = True
    while s:
        i = find_first_digit(s, non)
        if i == 0:
            non = not non
        elif i == -1:
            yield int(s) if s.isdigit() else s if case else s.lower()
            s = ''
        else:
            x, s = s[:i], s[i:]
            yield int(x) if x.isdigit() else x if case else x.lower()

def natural_key(s, *args, **kwargs):
    return tuple(split_digits(s, *args, **kwargs))

我们可以看到它是一般的，因为我们可以有多个数字块:

# Note that the key has lower case letters
natural_key('asl;dkfDFKJ:sdlkfjdf809lkasdjfa_543_hh')

('asl;dkfdfkj:sdlkfjdf', 809, 'lkasdjfa_', 543, '_hh')

或保留大小写敏感:

natural_key('asl;dkfDFKJ:sdlkfjdf809lkasdjfa_543_hh', True)

('asl;dkfDFKJ:sdlkfjdf', 809, 'lkasdjfa_', 543, '_hh')

我们可以看到它以适当的顺序对OP的列表进行排序

sorted(
    ['elm0', 'elm1', 'Elm2', 'elm9', 'elm10', 'Elm11', 'Elm12', 'elm13'],
    key=natural_key
)

['elm0', 'elm1', 'Elm2', 'elm9', 'elm10', 'Elm11', 'Elm12', 'elm13']

但它也可以处理更复杂的列表:

sorted(
    ['f_1', 'e_1', 'a_2', 'g_0', 'd_0_12:2', 'd_0_1_:2'],
    key=natural_key
)

['a_2', 'd_0_1_:2', 'd_0_12:2', 'e_1', 'f_1', 'g_0']

我的正则表达式等价于

def int_maybe(x):
    return int(x) if str(x).isdigit() else x

def split_digits_re(s, case=False):
    parts = re.findall('\d+|\D+', s)
    if not case:
        return map(int_maybe, (x.lower() for x in parts))
    else:
        return map(int_maybe, parts)
    
def natural_key_re(s, *args, **kwargs):
    return tuple(split_digits_re(s, *args, **kwargs))

下面是马克·拜尔回答的一个更加python化的版本:

import re

def natural_sort_key(s, _nsre=re.compile('([0-9]+)')):
    return [int(text) if text.isdigit() else text.lower()
            for text in _nsre.split(s)]

现在这个函数可以在任何使用它的函数中用作键，比如list。Sort, sorted, max，等等。

作为lambda:

lambda s: [int(t) if t.isdigit() else t.lower() for t in re.split('(\d+)', s)]

完全可重复的演示代码:

import re
natsort = lambda s: [int(t) if t.isdigit() else t.lower() for t in re.split('(\d+)', s)]
L = ["a1", "a10", "a11", "a2", "a22", "a3"]   
print(sorted(L, key=natsort))  
# ['a1', 'a2', 'a3', 'a10', 'a11', 'a22'] 

让我就这一需求提出自己的看法:

from typing import Tuple, Union, Optional, Generator


StrOrInt = Union[str, int]


# On Python 3.6, string concatenation is REALLY fast
# Tested myself, and this fella also tested:
# https://blog.ganssle.io/articles/2019/11/string-concat.html
def griter(s: str) -> Generator[StrOrInt, None, None]:
    last_was_digit: Optional[bool] = None
    cluster: str = ""
    for c in s:
        if last_was_digit is None:
            last_was_digit = c.isdigit()
            cluster += c
            continue
        if c.isdigit() != last_was_digit:
            if last_was_digit:
                yield int(cluster)
            else:
                yield cluster
            last_was_digit = c.isdigit()
            cluster = ""
        cluster += c
    if last_was_digit:
        yield int(cluster)
    else:
        yield cluster
    return


def grouper(s: str) -> Tuple[StrOrInt, ...]:
    return tuple(griter(s))

现在如果我们有这样的列表:

filelist = [
    'File3', 'File007', 'File3a', 'File10', 'File11', 'File1', 'File4', 'File5',
    'File9', 'File8', 'File8b1', 'File8b2', 'File8b11', 'File6'
]

我们可以简单地使用key= kwarg来进行自然排序:

>>> sorted(filelist, key=grouper)
['File1', 'File3', 'File3a', 'File4', 'File5', 'File6', 'File007', 'File8', 
'File8b1', 'File8b2', 'File8b11', 'File9', 'File10', 'File11']

当然，这里的缺点是，就像现在一样，该函数将对大写字母在小写字母之前进行排序。

我将把不区分大小写的grouper的实现留给读者:-)

克劳狄对马克·拜尔斯的回答的改进;-)

import re

def natural_sort_key(s, _re=re.compile(r'(\d+)')):
    return [int(t) if i & 1 else t.lower() for i, t in enumerate(_re.split(s))]

...
my_naturally_sorted_list = sorted(my_list, key=natural_sort_key)

顺便说一句，也许不是每个人都记得函数参数默认值是在def时求值的

我写了一个基于http://www.codinghorror.com/blog/2007/12/sorting-for-humans-natural-sort-order.html的函数，它增加了传递自己的“键”参数的能力。我需要这样才能执行包含更复杂对象(不仅仅是字符串)的列表的自然排序。

import re

def natural_sort(list, key=lambda s:s):
    """
    Sort the list into natural alphanumeric order.
    """
    def get_alphanum_key_func(key):
        convert = lambda text: int(text) if text.isdigit() else text 
        return lambda s: [convert(c) for c in re.split('([0-9]+)', key(s))]
    sort_key = get_alphanum_key_func(key)
    list.sort(key=sort_key)

例如:

my_list = [{'name':'b'}, {'name':'10'}, {'name':'a'}, {'name':'1'}, {'name':'9'}]
natural_sort(my_list, key=lambda x: x['name'])
print my_list
[{'name': '1'}, {'name': '9'}, {'name': '10'}, {'name': 'a'}, {'name': 'b'}]