mixin class

[Udo Gleich]

Hi,

I try to implement mixin classes. Thats why I
need to make a new class at runtime.

--tmp.py-------------------------------------

import new

class K1(object):
pass

class K2(object):
pass

mixed = new.classobj("K1_K2", (K1, K1), {})
new_instance = new.instance(mixed, {})

print new_instance

---------------------------------------------

Making a new instance from the new class works
only if K1 is not derived from object. If I use
new style classes I get the following traceback:

Traceback (most recent call last):
File "tmp.py", line 10, in ?
new_instance = new.instance(mixed, {})
TypeError: instance() argument 1 must be class, not type

I use Python 2.2.3 and Win2000.

My question: How do I implement mixin classes
with new style classes?

Thanks,

Udo

 

[Michele Simionato]

Hi,

I try to implement mixin classes. Thats why I
need to make a new class at runtime.

--tmp.py-------------------------------------

import new

class K1(object):
pass

class K2(object):
pass

mixed = new.classobj("K1_K2", (K1, K1), {})
new_instance = new.instance(mixed, {})

print new_instance

---------------------------------------------

Making a new instance from the new class works
only if K1 is not derived from object. If I use
new style classes I get the following traceback:

Traceback (most recent call last):
File "tmp.py", line 10, in ?
new_instance = new.instance(mixed, {})
TypeError: instance() argument 1 must be class, not type

I use Python 2.2.3 and Win2000.

My question: How do I implement mixin classes
with new style classes?

Thanks,

Udo

Why not simply

class K1(object):
pass

class K2(object):
pass

mixed = type("K1_K2", (K1, K1), {})
new_instance = mixed()

print new_instance

?

"type" is described in http://www.python.org/2.2.3/descrintro.html
(in one line very easy to miss


Michele

 

[Udo Gleich]

Hi,

Why not simply

class K1(object):
pass

class K2(object):
pass

mixed = type("K1_K2", (K1, K1), {})
new_instance = mixed()

print new_instance

almost.

If K1 has a constructor that takes arguments you get an
error when you call the constructor of the derived class
without an argument. Why do I want to do that? Usually
one would say that I should know the arguments of the
constructor of the derived class.

What I want to do is take *two objects*, derive from both
their classes, make a new object and combine the state of
the old objects.

I dont know if that is a good idea. I would appreciate comments
on the following solution. Especially the use of the dummy_init
function as an empty constructor looks not quite right to me.

---------------------------------------------------------------------

def dummy_init(self):
pass

class Mixin:

__shared_state = {"classes":{}}

def __init__(self):
self.__dict__ = self.__shared_state

def mix(self, original_instance, mixin_instance):
original_class = original_instance.__class__
mixin_class = mixin_instance.__class__
name = original_class.__name__ + '_' + mixin_class.__name__
mixed = self.classes.get(name,
type(name,
(mixin_class, original_class),
{"__init__": dummy_init}))
new_instance = mixed()

new_instance.__dict__.update(mixin_instance.__dict__)
new_instance.__dict__.update(original_instance.__dict__)

try:
new_instance.late_init_original()
except AttributeError:
pass
try:
new_instance.late_init_mixin()
except AttributeError:
pass
return new_instance

class K1(object):
def __init__(self, a):
self.a = a

class K2(object):
def late_init_mixin(self):
self.b = self.a + 1


mixer = Mixin()

new_instance = mixer.mix(K1(3), K2())

print new_instance
print new_instance.a, new_instance.b

-------------------------------------------------------------------------

 

[Udo Gleich]

mixed = new.classobj("K1_K2", (K1, K1), {})

There is an error in the code. Obviously it should
read

mixed = new.classobj("K1_K2", (K1, K2), {})
--

 

[Michele Simionato]

What I want to do is take *two objects*, derive from both
their classes, make a new object and combine the state of
the old objects.

I dont know if that is a good idea. I would appreciate comments
on the following solution. Especially the use of the dummy_init
function as an empty constructor looks not quite right to me.

You may avoid dummy_init and have the default object.__init__ but
this is not the point. The whole construction seems ugly to me.
Do you have the option of using only class variables ? I mean, no
explicit instance dictionary? Then you could simply create the
mixin from the original classes and not merge by hand the
instance dictionaries. Do you have the option of modifying the
original classes to make the all structure more multiple inheritance
friendly? The idea is that one should derive objects from classes,
not classes from objects. What you are doing will probably work,
but It is quite ugly to me.I cannot say more if I have no idea of what
are your constraints. If you cannot touch the original classes,
remember that you can create modifications of them, as mixin-frendly
as you wish. I also have a question: why don't you override the
__init__ method in M in such a way that it calls K1.__init__ and
K2.__init__ according to the number of arguments? That would be
the first thing would come to my mind, but maybe you have reasons
why you don't want that.
HTH,


Michele

 

[Jeremy Bowers]

My question: How do I implement mixin classes
with new style classes?

Um, if I understand what you're trying to do correctly, it's easy:

--------------
import random

class A(object): pass
class B(object): pass
class C(object): pass
class D(object): pass

mixin = random.choice([C, D])

class A_B_and_C_or_D(A, B, mixin): pass

newName = A_B_and_C_or_D

newInstance = newName()
----------------

(Warning: Didn't actually type this in)

The classes being derived from can be variable; in fact I do this rather
too frequently for my own good, perhaps. (I tend to use it for class I
want to be intimately involved with each other, but also usable
separately, having one of them dynamically derive from either object or
the other class, depending on whether the other class is available.)

So in this example, on any given run, class A_B_and_C_or_D will always
derive from A and B, and may derive from either of C or D.

You can dynamically do this in a loop or something if you're feeling
creative, if you assign the class to new names, as I did here for
"newName". AFAIK there's no way to programmatically create new classes
with generated names without resorting to eval (bad idea), but you can
bind them to new names after creation and that works OK.

----------------

classesILike = {}
for i in range(10):
class tmp(object): pass
classesILIke = tmp

----------------

and of course "tmp" may derive from what it will.

The inheritance list is as dynamic as anything else in Python; feel free
to use it that way. Personally I find this is a "killer feature"... you'd
find it hard to describe in advance when you'd want it, but when
you want it, you want it ***badly***, because the kludge will be a
killer... and there are very, very few languages that can do this cleanly
(and still share Python's other benefits).