language-servers/ruff

Fork 0

mirror of https://github.com/astral-sh/ruff.git synced 2025-10-22 08:12:17 +00:00

Eric Mark Martin 33030b34cd

CI / Determine changes (push) Waiting to run

Details

CI / cargo fmt (push) Waiting to run

Details

CI / cargo clippy (push) Blocked by required conditions

Details

CI / cargo test (linux) (push) Blocked by required conditions

Details

CI / cargo test (linux, release) (push) Blocked by required conditions

Details

CI / cargo test (windows) (push) Blocked by required conditions

Details

CI / cargo test (wasm) (push) Blocked by required conditions

Details

CI / cargo build (release) (push) Waiting to run

Details

CI / mkdocs (push) Waiting to run

Details

CI / cargo build (msrv) (push) Blocked by required conditions

Details

CI / cargo fuzz build (push) Blocked by required conditions

Details

CI / fuzz parser (push) Blocked by required conditions

Details

CI / test scripts (push) Blocked by required conditions

Details

CI / ecosystem (push) Blocked by required conditions

Details

CI / Fuzz for new ty panics (push) Blocked by required conditions

Details

CI / cargo shear (push) Blocked by required conditions

Details

CI / python package (push) Waiting to run

Details

CI / pre-commit (push) Waiting to run

Details

CI / formatter instabilities and black similarity (push) Blocked by required conditions

Details

CI / test ruff-lsp (push) Blocked by required conditions

Details

CI / check playground (push) Blocked by required conditions

Details

CI / benchmarks-instrumented (push) Blocked by required conditions

Details

CI / benchmarks-walltime (push) Blocked by required conditions

Details

[ty Playground] Release / publish (push) Waiting to run

Details

[ty] linear variance inference for PEP-695 type parameters (#18713 )

## Summary

Implement linear-time variance inference for type variables
(https://github.com/astral-sh/ty/issues/488).

Inspired by Martin Huschenbett's [PyCon 2025
Talk](https://www.youtube.com/watch?v=7uixlNTOY4s&t=9705s).

## Test Plan

update tests, add new tests, including for mutually recursive classes

---------

Co-authored-by: Carl Meyer <carl@astral.sh>

2025-08-19 17:54:09 -07:00

26 KiB

Raw Blame History

Variance: PEP 695 syntax

[environment]
python-version = "3.12"

Type variables have a property called variance that affects the subtyping and assignability relations. Much more detail can be found in the spec. To summarize, each typevar is either covariant, contravariant, invariant, or bivariant. (Note that bivariance is not currently mentioned in the typing spec, but is a fourth case that we must consider.)

For all of the examples below, we will consider typevars T and U, two generic classes using those typevars C[T] and D[U], and two types A and B.

(Note that dynamic types like Any never participate in subtyping, so C[Any] is neither a subtype nor supertype of any other specialization of C, regardless of T's variance. It is, however, assignable to any specialization of C, regardless of variance, via materialization.)

Covariance

With a covariant typevar, subtyping and assignability are in "alignment": if A <: B and C <: D, then C[A] <: C[B] and C[A] <: D[B].

Types that "produce" data on demand are covariant in their typevar. If you expect a sequence of ints, someone can safely provide a sequence of bools, since each bool element that you would get from the sequence is a valid int.

from ty_extensions import is_assignable_to, is_equivalent_to, is_subtype_of, static_assert, Unknown
from typing import Any

class A: ...
class B(A): ...

