跳转至

尚未翻译

本页面尚未翻译成中文，因此显示英文原文。 帮助翻译

delattr() Function Complexity

The delattr() function removes a named attribute from an object. delattr(obj, "name") is exactly del obj.name: both call type(obj).__delattr__, so a property deleter, a descriptor's __delete__ or a class's own __delattr__ runs in the same way.

d is the classes in type(obj).__mro__, s is the subclasses of a class, direct and indirect, and k is the cost of a user-defined __delattr__ or __delete__. Attribute names are treated as hashing and comparing in O(1). The bounds assume cached type lookups; a type attribute cache miss adds O(d).

Complexity Reference

Operation Time Space Notes
delattr(obj, name), instance attribute O(1) avg O(1) Deletes from the instance's __dict__ once no data descriptor on the class claims the name
delattr(obj, name), __slots__ attribute O(1) O(1) Clears the slot
delattr(obj, name), property or descriptor O(k) O(k) Runs the deleter or __delete__, not the dict deletion
delattr(obj, name), custom __delattr__ O(k) O(k) Runs the class's __delattr__
delattr(cls, name), class attribute O(s) O(1) Invalidates the attribute cache of the class and of each subclass whose cache is still valid
delattr(module, name) O(1) avg O(1) Deletes from the module's __dict__
Missing attribute O(1) avg O(1) Raises AttributeError

Instance Attributes

Deleting an attribute that lives in the instance's __dict__ is one hash-table deletion. A class attribute is not on the instance, so deleting it through the instance raises:

class Config:
    default = 42

config = Config()
config.host = "localhost"
config.port = 8000

delattr(config, "host")  # O(1) avg
del config.port  # O(1) avg - the same operation
assert vars(config) == {}

try:
    delattr(config, "default")  # lives on the class, not the instance
except AttributeError as e:
    assert "default" in str(e)
else:
    raise AssertionError("expected AttributeError")
assert config.default == 42

__slots__

A slot is cleared in place. Deleting a slot that holds no value raises AttributeError:

class Point:
    __slots__ = ("x", "y")

    def __init__(self, x, y):
        self.x = x
        self.y = y

p = Point(1, 2)
delattr(p, "x")  # O(1)
assert not hasattr(p, "x")

try:
    delattr(p, "x")
except AttributeError:
    pass
else:
    raise AssertionError("expected AttributeError")

Properties, Descriptors and __delattr__

A data descriptor on the class takes priority over the instance's __dict__, so delattr() costs whatever its deleter costs. A property without a deleter raises AttributeError:

class Cached:
    def __init__(self):
        self._value = 1
        self.log = []

    @property
    def value(self):
        return self._value

    @value.deleter
    def value(self):
        self.log.append("deleted")
        del self._value

obj = Cached()
delattr(obj, "value")  # O(k) - runs the deleter
assert obj.log == ["deleted"]
assert not hasattr(obj, "_value")


class ReadOnly:
    @property
    def value(self):
        return 1

try:
    delattr(ReadOnly(), "value")
except AttributeError:
    pass
else:
    raise AssertionError("expected AttributeError")

A class that defines __delattr__ sees every deletion, whether written as del or delattr():

class Tracked:
    def __init__(self):
        object.__setattr__(self, "deleted", [])

    def __delattr__(self, name):
        self.deleted.append(name)
        super().__delattr__(name)

obj = Tracked()
obj.x = 1
obj.y = 2
delattr(obj, "x")  # O(k) - calls Tracked.__delattr__
del obj.y
assert obj.deleted == ["x", "y"]

Deleting Class Attributes

Python caches attribute lookups per class. Deleting (or setting) an attribute on a class invalidates that cache for the class and its subclasses, so the cost grows with the size of the class hierarchy below it. Change class attributes at setup time, not in a hot loop:

class Base:
    flag = True

class Child(Base):
    pass

assert Child().flag
delattr(Base, "flag")  # O(s) - s = subclasses of Base
assert not hasattr(Child(), "flag")

Deleting Many Attributes

For ordinary instance attributes each delattr() call is O(1), so deleting n of them is O(n). To drop everything in the instance __dict__, vars(obj).clear() does it in one call without naming the attributes; it bypasses any __delattr__ the class defines and leaves __slots__ values alone:

class State:
    pass

state = State()
for i in range(100):
    setattr(state, f"attr{i}", i)

for i in range(50):
    delattr(state, f"attr{i}")  # O(1) each, O(n) in total
assert len(vars(state)) == 50

vars(state).clear()  # O(n)
assert vars(state) == {}

Common Patterns

Invalidating a Cached Value

Deleting a cached attribute makes the next access recompute it. functools.cached_property stores its value in the instance __dict__, so delattr() is the way to reset it:

from functools import cached_property

class Report:
    def __init__(self, rows):
        self.rows = rows
        self.computed = 0

    @cached_property
    def total(self):
        self.computed += 1
        return sum(self.rows)

report = Report([1, 2, 3])
assert report.total == 6
report.rows.append(4)
assert report.total == 6  # cached

delattr(report, "total")  # O(1) avg - drops the cached value
assert report.total == 10
assert report.computed == 2

Removing Optional Attributes

delattr() raises on a missing name. Catching the exception keeps it to one call per name; checking with hasattr() first adds a full attribute read to each one:

class Session:
    def __init__(self):
        self.user = "alice"
        self.token = "abc"

def drop(obj, names):
    for name in names:  # O(n)
        try:
            delattr(obj, name)  # O(1) avg
        except AttributeError:
            pass

session = Session()
drop(session, ["token", "password"])
assert vars(session) == {"user": "alice"}

Performance Best Practices

✅ Do:

  • Use delattr() when the attribute name is only known at run time; write del obj.name otherwise
  • Use vars(obj).clear() to empty the instance __dict__ at once
  • Delete a cached_property value to force it to be recomputed

❌ Avoid:

  • Deleting or setting class attributes in a hot loop: each change is O(s) in the subclasses
  • Expecting delattr() on an instance to remove a class attribute
  • Assuming delattr() is O(1) when the class defines __delattr__ or the name is a property

Version Notes

  • All Python 3: delattr(obj, name) is equivalent to del obj.name; name must be a string