Enumerations¶
The enum.Enum
class behaves differently from other Python classes in several
ways that require special-case handling in type checkers. This section discusses
the Enum behaviors that should be supported by type checkers and others which
may be supported optionally. It is recommended that library and type stub
authors avoid using optional behaviors because these may not be supported
by some type checkers.
Enum Definition¶
Enum classes can be defined using a “class syntax” or a “function syntax”. The function syntax offers several ways to specify enum members: names passed as individual arguments, a list or tuple of names, a string of comma-delimited or space-delimited names, a list or tuple of tuples that contain name/value pairs, and a dictionary of name/value items.
Type checkers should support the class syntax, but the function syntax (in its various forms) is optional:
class Color1(Enum): # Supported
RED = 1
GREEN = 2
BLUE = 3
Color2 = Enum('Color2', 'RED', 'GREEN', 'BLUE') # Optional
Color3 = Enum('Color3', ['RED', 'GREEN', 'BLUE']) # Optional
Color4 = Enum('Color4', ('RED', 'GREEN', 'BLUE')) # Optional
Color5 = Enum('Color5', 'RED, GREEN, BLUE') # Optional
Color6 = Enum('Color6', 'RED GREEN BLUE') # Optional
Color7 = Enum('Color7', [('RED', 1), ('GREEN', 2), ('BLUE', 3)]) # Optional
Color8 = Enum('Color8', (('RED', 1), ('GREEN', 2), ('BLUE', 3))) # Optional
Color9 = Enum('Color9', {'RED': 1, 'GREEN': 2, 'BLUE': 3}) # Optional
Enum classes can also be defined using a subclass of enum.Enum
or any class
that uses enum.EnumType
(or a subclass thereof) as a metaclass. Note that
enum.EnumType
was named enum.EnumMeta
prior to Python 3.11. Type
checkers should treat such classes as enums:
class CustomEnum1(Enum):
pass
class Color7(CustomEnum1): # Supported
RED = 1
GREEN = 2
BLUE = 3
class CustomEnumType(EnumType):
pass
class CustomEnum2(metaclass=CustomEnumType):
pass
class Color8(CustomEnum2): # Supported
RED = 1
GREEN = 2
BLUE = 3
Enum Behaviors¶
Enum classes are iterable and indexable, and they can be called with a value to look up the enum member with that value. Type checkers should support these behaviors:
class Color(Enum):
RED = 1
GREEN = 2
BLUE = 3
for color in Color:
reveal_type(color) # Revealed type is 'Color'
reveal_type(Color["RED"]) # Revealed type is 'Literal[Color.RED]' (or 'Color')
reveal_type(Color(3)) # Revealed type is 'Literal[Color.BLUE]' (or 'Color')
Unlike most Python classes, Calling an enum class does not invoke its constructor. Instead, the call performs a value-based lookup of an enum member.
An Enum class with one or more defined members cannot be subclassed. They are implicitly “final”. Type checkers should enforce this:
class EnumWithNoMembers(Enum):
pass
class Shape(EnumWithNoMembers): # OK (because no members are defined)
SQUARE = 1
CIRCLE = 2
class ExtendedShape(Shape): # Type checker error: Shape is implicitly final
TRIANGLE = 3
Defining Members¶
When using the “class syntax”, enum classes can define both members and
other (non-member) attributes. The EnumType
metaclass applies a set
of rules to distinguish between members and non-members. Type checkers
should honor the most common of these rules. The lesser-used rules are
optional. Some of these rules may be impossible to evaluate and enforce
statically in cases where dynamic values are used.
If an attribute is defined in the class body with a type annotation but with no assigned value, a type checker should assume this is a non-member attribute:
class Pet(Enum): genus: str # Non-member attribute species: str # Non-member attribute CAT = 1 # Member attribute DOG = 2 # Member attribute
Within a type stub, members can be defined using the actual runtime values, or a placeholder of
...
can be used:class Pet(Enum): genus: str # Non-member attribute species: str # Non-member attribute CAT = 1 # Member attribute with known value and type DOG = cast(int, ...) # Member attribute with unknown value and known type BIRD = ... # Member attribute with unknown value and type
Members defined within an enum class should not include explicit type annotations. Type checkers should infer a literal type for all members. A type checker should report an error if a type annotation is used for an enum member because this type will be incorrect and misleading to readers of the code:
class Pet(Enum): CAT = 1 # OK DOG: int = 2 # Type checker error
Methods, callables, descriptors (including properties), and nested classes that are defined in the class are not treated as enum members by the
EnumType
metaclass and should likewise not be treated as enum members by a type checker:def identity(x): return x class Pet(Enum): CAT = 1 # Member attribute DOG = 2 # Member attribute converter = lambda x: str(x) # Non-member attribute transform = identity # Non-member attribute @property def species(self) -> str: # Non-member property return "mammal" def speak(self) -> None: # Non-member method print("meow" if self is Pet.CAT else "woof") class Nested: ... # Non-member nested class
An attribute that is assigned the value of another member of the same enum is not a member itself. Instead, it is an alias for the first member:
class TrafficLight(Enum): RED = 1 GREEN = 2 YELLOW = 3 AMBER = YELLOW # Alias for YELLOW reveal_type(TrafficLight.AMBER) # Revealed type is Literal[TrafficLight.YELLOW]
If using Python 3.11 or newer, the
enum.member
andenum.nonmember