class C[T]:
    def receive(self) -> T:
        raise ValueError

class D[U](C[U]):
    pass

static_assert(is_assignable_to(C[B], C[A]))
static_assert(not is_assignable_to(C[A], C[B]))
static_assert(is_assignable_to(C[A], C[Any]))
static_assert(is_assignable_to(C[B], C[Any]))
static_assert(is_assignable_to(C[Any], C[A]))
static_assert(is_assignable_to(C[Any], C[B]))

static_assert(is_assignable_to(D[B], C[A]))
static_assert(not is_assignable_to(D[A], C[B]))
static_assert(is_assignable_to(D[A], C[Any]))
static_assert(is_assignable_to(D[B], C[Any]))
static_assert(is_assignable_to(D[Any], C[A]))
static_assert(is_assignable_to(D[Any], C[B]))

static_assert(is_subtype_of(C[B], C[A]))
static_assert(not is_subtype_of(C[A], C[B]))
static_assert(not is_subtype_of(C[A], C[Any]))
static_assert(not is_subtype_of(C[B], C[Any]))
static_assert(not is_subtype_of(C[Any], C[A]))
static_assert(not is_subtype_of(C[Any], C[B]))

static_assert(is_subtype_of(D[B], C[A]))
static_assert(not is_subtype_of(D[A], C[B]))
static_assert(not is_subtype_of(D[A], C[Any]))
static_assert(not is_subtype_of(D[B], C[Any]))
static_assert(not is_subtype_of(D[Any], C[A]))
static_assert(not is_subtype_of(D[Any], C[B]))

static_assert(is_equivalent_to(C[A], C[A]))
static_assert(is_equivalent_to(C[B], C[B]))
static_assert(not is_equivalent_to(C[B], C[A]))
static_assert(not is_equivalent_to(C[A], C[B]))
static_assert(not is_equivalent_to(C[A], C[Any]))
static_assert(not is_equivalent_to(C[B], C[Any]))
static_assert(not is_equivalent_to(C[Any], C[A]))
static_assert(not is_equivalent_to(C[Any], C[B]))

static_assert(not is_equivalent_to(D[A], C[A]))
static_assert(not is_equivalent_to(D[B], C[B]))
static_assert(not is_equivalent_to(D[B], C[A]))
static_assert(not is_equivalent_to(D[A], C[B]))
static_assert(not is_equivalent_to(D[A], C[Any]))
static_assert(not is_equivalent_to(D[B], C[Any]))
static_assert(not is_equivalent_to(D[Any], C[A]))
static_assert(not is_equivalent_to(D[Any], C[B]))

static_assert(is_equivalent_to(C[Any], C[Any]))
static_assert(is_equivalent_to(C[Any], C[Unknown]))

static_assert(not is_equivalent_to(D[Any], C[Any]))
static_assert(not is_equivalent_to(D[Any], C[Unknown]))

Contravariance

With a contravariant typevar, subtyping and assignability are in "opposition": if A <: B and C <: D, then C[B] <: C[A] and D[B] <: C[A].

Types that "consume" data are contravariant in their typevar. If you expect a consumer that receives bools, someone can safely provide a consumer that expects to receive ints, since each bool that you pass into the consumer is a valid int.

from ty_extensions import is_assignable_to, is_equivalent_to, is_subtype_of, static_assert, Unknown
from typing import Any

class A: ...
class B(A): ...

class C[T]:
    def send(self, value: T): ...

class D[U](C[U]):
    pass

static_assert(not is_assignable_to(C[B], C[A]))
static_assert(is_assignable_to(C[A], C[B]))
static_assert(is_assignable_to(C[A], C[Any]))
static_assert(is_assignable_to(C[B], C[Any]))
static_assert(is_assignable_to(C[Any], C[A]))
static_assert(is_assignable_to(C[Any], C[B]))

static_assert(not is_assignable_to(D[B], C[A]))
static_assert(is_assignable_to(D[A], C[B]))
static_assert(is_assignable_to(D[A], C[Any]))
static_assert(is_assignable_to(D[B], C[Any]))
static_assert(is_assignable_to(D[Any], C[A]))
static_assert(is_assignable_to(D[Any], C[B]))

static_assert(not is_subtype_of(C[B], C[A]))
static_assert(is_subtype_of(C[A], C[B]))
static_assert(not is_subtype_of(C[A], C[Any]))
static_assert(not is_subtype_of(C[B], C[Any]))
static_assert(not is_subtype_of(C[Any], C[A]))
static_assert(not is_subtype_of(C[Any], C[B]))

