Class factory to produce simple struct-like classe

问题:

While investigating Ruby I came across this to create a simple Struct-like class:

Person = Struct.new(:forname, :surname)
person1 = Person.new('John', 'Doe')
puts person1  #<struct Person forname="John", surname="Doe">

Which raised a few Python questions for me. I have written a [VERY] basic clone of this mechanism in Python:

def Struct(*args):
    class NewStruct:
        def __init__(self):
            for arg in args:
                self.__dict__[arg] = None

    return NewStruct

>>> Person = Struct('forename', 'surname')
>>> person1 = Person()
>>> person2 = Person()
>>> person1.forename, person1.surname = 'John','Doe'
>>> person2.forename, person2.surname = 'Foo','Bar'
>>> person1.forename
'John'
>>> person2.forename
'Foo'

1. Is there already a similar mechanism in Python to handle this? (I usually just use dictionaries).

2. How would I get the Struct() function to create the correct __init__() arguments. (in this case I would like to perform person1 = Person('John', 'Doe') Named Arguments if possible: person1 = Person(surname='Doe', forename='John')

I Would like, as a matter of interest, to have Question 2 answered even if there is a better Python mechanism to do this.

回答1:

If you're using Python 2.6, try the standard library namedtuple class.

>>> from collections import namedtuple
>>> Person = namedtuple('Person', ('forename', 'surname'))
>>> person1 = Person('John', 'Doe')
>>> person2 = Person(forename='Adam', surname='Monroe')
>>> person1.forename
'John'
>>> person2.surname
'Monroe'

Edit: As per comments, there is a backport for earlier versions of Python

回答2:

If you're running python <2.6 or would like to extend your class to do more stuff, I would suggest using the type() builtin. This has the advantage over your solution in that the setting up of __dict__ happens at class creation rather than instantiation. It also doesn't define an __init__ method and thus doesn't lead to strange behavior if the class calls __init__ again for some reason. For example:

def Struct(*args, **kwargs):
    name = kwargs.pop("name", "MyStruct")
    kwargs.update(dict((k, None) for k in args))
    return type(name, (object,), kwargs)

Used like so:

>>> MyStruct = Struct("forename", "lastname")

Equivalent to:

class MyStruct(object):
    forename = None
    lastname = None

While this:

>>> TestStruct = Struct("forename", age=18, name="TestStruct")

Is equivalent to:

class TestStruct(object):
    forename = None
    age = 18

Update

Additionally, you can edit this code to very easily prevent assignment of other variables than the ones specificed. Just change the Struct() factory to assign __slots__.

def Struct(*args, **kwargs):
    name = kwargs.pop("name", "MyStruct")
    kwargs.update(dict((k, None) for k in args))
    kwargs['__slots__'] = kwargs.keys()
    return type(name, (object,), kwargs)

回答3:

As others have said, named tuples in Python 2.6/3.x. With older versions, I usually use the Stuff class:

class Stuff(object):
    def __init__(self, **kwargs):
        self.__dict__.update(kwargs)

john = Stuff(forename='John', surname='Doe')

This doesn't protect you from mispellings though. There's also a recipe for named tuples on ActiveState:

http://code.activestate.com/recipes/500261/

回答4:

This is following up on Cide's answer (and probably only interesting for people who want to dig deeper).

I experienced a problem using Cide's updated definition of Struct(), the one using __slots__. The problem is that instances of returned classes have read-only attributes:

>>> MS = Struct('forename','lastname')
>>> m=MS()
>>> m.forename='Jack'
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: 'MyStruct' object attribute 'forename' is read-only

Seems that __slots__ is blocking instance-level attributes when there are class attributes of same names. I've tried to overcome this by providing an __init__ method, so instance attributes can be set at object creation time:

def Struct1(*args, **kwargs):
    def init(self):
        for k,v in kwargs.items():
            setattr(self, k, v)
    name = kwargs.pop("name", "MyStruct")
    kwargs.update(dict((k, None) for k in args))
    return type(name, (object,), {'__init__': init, '__slots__': kwargs.keys()})

As a net effect the constructed class only sees the __init__ method and the __slots__ member, which is working as desired:

>>> MS1 = Struct1('forename','lastname')
>>> m=MS1()
>>> m.forename='Jack'
>>> m.forename
'Jack'

回答5:

An update of ThomasH's variant:

def Struct(*args, **kwargs):
    def init(self, *iargs, **ikwargs):
        for k,v in kwargs.items():
            setattr(self, k, v)
        for i in range(len(iargs)):
            setattr(self, args[i], iargs[i])
        for k,v in ikwargs.items():
            setattr(self, k, v)

    name = kwargs.pop("name", "MyStruct")
    kwargs.update(dict((k, None) for k in args))
    return type(name, (object,), {'__init__': init, '__slots__': kwargs.keys()})

This allows parameters (and named parameters) passed into __init__() (without any validation - seems crude):

>>> Person = Struct('fname', 'age')
>>> person1 = Person('Kevin', 25)
>>> person2 = Person(age=42, fname='Terry')
>>> person1.age += 10
>>> person2.age -= 10
>>> person1.fname, person1.age, person2.fname, person2.age
('Kevin', 35, 'Terry', 32)
>>> 

Update

Having a look into how namedtuple() does this in collections.py. The class is created and expanded as a string and evaluated. Also has support for pickling and so on, etc.

回答6:

There is namedtuple

>>> from collections import namedtuple
>>> Person = namedtuple("Person", ("forename", "surname"))
>>> john = Person("John", "Doe")
>>> john.forename 
'John'
>>> john.surname 
'Doe'