What does frozen mean for dataclasses?
Question:
What’s the difference between @dataclass(frozen=True)
and @dataclass(frozen=False)
? When should I use which?
Answers:
In Python, "frozen" means an object cannot be modified. For example, consider set
and frozenset
:
>>> s = set((1, 2, 3))
>>> s
{1, 2, 3}
>>> s.add(4)
>>> s
{1, 2, 3, 4}
>>> fs = frozenset((1, 2, 3))
>>> fs
frozenset({1, 2, 3})
>>> fs.add(4)
...
AttributeError: 'frozenset' object has no attribute 'add'
Likewise, creating a dataclass
with frozen=True
means its instances are frozen and cannot be changed.
Be aware that frozen
only applies to the dataclass instance itself – a frozen
dataclass can contain mutable items such as lists, and a regular dataclass can contain frozen/immutable items such as tuples.
The point of frozen objects is to avoid accidental modification, and to guarantee a consistent value.
- The former is advantageous to avoid bugs. When an object is not intended to be modified, making it
frozen
reveals accidental modification via an immediate error.
- The latter allows use as immutable object, for example the keys of a
dict
. A frozen
dataclass is by default hashable and suitable as a dict
key.
from dataclasses import dataclass
@dataclass(frozen=True)
class Frozen:
x: int
y: int
named_points = {Frozen(0, 0): "Origin"}
Note that hashability does not just depend on the dataclass but is recursive – a frozen
dataclass containing a list
is not hashable, because the list
is not hashable.
What’s the difference between @dataclass(frozen=True)
and @dataclass(frozen=False)
? When should I use which?
In Python, "frozen" means an object cannot be modified. For example, consider set
and frozenset
:
>>> s = set((1, 2, 3))
>>> s
{1, 2, 3}
>>> s.add(4)
>>> s
{1, 2, 3, 4}
>>> fs = frozenset((1, 2, 3))
>>> fs
frozenset({1, 2, 3})
>>> fs.add(4)
...
AttributeError: 'frozenset' object has no attribute 'add'
Likewise, creating a dataclass
with frozen=True
means its instances are frozen and cannot be changed.
Be aware that frozen
only applies to the dataclass instance itself – a frozen
dataclass can contain mutable items such as lists, and a regular dataclass can contain frozen/immutable items such as tuples.
The point of frozen objects is to avoid accidental modification, and to guarantee a consistent value.
- The former is advantageous to avoid bugs. When an object is not intended to be modified, making it
frozen
reveals accidental modification via an immediate error. - The latter allows use as immutable object, for example the keys of a
dict
. Afrozen
dataclass is by default hashable and suitable as adict
key.
from dataclasses import dataclass
@dataclass(frozen=True)
class Frozen:
x: int
y: int
named_points = {Frozen(0, 0): "Origin"}
Note that hashability does not just depend on the dataclass but is recursive – a frozen
dataclass containing a list
is not hashable, because the list
is not hashable.