attrs/docs/init.md

Initialization

In Python, instance initialization happens in the __init__ method. Generally speaking, you should keep as little logic as possible in it, and you should think about what the class needs and not how it is going to be instantiated.

Passing complex objects into __init__ and then using them to derive data for the class unnecessarily couples your new class with the old class which makes it harder to test and also will cause problems later.

So assuming you use an ORM and want to extract 2D points from a row object, do not write code like this:

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

pt = Point(row)

Instead, write a {obj}classmethod that will extract it for you:

@define
class Point:
    x: float
    y: float

    @classmethod
    def from_row(cls, row):
        return cls(row.x, row.y)

pt = Point.from_row(row)

This is sometimes called a named constructor or a factory method.

Now, you can instantiate Points without creating fake row objects in your tests. You can also have as many smart creation helpers as you want. This flexibility is useful because additional data sources tend to appear over time.

---

For similar reasons, we strongly discourage from patterns like:

pt = Point(**row.attributes)

which couples your classes to the database data model. Try to design your classes in a way that is clean and convenient to use -- not based on your database format. The database format can change anytime and you're stuck with a bad class design that is hard to change. Embrace functions and classmethods as a filter between reality and what's best for you to work with.

:::{important} While attrs's initialization concepts (including the following sections about validators and converters) are powerful, they are not intended to replace a fully-featured serialization or validation system.

We want to help you to write a __init__ that you'd write by hand, but with less boilerplate.

If you look for powerful-yet-unintrusive serialization and validation for your attrs classes, have a look at our sibling project cattrs or our third-party extensions.

This separation of creating classes and serializing them is a conscious design decision. We don't think that your business model and your serialization format should be coupled. :::

(private-attributes)=

Private Attributes and Aliases

One thing people tend to find confusing is the treatment of private attributes that start with an underscore. attrs follows the doctrine that there is no such thing as a private argument and strips the underscores from the name when writing the __init__ method signature:

>>> import inspect
>>> from attrs import define
>>> @define
... class C:
...    _x: int
>>> inspect.signature(C.__init__)
<Signature (self, x: int) -> None>

There really isn't a right or wrong, it's a matter of taste. But it's important to be aware of it because it can lead to surprising syntax errors:

>>> @define
... class C:
...    _1: int
Traceback (most recent call last):
   ...
SyntaxError: invalid syntax

In this case a valid attribute name _1 got transformed into an invalid argument name 1.

If your taste differs, you can use the alias argument to {func}attrs.field to explicitly set the argument name. This can be used to override private attribute handling, or make other arbitrary changes to __init__ argument names.

>>> from attrs import field
>>> @define
... class C:
...    _x: int = field(alias="_x")
...    y: int = field(alias="distasteful_y")
>>> inspect.signature(C.__init__)
<Signature (self, _x: int, distasteful_y: int) -> None>

(defaults)=

Defaults

Sometimes you don't want to pass all attribute values to a class. And sometimes, certain attributes aren't even intended to be passed but you want to allow for customization anyways for easier testing.

This is when default values come into play:

>>> from attrs import Factory
>>> @define
... class C:
...     a: int = 42
...     b: list = field(factory=list)
...     c: list = Factory(list)  # syntactic sugar for above
...     d: dict = field()
...     @d.default
...     def _any_name_except_a_name_of_an_attribute(self):
...        return {}
>>> C()
C(a=42, b=[], c=[], d={})

It's important that the decorated method -- or any other method or property! -- doesn't have the same name as the attribute, otherwise it would overwrite the attribute definition.

Similar to the common gotcha with mutable default arguments, default=[] will not do what you may think it might do:

>>> @define
... class C:
...     x = []
>>> i = C()
>>> k = C()
>>> i.x.append(42)
>>> k.x
[42]

This is why attrs comes with factory options.

:::{warning} Please note that the decorator based defaults have one gotcha: they are executed when the attribute is set, that means depending on the order of attributes, the self object may not be fully initialized when they're called.

Therefore you should use self as little as possible.