static_assert(not is_subtype_of(D[B], C[A]))
static_assert(is_subtype_of(D[A], C[B]))
static_assert(not is_subtype_of(D[A], C[Any]))
static_assert(not is_subtype_of(D[B], C[Any]))
static_assert(not is_subtype_of(D[Any], C[A]))
static_assert(not is_subtype_of(D[Any], C[B]))

static_assert(is_equivalent_to(C[A], C[A]))
static_assert(is_equivalent_to(C[B], C[B]))
static_assert(not is_equivalent_to(C[B], C[A]))
static_assert(not is_equivalent_to(C[A], C[B]))
static_assert(not is_equivalent_to(C[A], C[Any]))
static_assert(not is_equivalent_to(C[B], C[Any]))
static_assert(not is_equivalent_to(C[Any], C[A]))
static_assert(not is_equivalent_to(C[Any], C[B]))

static_assert(not is_equivalent_to(D[A], C[A]))
static_assert(not is_equivalent_to(D[B], C[B]))
static_assert(not is_equivalent_to(D[B], C[A]))
static_assert(not is_equivalent_to(D[A], C[B]))
static_assert(not is_equivalent_to(D[A], C[Any]))
static_assert(not is_equivalent_to(D[B], C[Any]))
static_assert(not is_equivalent_to(D[Any], C[A]))
static_assert(not is_equivalent_to(D[Any], C[B]))

static_assert(is_equivalent_to(C[Any], C[Any]))
static_assert(is_equivalent_to(C[Any], C[Unknown]))

static_assert(not is_equivalent_to(D[Any], C[Any]))
static_assert(not is_equivalent_to(D[Any], C[Unknown]))

Invariance

With an invariant typevar, only equivalent specializations of the generic class are subtypes of or assignable to each other.

This often occurs for types that are both producers and consumers, like a mutable list. Iterating over the elements in a list would work with a covariant typevar, just like with the "producer" type above. Appending elements to a list would work with a contravariant typevar, just like with the "consumer" type above. However, a typevar cannot be both covariant and contravariant at the same time!

If you expect a mutable list of ints, it's not safe for someone to provide you with a mutable list of bools, since you might try to add an element to the list: if you try to add an int, the list would no longer only contain elements that are subtypes of bool.

Conversely, if you expect a mutable list of bools, it's not safe for someone to provide you with a mutable list of ints, since you might try to extract elements from the list: you expect every element that you extract to be a subtype of bool, but the list can contain any int.

In the end, if you expect a mutable list, you must always be given a list of exactly that type, since we can't know in advance which of the allowed methods you'll want to use.

from ty_extensions import is_assignable_to, is_equivalent_to, is_subtype_of, static_assert, Unknown
from typing import Any

class A: ...
class B(A): ...

class C[T]:
    def send(self, value: T): ...
    def receive(self) -> T:
        raise ValueError

class D[U](C[U]):
    pass

static_assert(not is_assignable_to(C[B], C[A]))
static_assert(not is_assignable_to(C[A], C[B]))
static_assert(is_assignable_to(C[A], C[Any]))
static_assert(is_assignable_to(C[B], C[Any]))
static_assert(is_assignable_to(C[Any], C[A]))
static_assert(is_assignable_to(C[Any], C[B]))

static_assert(not is_assignable_to(D[B], C[A]))
static_assert(not is_assignable_to(D[A], C[B]))
static_assert(is_assignable_to(D[A], C[Any]))
static_assert(is_assignable_to(D[B], C[Any]))
static_assert(is_assignable_to(D[Any], C[A]))
static_assert(is_assignable_to(D[Any], C[B]))

static_assert(not is_subtype_of(C[B], C[A]))
static_assert(not is_subtype_of(C[A], C[B]))
static_assert(not is_subtype_of(C[A], C[Any]))
static_assert(not is_subtype_of(C[B], C[Any]))
static_assert(not is_subtype_of(C[Any], C[A]))
static_assert(not is_subtype_of(C[Any], C[B]))