classes can be used to unambiguously distinguish members from non-members. Type checkers should support these classes:class Example(Enum): a = member(1) # Member attribute b = nonmember(2) # Non-member attribute @member def c(self) -> None: # Member method pass reveal_type(Example.a) # Revealed type is Literal[Example.a] reveal_type(Example.b) # Revealed type is int or Literal[2] reveal_type(Example.c) # Revealed type is Literal[Example.c]
An attribute with a private name (beginning with, but not ending in, a double underscore) is treated as a non-member:
class Example(Enum): A = 1 # Member attribute __B = 2 # Non-member attribute reveal_type(Example.A) # Revealed type is Literal[Example.A] reveal_type(Example.__B) # Type Error: Private name is mangled
An enum class can define a class symbol named
_ignore_
. This can be a list of names or a string containing a space-delimited list of names that are deleted from the enum class at runtime. Type checkers may support this mechanism:class Pet(Enum): _ignore_ = "DOG FISH" CAT = 1 # Member attribute DOG = 2 # temporary variable, will be removed from the final enum class FISH = 3 # temporary variable, will be removed from the final enum class
Member Names¶
All enum member objects have an attribute _name_
that contains the member’s
name. They also have a property name
that returns the same name. Type
checkers may infer a literal type for the name of a member:
class Color(Enum):
RED = 1
GREEN = 2
BLUE = 3
reveal_type(Color.RED._name_) # Revealed type is Literal["RED"] (or str)
reveal_type(Color.RED.name) # Revealed type is Literal["RED"] (or str)
def func1(red_or_blue: Literal[Color.RED, Color.BLUE]):
reveal_type(red_or_blue.name) # Revealed type is Literal["RED", "BLUE"] (or str)
def func2(any_color: Color):
reveal_type(any_color.name) # Revealed type is Literal["RED", "BLUE", "GREEN"] (or str)
Member Values¶
All enum member objects have an attribute _value_
that contains the member’s
value. They also have a property value
that returns the same value. Type
checkers may infer the type of a member’s value:
class Color(Enum):
RED = 1
GREEN = 2
BLUE = 3
reveal_type(Color.RED._value_) # Revealed type is Literal[1] (or int or object or Any)
reveal_type(Color.RED.value) # Revealed type is Literal[1] (or int or object or Any)
def func1(red_or_blue: Literal[Color.RED, Color.BLUE]):
reveal_type(red_or_blue.value) # Revealed type is Literal[1, 2] (or int or object or Any)
def func2(any_color: Color):
reveal_type(any_color.value) # Revealed type is Literal[1, 2, 3] (or int or object or Any)
The value of _value_
can be assigned in a constructor method. This technique
is sometimes used to initialize both the member value and non-member attributes.
If the value assigned in the class body is a tuple, the unpacked tuple value is
passed to the constructor. Type checkers may validate consistency between assigned
tuple values and the constructor signature:
class Planet(Enum):
def __init__(self, value: int, mass: float, radius: float):
self._value_ = value
self.mass = mass
self.radius = radius
MERCURY = (1, 3.303e+23, 2.4397e6)
VENUS = (2, 4.869e+24, 6.0518e6)
EARTH = (3, 5.976e+24, 6.37814e6)
MARS = (6.421e+23, 3.3972e6) # Type checker error (optional)
JUPITER = 5 # Type checker error (optional)
reveal_type(Planet.MERCURY.value) # Revealed type is Literal[1] (or int or object or Any)
The class enum.auto
and method _generate_next_value_
can be used within
an enum class to automatically generate values for enum members. Type checkers
may support these to infer literal types for member values:
class Color(Enum):
RED = auto()
GREEN = auto()
BLUE = auto()
reveal_type(Color.RED.value) # Revealed type is Literal[1] (or int or object or Any)
If an enum class provides an explicit type annotation for _value_
, type
checkers should enforce this declared type when values are assigned to
_value_
:
class Color(Enum):
_value_: int
RED = 1 # OK
GREEN = "green" # Type error
class Planet(Enum):
_value_: str
def __init__(self, value: int, mass: float, radius: float):
self._value_ = value # Type error
MERCURY = (1, 3.303e+23, 2.4397e6)
If the literal values for enum members are not supplied, as they sometimes
are not within a type stub file, a type checker can use the type of the
_value_
attribute:
class ColumnType(Enum):
_value_: int
DORIC = ...
IONIC = ...
CORINTHIAN = ...
reveal_type(ColumnType.DORIC.value) # Revealed type is int (or object or Any)
Enum Literal Expansion¶
From the perspective of the type system, most enum classes are equivalent
to the union of the literal members within that enum. (This rule
does not apply to classes that derive from enum.Flag
because these enums
allow flags to be combined in arbitrary ways.) Because of the equivalency
between an enum class and the union of literal members within that enum, the
two types may be used interchangeably. Type checkers may therefore expand
an enum type (that does not derive from enum.Flag
) into a union of
literal values during type narrowing and exhaustion detection:
class Color(Enum):
RED = 1
GREEN = 2
BLUE = 3
def print_color1(c: Color):
if c is Color.RED or c is Color.BLUE:
print("red or blue")
else:
reveal_type(c) # Revealed type is Literal[Color.GREEN]
def print_color2(c: Color):
match c:
case Color.RED | Color.BLUE:
print("red or blue")
case Color.GREEN:
print("green")
case _:
reveal_type(c) # Revealed type is Never
Likewise, a type checker should treat a complete union of all literal members as equivalent to the enum type:
class Answer(Enum):
Yes = 1
No = 2
def func(val: object) -> Answer:
if val is not Answer.Yes and val is not Answer.No:
raise ValueError("Invalid value")
reveal_type(val) # Revealed type is Answer (or Literal[Answer.Yes, Answer.No])
return val # OK