Private Constructor in Python

Question

The _ and __ prefixes don't offer a solution to restricting instantiation of an object to a specific 'factory', however Python is a powerful toolbox and the desired behaviour can be achieved in more than one way (as @Jesse W at Z has demonstrated). Here is a possible solution that keeps the class publicly visible (allowing isinstance etc.) but ensures construction is only possible by class-methods:

class OnlyCreatable(object):

    __create_key = object()

    @classmethod
    def create(cls, value):
        return OnlyCreatable(cls.__create_key, value)

    def __init__(self, create_key, value):
        assert(create_key == OnlyCreatable.__create_key), \
            "OnlyCreatable objects must be created using OnlyCreatable.create"
        self.value = value

Constructing an object with the create class-method:

>>> OnlyCreatable.create("I'm a test") 
<__main__.OnlyCreatable object at 0x1023a6f60>

When attempting to construct an object without using the create class-method creation fails due to the assertion:

>>> OnlyCreatable(0, "I'm a test")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "<stdin>", line 11, in __init__
AssertionError: OnlyCreatable objects can only be created using OnlyCreatable.create

If attempting to create an object by mimicking the create class-method creation fails due to compiler mangling of OnlyCreatable.__createKey.

>>> OnlyCreatable(OnlyCreatable.__createKey, "I'm a test")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: type object 'OnlyCreatable' has no attribute '__createKey'

The only way to construct OnlyCreatable outside of a class-method is to know the value of OnlyCreatable.__create_key. Since this class-attribute's value is generated at runtime and it's name is prefixed with __ marking it as inaccessible it is effectively 'impossible' to obtain this value and/or construct the object.

How do I create a private constructor?

In essence, it's impossible both because python does not use constructors the way you may think it does if you come from other OOP languages and because python does not enforce privacy, it just has a specific syntax to suggest that a given method/property should be considered as private. Let me elaborate...

First: the closest to a constructor that you can find in python is the __new__ method but this is very very seldom used (you normally use __init__, which modify the just created object (in fact it already has self as first parameter).

Regardless, python is based on the assumption everybody is a consenting adult, thus private/public is not enforced as some other language do.

As mentioned by some other responder, methods that are meant to be "private" are normally prepended by either one or two underscores: _private or __private. The difference between the two is that the latter will scramble the name of the method, so you will be unable to call it from outside the object instantiation, while the former doesn't.

So for example if your class A defines both _private(self) and __private(self):

>>> a = A()
>>> a._private()   # will work
>>> a.__private()  # will raise an exception

You normally want to use the single underscore, as - especially for unit testing - having double underscores can make things very tricky....

HTH!

As no-one has mentioned this yet -- you can have considerable control over what names are visible in what scopes -- and there are lots of scopes available. Here are two three other ways to limit construction of a class to a factory method:

#Define the class within the factory method
def factory():
  class Foo:
    pass
  return Foo()

OR

#Assign the class as an attribute of the factory method
def factory():
  return factory.Foo()
class Foo:
  pass
factory.Foo = Foo
del Foo

(Note: This still allows the class to be referred to from outside (for isinstance checks, for example), but it makes it pretty obvious that you aren't supposed to instantiate it directly.)

OR

#Assign the class to a local variable of an outer function
class Foo:
  pass
def factory_maker():
  inner_Foo=Foo
  def factory():
    return inner_Foo()
  return factory
factory = factory_maker()
del Foo
del factory_maker

This makes it impossible (at least, without using at least one magic (double underscore) property) to access the Foo class, but still allows multiple functions to use it (by defining them before deleting the global Foo name.

Quoting the Python style guide (PEP 8):

In addition, the following special forms using leading or trailing underscores are recognized (these can generally be combined with any case convention):

• _single_leading_underscore: weak "internal use" indicator. E.g. "from M import *" does not import objects whose name starts with an underscore.

• single_trailing_underscore_: used by convention to avoid conflicts with Python keyword, e.g. Tkinter.Toplevel(master, class_='ClassName')

• __double_leading_underscore: when naming a class attribute, invokes name mangling (inside class FooBar, __boo becomes _FooBar__boo; see below).

• __double_leading_and_trailing_underscore__: "magic" objects or attributes that live in user-controlled namespaces. E.g. __init__, __import__ or __file__. Never invent such names; only use them as documented.

Though strictly private attributes do not exist in Python, you can use a metaclass to prevent the use of the MyClass() syntax to create a MyClass object.

Here is an example adapted from the Trio project:

from typing import Type, Any, TypeVar


T = TypeVar("T")


class NoPublicConstructor(type):
    """Metaclass that ensures a private constructor

    If a class uses this metaclass like this:

        class SomeClass(metaclass=NoPublicConstructor):
            pass

    If you try to instantiate your class (`SomeClass()`),
    a `TypeError` will be thrown.
    """

    def __call__(cls, *args, **kwargs):
        raise TypeError(
            f"{cls.__module__}.{cls.__qualname__} has no public constructor"
        )

    def _create(cls: Type[T], *args: Any, **kwargs: Any) -> T:
        return super().__call__(*args, **kwargs)  # type: ignore

Here is an example of use:

from math import cos, sin


class Point(metaclass=NoPublicConstructor):
     def __init__(self, x, y):
         self.x = x
         self.y = y

     @classmethod
     def from_cartesian(cls, x, y):
         return cls._create(x, y)
     
     @classmethod
     def from_polar(cls, rho, phi):
         return cls._create(rho * cos(phi), rho * sin(phi))

Point(1, 2) # raises a type error
Point.from_cartesian(1, 2) # OK
Point.from_polar(1, 2) # OK

Fist of all, the term "constructor" does not apply to Python, because, although __init__() method plays a role of one, it is just a method which is called when an object has already been created and requires initialization.

Every method of a class in Python is public. Generally programmers mark "private" methods with _ or __ in the name of a method, e.g.:

# inheriting from object is relevant for Python 2.x only
class MyClass(object): 
    # kinda "constructor"
    def __init__(self):
        pass

    # here is a "private" method
    def _some_method(self):
        pass

    # ... and a public one
    def another_method(self):
        pass

If a module-level function is OK instead of a static method, see my other answer

You can achieve something to this effect with an abstract class. Any instance attributes that need defining in the "private constructor" can be abstract properties. Your factory class method then builds its own concrete class by populating these abstract attributes, as well as doing any other initialisation work such as data validation.

from abc import ABC, abstractmethod

class Foo(ABC):
    @property
    @abstractmethod
    def _a(self) -> int:
        pass

    def bar(self) -> int:
        return self._a + 1

    @classmethod
    def from_values(cls, a: int) -> 'Foo':
        class _Foo(cls):
            def __init__(self, __a):
                self.__a = __a

            @property
            def _a(self):
                return self.__a
        
        return _Foo(a)

Foo()  # TypeError: Can't instantiate abstract class ...
Foo.from_values(1).bar()  # 1

If you find you need no abstract attributes on Foo, you won't get the TypeError when calling Foo(). In that case you can either rely on the inheritance from ABC as documentation, or define a dummy attribute for safety.

Optional tweaks

• Need mutable instance attributes? Add setters.
• Don't care about the difference between class and instance attributes? Simplify with

  class _Foo(cls):
      _a = a
  
  return _Foo()