static_assert(not is_subtype_of(D[B], C[A]))
static_assert(not is_subtype_of(D[A], C[B]))
static_assert(not is_subtype_of(D[A], C[Any]))
static_assert(not is_subtype_of(D[B], C[Any]))
static_assert(not is_subtype_of(D[Any], C[A]))
static_assert(not is_subtype_of(D[Any], C[B]))

static_assert(is_equivalent_to(C[A], C[A]))
static_assert(is_equivalent_to(C[B], C[B]))
static_assert(not is_equivalent_to(C[B], C[A]))
static_assert(not is_equivalent_to(C[A], C[B]))
static_assert(not is_equivalent_to(C[A], C[Any]))
static_assert(not is_equivalent_to(C[B], C[Any]))
static_assert(not is_equivalent_to(C[Any], C[A]))
static_assert(not is_equivalent_to(C[Any], C[B]))

static_assert(not is_equivalent_to(D[A], C[A]))
static_assert(not is_equivalent_to(D[B], C[B]))
static_assert(not is_equivalent_to(D[B], C[A]))
static_assert(not is_equivalent_to(D[A], C[B]))
static_assert(not is_equivalent_to(D[A], C[Any]))
static_assert(not is_equivalent_to(D[B], C[Any]))
static_assert(not is_equivalent_to(D[Any], C[A]))
static_assert(not is_equivalent_to(D[Any], C[B]))

static_assert(is_equivalent_to(C[Any], C[Any]))
static_assert(is_equivalent_to(C[Any], C[Unknown]))

static_assert(not is_equivalent_to(D[Any], C[Any]))
static_assert(not is_equivalent_to(D[Any], C[Unknown]))

Bivariance

With a bivariant typevar, all specializations of the generic class are assignable to (and in fact, gradually equivalent to) each other, and all fully static specializations are subtypes of (and equivalent to) each other.

This is a bit of pathological case, which really only happens when the class doesn't use the typevar at all. (If it did, it would have to be covariant, contravariant, or invariant, depending on how the typevar was used.)

from ty_extensions import is_assignable_to, is_equivalent_to, is_subtype_of, static_assert, Unknown
from typing import Any

class A: ...
class B(A): ...

class C[T]:
    pass

class D[U](C[U]):
    pass

static_assert(is_assignable_to(C[B], C[A]))
static_assert(is_assignable_to(C[A], C[B]))
static_assert(is_assignable_to(C[A], C[Any]))
static_assert(is_assignable_to(C[B], C[Any]))
static_assert(is_assignable_to(C[Any], C[A]))
static_assert(is_assignable_to(C[Any], C[B]))

static_assert(is_assignable_to(D[B], C[A]))
static_assert(is_subtype_of(C[A], C[A]))
static_assert(is_assignable_to(D[A], C[B]))
static_assert(is_assignable_to(D[A], C[Any]))
static_assert(is_assignable_to(D[B], C[Any]))
static_assert(is_assignable_to(D[Any], C[A]))
static_assert(is_assignable_to(D[Any], C[B]))

static_assert(is_subtype_of(C[B], C[A]))
static_assert(is_subtype_of(C[A], C[B]))
static_assert(not is_subtype_of(C[A], C[Any]))
static_assert(not is_subtype_of(C[B], C[Any]))
static_assert(not is_subtype_of(C[Any], C[A]))
static_assert(not is_subtype_of(C[Any], C[B]))
static_assert(not is_subtype_of(C[Any], C[Any]))

static_assert(is_subtype_of(D[B], C[A]))
static_assert(is_subtype_of(D[A], C[B]))
static_assert(not is_subtype_of(D[A], C[Any]))
static_assert(not is_subtype_of(D[B], C[Any]))
static_assert(not is_subtype_of(D[Any], C[A]))
static_assert(not is_subtype_of(D[Any], C[B]))

