immutable-defaults 0.1.1

Immutable defaults for Python

immutable_defaults

Simple decorator to force immutability to function arguments by deepcopying. Never again pass None when your heart wants to pass an empty list. Also works for arbitrary objects that can be deepcopied. Has simple config options for granularity or performance (copy vs deepcopy).

No dependencies.

In order to use various type hints we require Python >=3.12.

How to install

pip install immutable-defaults or your equivalent (e.g. pdm add immutable-defaults)

Example usage

from immutable_defaults import immutable_defaults 

@immutable_defaults
def my_function(a: list = []):
    a.append("world")
    return a

print(my_function())  # ['world']
print(my_function(a=["hello"]))  # ['hello', 'world']
print(my_function(["HELLO"]))  # ['HELLO', 'world']
print(my_function())  #  ['world']

@immutable_defaults(ignore=["b"])
def my_function2(a = ["hello"], b = []):
    """basic function with ignore parameter"""
    a.append("world")
    b.append("!")
    return a + b

print(my_function2())  # ['hello', 'world', '!']
print(my_function2())  # ['hello', 'world', '!', '!']
print(my_function2())  # ['hello', 'world', '!', '!', '!']

# more exhaustive tests in tests/tests.py

Methods, Classmethods, and Staticmethods

The decorator works with methods, classmethods and staticmethods. Since @immutable_defaults requires that the wrapped function is callable, make sure that the outer decorator is @classmethod/@staticmethod.

Optional keyword arguments

• @immutable_defaults can be called with keyword arguments deepcopy and ignore.

• deepcopy: boolean | Iterable[str] = True

  • if True then defaults are copied with copy.deepcopy. If False, then with copy.copy.
  • If passed an iterable of argument names then those arguments will be deep copied and other mutable defaults will be shallow copied, e.g. in the below a and arg will be deep copied while b will be shallow copied.

    @immutable_defaults(deepcopy=["a","arg"]) 
    def f(a=[[1]], b=[], arg={1: {2}}): ...
  

• ignore: Iterable[str] | None = None

  • all argument names passed will have the default Python behavior.

Input validation

• We check that you cannot have the same mutable object (as per a is b comparison) marked for both shallow and deep copying. For example, the below will raise an ImmutableDefaultsError:

xss = [[1]]
@immutable_defaults(deepcopy=["xss2"]) # raises ImmutableDefaultsError
def f(x, xss1 = xss, xss2 = xss): ...

• Similarly, we check that you cannot ignore and not ignore the same mutable object. For example, the below will raise an ImmutableDefaultsError:

xss = [[1]]
@immutable_defaults(ignore=["xss2"]) # raises ImmutableDefaultsError
def f(x, xss1 = xss, xss2 = xss): ...

• A KeyError is raised if either deepcopy or ignore have arguments that cannot be found in the signature of the decorated function.
  • It would have been easy to silently do nothing when variables in ignore are not present, but this would make typos very hard to debug.
• ignore takes precedence over deepcopy, i.e. @immutable_defaults(ignore=["x"], deepcopy=["x"]) will do the same thing as @immutable_defaults(ignore=["x"])

Prior art

(Comments valid May 13 2024)

• comparison with https://pypi.org/project/immutable_default_args/
  • we deep copy all defaults except for standard immutable types (int, float, complex, bool, str, tuple, frozenset). This means we cover sets, and also other custom mutable objects that implement __deepcopy__ (or optionally __copy__).
  • Instead of a metaclass, we have a class decorator.
• comparison with https://pypi.org/project/python-immutable/
• comparison with https://github.com/roodrepo/default_mutable/
  • we have typed code and IMO better naming.
• comparison with https://pypi.org/project/python-none-objects/
  • completely different solution to the mutable default arguments problem.
  • only works for empty containers.

Todo

• Performance benchmarking - what is the overhead?
• Make publishing to pypi part of github workflow