Copy constructor in python?

Question

Is there a copy constructor in python ? If not what would I do to achieve something similar ?

The situation is that I am using a library and I have extended one of t

Answer 1

The following solution probably repeats some of the previous ones in a simple form. I don't know how it is "pythocally" right, but it works and was quite convenient in the certain case I used it.

class Entity:
    def __init__(self, code=None, name=None, attrs=None):
        self.code = code
        self.name = name
        self.attrs = {} if attrs is None else attrs


    def copy(self, attrs=None):
        new_attrs = {k: v.copy() for k, v in self.attrs.items()} if attrs is None else attrs
        return Entity(code=self.code, name=self.name, attrs=new_attrs)

Usage:

new_entity = entity.copy()

This is a more complicated version that allows to interfere in the copying process. I used it in only one place. Also note that objects contained in self.attrs also have such kind of "copying constructor".

This solution is not generic but is very simple and provides quite much control.

Answer 2

A simple example of my usual implementation of a copy constructor:

import copy

class Foo:

  def __init__(self, data):
    self._data = data

  @classmethod
  def from_foo(cls, class_instance):
    data = copy.deepcopy(class_instance._data) # if deepcopy is necessary
    return cls(data)

Answer 3

I have a similar situation differing in that the new class only needs to copy attributes. Thus using @Dunham's idea and adding some specificity to @meisterluk's suggestion, @meisterluk's "copy_constructor" method could be:

from copy import deepcopy
class Foo(object):
    def __init__(self, myOne=1, other=None):
    self.two = 2
    if other <> None:
        assert isinstance(other, Foo), "can only copy instances of Foo"
        self.__dict__ = deepcopy(other.__dict__)
    self.one = myOne

def __repr__(self):
    out = ''
    for k,v in self.__dict__.items():
        out += '{:>4s}: {}, {}\n'.format(k,v.__class__,v)
    return out

def bar(self):
    pass

foo1 = Foo()
foo2 = Foo('one', foo1)

print '\nfoo1\n',foo1
print '\nfoo2\n',foo2

The output:

foo1
 two: <type 'int'>, 2
 one: <type 'int'>, 1


foo2
 two: <type 'int'>, 2
 one: <type 'str'>, one

Answer 4

In python the copy constructor can be defined using default arguments. Lets say you want the normal constructor to run the function non_copy_constructor(self) and the copy constructor should run copy_constructor(self, orig). Then you can do the following:

class Foo:
    def __init__(self, orig=None):
        if orig is None:
            self.non_copy_constructor()
        else:
            self.copy_constructor(orig)
    def non_copy_constructor(self):
        # do the non-copy constructor stuff
    def copy_constructor(self, orig):
        # do the copy constructor

a=Foo()  # this will call the non-copy constructor
b=Foo(a) # this will call the copy constructor

Answer 5

For your situation, I would suggest writing a class method (or it could be a static method or a separate function) that takes as an argument an instance of the library's class and returns an instance of your class with all applicable attributes copied over.

Answer 6

Building on @Godsmith's train of thought and addressing @Zitrax's need (I think) to do the data copy for all attributes within the constructor:

class ConfusionMatrix(pd.DataFrame):
    def __init__(self, df, *args, **kwargs):
        try:
            # Check if `df` looks like a `ConfusionMatrix`
            # Could check `isinstance(df, ConfusionMatrix)`
            # But might miss some "ConfusionMatrix-elligible" `DataFrame`s
            assert((df.columns == df.index).all())
            assert(df.values.dtype == int)
            self.construct_copy(df, *args, **kwargs)
            return
        except (AssertionError, AttributeError, ValueError):
            pass
        # df is just data, so continue with normal constructor here ...

    def construct_copy(self, other, *args, **kwargs):
        # construct a parent DataFrame instance
        parent_type = super(ConfusionMatrix, self)
        parent_type.__init__(other)
        for k, v in other.__dict__.iteritems():
            if hasattr(parent_type, k) and hasattr(self, k) and getattr(parent_type, k) == getattr(self, k):
                continue
            setattr(self, k, deepcopy(v))

This ConfusionMatrix class inherits a pandas.DataFrame and adds a ton of other attributes and methods that need to be recomputed unless the other matrix data can be copied over. Searching for a solution is how I found this question.