static_assert(is_equivalent_to(C[A], C[A]))
static_assert(is_equivalent_to(C[B], C[B]))
static_assert(is_equivalent_to(C[B], C[A]))
static_assert(is_equivalent_to(C[A], C[B]))
static_assert(is_equivalent_to(C[A], C[Any]))
static_assert(is_equivalent_to(C[B], C[Any]))
static_assert(is_equivalent_to(C[Any], C[A]))
static_assert(is_equivalent_to(C[Any], C[B]))

static_assert(not is_equivalent_to(D[A], C[A]))
static_assert(not is_equivalent_to(D[B], C[B]))
static_assert(not is_equivalent_to(D[B], C[A]))
static_assert(not is_equivalent_to(D[A], C[B]))
static_assert(not is_equivalent_to(D[A], C[Any]))
static_assert(not is_equivalent_to(D[B], C[Any]))
static_assert(not is_equivalent_to(D[Any], C[A]))
static_assert(not is_equivalent_to(D[Any], C[B]))

static_assert(is_equivalent_to(C[Any], C[Any]))
static_assert(is_equivalent_to(C[Any], C[Unknown]))

static_assert(not is_equivalent_to(D[Any], C[Any]))
static_assert(not is_equivalent_to(D[Any], C[Unknown]))

Mutual Recursion

This example due to Martin Huschenbett's PyCon 2025 talk, Linear Time variance Inference for PEP 695

from ty_extensions import is_subtype_of, static_assert
from typing import Any

class A: ...
class B(A): ...

class C[X]:
    def f(self) -> "D[X]":
        return D()

    def g(self, x: X) -> None: ...

class D[Y]:
    def h(self) -> C[Y]:
        return C()

C is contravariant in X, and D in Y:

C has two occurrences of X
- X occurs in the return type of f as D[X] (X is substituted in for Y)
  - D has one occurrence of Y
    - Y occurs in the return type of h as C[Y]
- X occurs contravariantly as a parameter in g

Thus the variance of X in C depends on itself. We want to infer the least restrictive possible variance, so in such cases we begin by assuming that the point where we detect the cycle is bivariant.

If we thus assume X is bivariant in C, then Y will be bivariant in D, as D's only occurrence of Y is in C. Then we consider X in C once more. We have two occurrences: D[X] covariantly in a return type, and X contravariantly in an argument type. With one bivariant and one contravariant occurrence, we update our inference of X in C to contravariant---the supremum of contravariant and bivariant in the lattice.

Now that we've updated the variance of X in C, we re-evaluate Y in D. It only has the one occurrence C[Y], which we now infer is contravariant, and so we infer contravariance for Y in D as well.

Because the variance of X in C depends on that of Y in D, we have to re-evaluate now that we've updated the latter to contravariant. The variance of X in C is now the supremum of contravariant and contravariant---giving us contravariant---and so remains unchanged.

Once we've completed a turn around the cycle with nothing changed, we've reached a fixed-point---the variance inference will not change any further---and so we finally conclude that both X in C and Y in D are contravariant.

static_assert(not is_subtype_of(C[B], C[A]))
static_assert(is_subtype_of(C[A], C[B]))
static_assert(not is_subtype_of(C[A], C[Any]))
static_assert(not is_subtype_of(C[B], C[Any]))
static_assert(not is_subtype_of(C[Any], C[A]))
static_assert(not is_subtype_of(C[Any], C[B]))

static_assert(not is_subtype_of(D[B], D[A]))
static_assert(is_subtype_of(D[A], D[B]))
static_assert(not is_subtype_of(D[A], D[Any]))
static_assert(not is_subtype_of(D[B], D[Any]))
static_assert(not is_subtype_of(D[Any], D[A]))
static_assert(not is_subtype_of(D[Any], D[B]))

Class Attributes

Mutable Attributes

Normal attributes are mutable, and so make the enclosing class invariant in this typevar (see [inv]).

from ty_extensions import is_subtype_of, static_assert

class A: ...
class B(A): ...

class C[T]:
    x: T

static_assert(not is_subtype_of(C[B], C[A]))
static_assert(not is_subtype_of(C[A], C[B]))

One might think that occurrences in the types of normal attributes are covariant, but they are mutable, and thus the occurrences are invariant.

