python: combine sort-key-functions itemgetter and

问题:

I want to sort a list of dictionaries by dictionary key, where I don't want to distinguish between upper and lower case characters.

dict1 = {'name':'peter','phone':'12355'}
dict2 = {'name':'Paul','phone':'545435'}
dict3 = {'name':'klaus','phone':'55345'}
dict4 = {'name':'Krishna','phone':'12345'}
dict5 = {'name':'Ali','phone':'53453'}
dict6 = {'name':'Hans','phone':'765756'}
list_of_dicts = [dict1,dict2,dict3,dict4,dict5,dict6]

key_field = 'name'
list_of_dicts.sort(key=itemgetter(key_field))
# how to combine key=itemgetter(key_field) and key=str.lower? 
for list_field in list_of_dicts:
    print list_field[key_field]

should provide

Ali, Hans, klaus, Krishna, Paul, peter

and not

klaus, peter, Ali, Hans, Krishna, Paul

回答1:

In the general case, you'll want to write your key-extraction function for sorting purposes; only in special (though important) cases it happens that you can just reuse an existing callable to extract the keys for you, or just conjoin a couple of existing ones (in a "quick and dirty" way using lambda, since there's no built-in way to do function composition).

If you often need to perform these two kinds of operations for key extraction (get an item and call a method on that item), I suggest:

def combiner(itemkey, methodname, *a, **k):
  def keyextractor(container):
    item = container[itemkey]
    method = getattr(item, methodname)
    return method(*a, **k)
  return keyextractor

so listofdicts.sort(key=combiner('name', 'lower')) will work in your case.

Note that while excessive generalization has costs, tasteful and moderate generalization (leaving the item key, method name, and method arguments if any, as runtime-determined, in this case) generally has benefits -- one general function, not more complex than a dozen specific and specialized ones (with the extractor, method to call, or both, hardwired in their code), will be easier to maintain (and, of course, much easier to reuse!-).

回答2:

How about this:

list_of_dicts.sort(key=lambda a: a['name'].lower())

回答3:

from functools import partial

def nested_funcs(*funcs):
    return partial(reduce, lambda arg, func: func(arg), funcs)


sorted(list_of_dicts, key=nested_funcs(itemgetter('name'), str.strip, str.lower))

回答4:

You probably should go with a lambda for the sake of readability. But as an interesting study into higher order functions, here's the extended version of q-combinator in Python (also known as the queer bird combinator). This allows you to create a new function by composing two functions

 def compose(inner_func, *outer_funcs):
     if not outer_funcs:
         return inner_func
     outer_func = compose(*outer_funcs)
     return lambda *args, **kwargs: outer_func(inner_func(*args, **kwargs))

 from operator import itemgetter, methodcaller
 name_lowered = compose(itemgetter('name'), methodcaller('lower'))
 print(name_lowered( {'name': 'Foo'} ))

If you reverse the definitions of inner and outer in the compose function, you get the more traditional b-combinator (bluebird). I like the q-combinator more because of the similarity to unix pipes.

回答5:

This solution will use your system locale, and as a bonus, it will sort eventual other characters according to the current locale as well (Will put "ü" after "u" in a german locale etc).

from locale import setlocale, strxfrm, LC_ALL
import operator

# call setlocale to init current locale
setlocale(LC_ALL, "")

def locale_keyfunc(keyfunc):  
  def locale_wrapper(obj):
    return strxfrm(keyfunc(obj))
  return locale_wrapper

list_of_dicts.sort(key=locale_keyfunc(operator.itemgetter("name")))

This of course uses that the locale sort is the User interface "natural" sort that you wish to emulate with .lower().

I'm amazed that python's locale module is unknown and unused, it for sure is an important component in the application I write (translated to multiple languages, but the locale module is important for even getting one module right. Case in point: in swedish 'V' and 'W' sort alike, so you have to collate them. locale does all that for you.). In the POSIX locale (not default), this will revert to sorting "a" after "Z".

回答6:

Personally, I wish there were two functions in the Python standard library (probably in functools):

def compose(*funcs):
    """
    Compose any number of unary functions into a single unary
    function.

    >>> import textwrap
    >>> str.strip(textwrap.dedent(compose.__doc__)) == compose(str.strip, textwrap.dedent)(compose.__doc__)
    True
    """

    compose_two = lambda f1, f2: lambda v: f1(f2(v))
    return reduce(compose_two, funcs)

def method_caller(method_name, *args, **kwargs):
    """
    Return a function that will call a named method on the
    target object with optional positional and keyword
    arguments.

    >>> lower = method_caller('lower')
    >>> lower('MyString')
    'mystring'
    """
    def call_method(target):
        func = getattr(target, method_name)
        return func(*args, **kwargs)
    return call_method

I have implemented these for my own use in jaraco.util.functools.

Either way, now your code is quite clear, self-documented, and robust (IMO).

lower = method_caller('lower')
get_name = itemgetter('name')
lowered_name = compose(lower, get_name)

list_of_dicts.sort(key=lowered_name)

回答7:

def lower_getter(field):
    def _getter(obj):
        return obj[field].lower()
    return _getter

list_of_dicts.sort(key=lower_getter(key_field))