__pre_init__ functionalty in python?

Question

I would like to make string comparison case insensitive. For that, I would like to create an immutable class with just one string field. In the constructor, I would like to call lower() before assigning the value to the field.

I would like to use as much as possible of standard classes like namedtuple or dataclass. Using the post_init function (see e.g. How to use the __post_init__ method in Dataclasses in Python) feels like a hack. It also makes me wonder is the field still frozen after I changed it in the post_init function.

However, I can't find a pre_init function. Is there a better way?

" I would like to call lower() before assigning the value to the field" -- So, is something stopping you from doing that? If so, pls elaborate. If not, what else is the problem? — fountainhead, Nov 04 '20 at 14:07
Why wouldn't a .lower() in your regular init not be sufficient? Just a hobbyist here, so perhaps i lack some SE knowledge... — Kraay89, Nov 04 '20 at 14:08
A `pre_init` - a mthod that is called before `__init__` is `__new__`. Look here for more https://spyhce.com/blog/understanding-new-and-init — balderman, Nov 04 '20 at 14:11
__new__ is the initialization of the class, __init__ the initialization of the instance. The conversion to lower should happen for all instances. — Pierre van de Laar, Nov 04 '20 at 14:35
For example, overwriting __init__ of a NamedTuple as ```` class Name(NamedTuple): name: str def __init__(self, name: str) -> None: def canonical_representation() -> str: return name.lower() self.name: str = canonical_representation() ``` is not allowed: ``` File "C:\Users\laarpjljvd\AppData\Local\Programs\Python\Python39\lib\typing.py", line 1775, in __new__ raise AttributeError("Cannot overwrite NamedTuple attribute " + key) ``` — Pierre van de Laar, Nov 04 '20 at 14:39
I am looking for readable and minimal declarations: something like ``` @dataclass(frozen=True) class Name: name: str = field.lower() ``` such that I don't have to write all good functions for repr, hash, equality, etc, etc, etc... — Pierre van de Laar, Nov 04 '20 at 14:51
@PierrevandeLaar, I'm not 100% sure what you're asking, but I took a guess at it with the answer I've left. If that is not what you're after, can you please update your question to clarify? Thanks! — Mack, Jan 29 '21 at 17:53

score 4 · Answer 1 · answered Nov 05 '20 at 10:47

4

Turns out that dataclasses doesn't provide the functionality I was looking for. Attrs however does:

from attr import attrs, attrib


@attrs(frozen=True)
class Name:
    name: str = attrib(converter=str.lower)

answered Nov 05 '20 at 10:47

Pierre van de Laar

131
7

Mack · Answer 2 · 2021-01-29T18:04:52.013

Clarification: If I'm interpreting the question correctly, you specifically want to work with a class annotated with the stdlib dataclasses.dataclass(frozen=True). Specifically, for your case, you might have something like this:

from dataclasses import dataclass


@dataclass(frozen=True)
class HasStringField:
    some_string: str


instance = HasStringField("SOME_STRING")
instance.some_string  # value: "SOME_STRING"

My understanding is that you want to implement something such that the value in instance.some_string above is actually "some_string".

Answer: There is no "built-in" way to do what you want, but there are two options that come to mind: using object.__setattr__ in __post_init__ or else using an alternative decorator that provides you with a way to create a custom constructor that still calls through to the constructor created by dataclass.

Option 1: using object.__setattr__ in __post_init__

@dataclasses.dataclass(frozen=True)
class HasStringField:
    some_string: str

    def __post_init__(self):
        object.__setattr__(self, "some_string", self.some_string.lower())


inst = HasStringField("SOME_STRING")
inst.some_string  # value: "some_string"

Option 2: custom decorator

The dataclass_with_default_init decorator in an answer to a similar question provides one such decorator.

Using the dataclass_with_default_init decorator, you'd do:

# Import / implement the decorator

@dataclass_with_default_init(frozen=True)
class HasStringField:
    some_string: str

    def __init__(self, some_string: str):
        self.__default_init__(some_string=some_string.lower())


instance = HasStringField("SOME_STRING")
instance.some_string  # value: "some_string"

...and everything else about this class would remain as if you had used the dataclass decorator instead.

progmatico · Answer 3 · 2021-02-08T17:33:03.153

This should be one of the simplest ways, the field you mention is the internal data (inherited from the UserString class):

from collections import UserString

class LowerStr(UserString):
    def __init__(self, value):
        super().__init__(value)
        self.data = self.data.lower()

s1 = LowerStr("ABC")
print("s1: {}".format(s1))
print("s1 class: {}".format(s1.__class__))
print("s1 class mro: {}".format(s1.__class__.mro()))

Output:

s1: abc
s1 class: <class '__main__.LowerStr'>
s1 class mro: [<class '__main__.LowerStr'>, <class 'collections.UserString'>, <class 'collections.abc.Sequence'>, <class 'collections.abc.Reversible'>, <class 'collections.abc.Collection'>, <class 'collections.abc.Sized'>, <class 'collections.abc.Iterable'>, <class 'collections.abc.Container'>, <class 'object'>]

This gives you all the methods from str and you can customize them at will.

If you prefer to subclass, so there is no internal data attribute:

class InsensitiveStr(str):
    def __new__(cls, value):
        if isinstance(value, str):
            return super().__new__(cls, value.lower())
        else:
            raise ValueError

s1 = InsensitiveStr("ABC")
print(s1)
print(type(s1))
print(s1.__class__)
print(s1.__class__.__mro__)

Note the differences in mro.

Outputs

abc
<class '__main__.InsensitiveStr'>
<class '__main__.InsensitiveStr'>
(<class '__main__.InsensitiveStr'>, <class 'str'>, <class 'object'>)

This class is not immutable. One can change the value of the data field. — Pierre van de Laar, Feb 01 '21 at 07:40
`s1` is like an immutable string. Apart from `s1.data` of course. You can as well sub-class with the same idea. See my edit. — progmatico, Feb 08 '21 at 17:29
See also [here](https://stackoverflow.com/questions/7255655/how-to-subclass-str-in-python#7255782) for similar discussion. — progmatico, Feb 08 '21 at 17:34

score 0 · Answer 4 · answered Oct 24 '21 at 14:38

Note: I still think the best approach is to use the __post_init__ / object.__set_attribute__ mentioned above or to use an external library like https://www.attrs.org/en/stable/)

One pattern I have seen reading the python code for dataclasses itself is to use a function with the same name but in snake case to do these sorts of changes. e.g.:

@dataclass(frozen=True)
class HasStringField:
  some_string: str


def has_string_field(some_string: str) -> HasStringField:
  return HasStringField(some_string.lower())

Then you always import and use has_string_field() rather than HasStringField(). (An example of this pattern is dataclasses field vs Field):

It has the drawback of being verbose and making the class invariant's bypassable.

__pre_init__ functionalty in python?

4 Answers4

Linked