Even the smartest of us can get confused by what happens if you pass partially initialized objects around. :::

(validators)=

Validators

Another thing that definitely does belong in __init__ is checking the resulting instance for invariants. This is why attrs has the concept of validators.

Decorator

The most straightforward way is using the attribute's validator method as a decorator.

The method has to accept three arguments:

1. the instance that's being validated (aka self),
2. the attribute that it's validating, and finally
3. the value that is passed for it.

These values are passed as positional arguments, therefore their names don't matter.

If the value does not pass the validator's standards, it just raises an appropriate exception.

>>> @define
... class C:
...     x: int = field()
...     @x.validator
...     def _check_x(self, attribute, value):
...         if value > 42:
...             raise ValueError("x must be smaller or equal to 42")
>>> C(42)
C(x=42)
>>> C(43)
Traceback (most recent call last):
   ...
ValueError: x must be smaller or equal to 42

Again, it's important that the decorated method doesn't have the same name as the attribute and that the {func}attrs.field helper is used.

Callables

If you want to reuse your validators, you should have a look at the validator argument to {func}attrs.field.

It takes either a callable or a list of callables (usually functions) and treats them as validators that receive the same arguments as with the decorator approach. Also as with the decorator approach, they are passed as positional arguments so you can name them however you want.

Since the validators run after the instance is initialized, you can refer to other attributes while validating:

>>> import attrs
>>> def x_smaller_than_y(instance, attribute, value):
...     if value >= instance.y:
...         raise ValueError("'x' has to be smaller than 'y'!")
>>> @define
... class C:
...     x = field(validator=[attrs.validators.instance_of(int),
...                          x_smaller_than_y])
...     y = field()
>>> C(x=3, y=4)
C(x=3, y=4)
>>> C(x=4, y=3)
Traceback (most recent call last):
   ...
ValueError: 'x' has to be smaller than 'y'!

This example demonstrates a convenience shortcut: Passing a list of validators directly is equivalent to passing them wrapped in the {obj}attrs.validators.and_ validator and all validators must pass.

attrs won't intercept your changes to those attributes but you can always call {func}attrs.validate on any instance to verify that it's still valid:

When using {func}attrs.define or attrs.frozen, however, attrs will run the validators even when setting the attribute.

>>> i = C(4, 5)
>>> i.x = 5
Traceback (most recent call last):
   ...
ValueError: 'x' has to be smaller than 'y'!

attrs ships with a bunch of validators, make sure to check them out before writing your own:

>>> @define
... class C:
...     x = field(validator=attrs.validators.instance_of(int))
>>> C(42)
C(x=42)
>>> C("42")
Traceback (most recent call last):
   ...
TypeError: ("'x' must be <type 'int'> (got '42' that is a <type 'str'>).", Attribute(name='x', default=NOTHING, factory=NOTHING, validator=<instance_of validator for type <type 'int'>>, type=None), <type 'int'>, '42')

Of course you can mix and match the two approaches at your convenience. If you use both ways to define validators for an attribute, they are both ran:

>>> @define
... class C:
...     x = field(validator=attrs.validators.instance_of(int))
...     @x.validator
...     def fits_byte(self, attribute, value):
...         if not 0 <= value < 256:
...             raise ValueError("value out of bounds")
>>> C(128)
C(x=128)
>>> C("128")
Traceback (most recent call last):
   ...
TypeError: ("'x' must be <class 'int'> (got '128' that is a <class 'str'>).", Attribute(name='x', default=NOTHING, validator=[<instance_of validator for type <class 'int'>>, <function fits_byte at 0x10fd7a0d0>], repr=True, cmp=True, hash=True, init=True, metadata=mappingproxy({}), type=None, converter=None), <class 'int'>, '128')
>>> C(256)
Traceback (most recent call last):
   ...
ValueError: value out of bounds

Finally, validators can be globally disabled:

>>> attrs.validators.set_disabled(True)
>>> C("128")
C(x='128')
>>> attrs.validators.set_disabled(False)
>>> C("128")
Traceback (most recent call last):
   ...
TypeError: ("'x' must be <class 'int'> (got '128' that is a <class 'str'>).", Attribute(name='x', default=NOTHING, validator=[<instance_of validator for type <class 'int'>>, <function fits_byte at 0x10fd7a0d0>], repr=True, cmp=True, hash=True, init=True, metadata=mappingproxy({}), type=None, converter=None), <class 'int'>, '128')

... or within a context manager:

>>> with attrs.validators.disabled():
...     C("128")
C(x='128')
>>> C("128")
Traceback (most recent call last):
   ...
TypeError: ("'x' must be <class 'int'> (got '128' that is a <class 'str'>).", Attribute(name='x', default=NOTHING, validator=[<instance_of validator for type <class 'int'>>, <function fits_byte at 0x10fd7a0d0>], repr=True, cmp=True, hash=True, init=True, metadata=mappingproxy({}), type=None, converter=None), <class 'int'>, '128')

(converters)=

Converters

Sometimes, it is necessary to normalize the values coming in, therefore attrs comes with converters.

Attributes can have a converter function specified, which will be called with the attribute's passed-in value to get a new value to use. This can be useful for doing type-conversions on values that you don't want to force your callers to do.

>>> @define
... class C:
...     x = field(converter=int)
>>> o = C("1")
>>> o.x
1
>>> o.x = "2"
>>> o.x
2

Converters are run before validators, so you can use validators to check the final form of the value.

>>> def validate_x(instance, attribute, value):
...     if value < 0:
...         raise ValueError("x must be at least 0.")
>>> @define
... class C:
...     x = field(converter=int, validator=validate_x)
>>> o = C("0")
>>> o.x
0
>>> C("-1")
Traceback (most recent call last):
   ...
ValueError: x must be at least 0.

Arguably, you can abuse converters as one-argument validators:

>>> C("x")
Traceback (most recent call last):
   ...
ValueError: invalid literal for int() with base 10: 'x'

If a converter's first argument has a type annotation, that type will appear in the signature for __init__. A converter will override an explicit type annotation or type argument.

>>> def str2int(x: str) -> int:
...     return int(x)
>>> @define
... class C:
...     x = field(converter=str2int)
>>> C.__init__.__annotations__
{'return': None, 'x': <class 'str'>}

If you need more control over the conversion process, you can wrap the converter with a {class}attrs.Converter and ask for the instance and/or the field that are being initialized:

>>> def complicated(value, self_, field):
...     return int(value) * self_.factor + field.metadata["offset"]
>>> @define
... class C:
...     factor = 5  # not an *attrs* field
...     x = field(
...         metadata={"offset": 200},
...         converter=attrs.Converter(
...             complicated,
...             takes_self=True, takes_field=True
...     ))
>>> C("42")
C(x=410)

Hooking Yourself Into Initialization

Generally speaking, the moment you realize the need of finer control – than what attrs offers – over how a class is instantiated, it's usually best to use a {obj}classmethod factory or to apply the builder pattern.

However, sometimes you need to do that one quick thing before or after your class is initialized. For that purpose, attrs offers the following options:

• __attrs_pre_init__ is automatically detected and run before attrs starts initializing. If __attrs_pre_init__ takes more than the self argument, the attrs-generated __init__ will call it with the same arguments it received itself. This is useful if you need to inject a call to super().__init__() -- with or without arguments.

• __attrs_post_init__ is automatically detected and run after attrs is done initializing your instance. This is useful if you want to derive some attribute from others or perform some kind of validation over the whole instance.

• __attrs_init__ is written and attached to your class instead of __init__, if attrs is told to not write one (when init=False or a by a combination of auto_detect=True and a custom __init__). This is useful if you want full control over the initialization process, but don't want to set the attributes by hand.

Pre Init

The sole reason for the existence of __attrs_pre_init__ is to give users the chance to call super().__init__(), because some subclassing-based APIs require that.

>>> @define
... class C:
...     x: int
...     def __attrs_pre_init__(self):
...         super().__init__()
>>> C(42)
C(x=42)

If you need more control, use the custom init approach described next.

:::{warning} You never need to use super() with attrs classes that inherit from other attrs classes. Each attrs class implements an __init__ based on its own fields and those of all its base classes.

You only need this escape hatch when subclassing non-attrs classes. :::

Custom Init

If you tell attrs to not write an __init__, it will write an __attrs_init__ instead, with the same code that it would have used for __init__. You have full control over the initialization, but also have to type out the types of your arguments etc. Here's an example of a manual default value:

>>> @define
... class C:
...     x: int
...
...     def __init__(self, x: int = 42):
...         self.__attrs_init__(x)
>>> C()
C(x=42)

Post Init

>>> @define
... class C:
...     x: int
...     y: int = field(init=False)
...     def __attrs_post_init__(self):
...         self.y = self.x + 1
>>> C(1)
C(x=1, y=2)

Please note that you can't directly set attributes on frozen classes:

>>> @frozen
... class FrozenBroken:
...     x: int
...     y: int = field(init=False)
...     def __attrs_post_init__(self):
...         self.y = self.x + 1
>>> FrozenBroken(1)
Traceback (most recent call last):
   ...
attrs.exceptions.FrozenInstanceError: can't set attribute

If you need to set attributes on a frozen class, you'll have to resort to the same trick as attrs and use {meth}object.__setattr__:

>>> @define
... class Frozen:
...     x: int
...     y: int = field(init=False)
...     def __attrs_post_init__(self):
...         object.__setattr__(self, "y", self.x + 1)
>>> Frozen(1)
Frozen(x=1, y=2)

Note that you must not access the hash code of the object in __attrs_post_init__ if cache_hash=True.

Order of Execution

If present, the hooks are executed in the following order:

1. __attrs_pre_init__ (if present on current class)

2. For each attribute, in the order it was declared:

   1. default factory
   2. converter

3. all validators

4. __attrs_post_init__ (if present on current class)

Notably this means, that you can access all attributes from within your validators, but your converters have to deal with invalid values and have to return a valid value.

Derived Attributes

One of the most common attrs questions on Stack Overflow is how to have attributes that depend on other attributes. For example if you have an API token and want to instantiate a web client that uses it for authentication. Based on the previous sections, there are two approaches.

The simpler one is using __attrs_post_init__:

@define
class APIClient:
    token: str
    client: WebClient = field(init=False)

    def __attrs_post_init__(self):
        self.client = WebClient(self.token)

The second one is using a decorator-based default:

@define
class APIClient:
    token: str
    client: WebClient = field()  # needed! attr.ib works too

    @client.default
    def _client_factory(self):
        return WebClient(self.token)

That said, and as pointed out in the beginning of the chapter, a better approach would be to have a factory class method:

@define
class APIClient:
    client: WebClient

    @classmethod
    def from_token(cls, token: str) -> "APIClient":
        return cls(client=WebClient(token))

This makes the class more testable.

(init-subclass)=

attrs and __init_subclass__

{meth}object.__init_subclass__ is a special method that is called when a subclass of the class that defined it is created.

For example:

>>> class Base:
...    @classmethod
...    def __init_subclass__(cls):
...        print(f"Base has been subclassed by {cls}.")
>>> class Derived(Base):
...    pass
Base has been subclassed by <class 'Derived'>.

Unfortunately, a class decorator-based approach like attrs (or dataclasses) doesn't play well with __init_subclass__. With {term}dict classes, it is run before the class has been processed by attrs and in the case of {term}slotted classes, where attrs has to replace the original class, __init_subclass__ is called twice: once for the original class and once for the attrs class.

To alleviate this, attrs provides __attrs_init_subclass__ which is also called once the class is done assembling. The base class doesn't even have to be an attrs class:

>>> class Base:
...    @classmethod
...    def __attrs_init_subclass__(cls):
...        print(f"Base has been subclassed by attrs {cls}.")
>>> @define
... class Derived(Base):
...    pass
Base has been subclassed by attrs <class 'Derived'>.