Immutable Attributes

Immutable attributes can't be written to, and thus constrain the typevar to covariance, not invariance.

Final attributes

from typing import Final
from ty_extensions import is_subtype_of, static_assert

class A: ...
class B(A): ...

class C[T]:
    x: Final[T]

static_assert(is_subtype_of(C[B], C[A]))
static_assert(not is_subtype_of(C[A], C[B]))

Underscore-prefixed attributes

Underscore-prefixed instance attributes are considered private, and thus are assumed not externally mutated.

from ty_extensions import is_subtype_of, static_assert

class A: ...
class B(A): ...

class C[T]:
    _x: T

    @property
    def x(self) -> T:
        return self._x

static_assert(is_subtype_of(C[B], C[A]))
static_assert(not is_subtype_of(C[A], C[B]))

class D[T]:
    def __init__(self, x: T):
        self._x = x

    @property
    def x(self) -> T:
        return self._x

static_assert(is_subtype_of(D[B], D[A]))
static_assert(not is_subtype_of(D[A], D[B]))

Frozen dataclasses in Python 3.12 and earlier

from dataclasses import dataclass, field
from ty_extensions import is_subtype_of, static_assert

class A: ...
class B(A): ...

@dataclass(frozen=True)
class D[U]:
    y: U

static_assert(is_subtype_of(D[B], D[A]))
static_assert(not is_subtype_of(D[A], D[B]))

@dataclass(frozen=True)
class E[U]:
    y: U = field()

static_assert(is_subtype_of(E[B], E[A]))
static_assert(not is_subtype_of(E[A], E[B]))

Frozen dataclasses in Python 3.13 and later

[environment]
python-version = "3.13"

Python 3.13 introduced a new synthesized __replace__ method on dataclasses, which uses every field type in a contravariant position (as a parameter to __replace__). This means that frozen dataclasses on Python 3.13+ can't be covariant in their field types.

from dataclasses import dataclass
from ty_extensions import is_subtype_of, static_assert

class A: ...
class B(A): ...

@dataclass(frozen=True)
class D[U]:
    y: U

static_assert(not is_subtype_of(D[B], D[A]))
static_assert(not is_subtype_of(D[A], D[B]))

NamedTuple

from typing import NamedTuple
from ty_extensions import is_subtype_of, static_assert

class A: ...
class B(A): ...

class E[V](NamedTuple):
    z: V

static_assert(is_subtype_of(E[B], E[A]))
static_assert(not is_subtype_of(E[A], E[B]))

A subclass of a NamedTuple can still be covariant:

class D[T](E[T]):
    pass

static_assert(is_subtype_of(D[B], D[A]))
static_assert(not is_subtype_of(D[A], D[B]))

But adding a new generic attribute on the subclass makes it invariant (the added attribute is not a NamedTuple field, and thus not immutable):

class C[T](E[T]):
    w: T

static_assert(not is_subtype_of(C[B], C[A]))
static_assert(not is_subtype_of(C[A], C[B]))

Properties

Properties constrain to covariance if they are get-only and invariant if they are get-set:

from ty_extensions import static_assert, is_subtype_of

class A: ...
class B(A): ...

class C[T]:
    @property
    def x(self) -> T | None:
        return None

class D[U]:
    @property
    def y(self) -> U | None:
        return None

    @y.setter
    def y(self, value: U): ...

static_assert(is_subtype_of(C[B], C[A]))
static_assert(not is_subtype_of(C[A], C[B]))
static_assert(not is_subtype_of(D[B], D[A]))
static_assert(not is_subtype_of(D[A], D[B]))

Implicit Attributes

Implicit attributes work like normal ones

from ty_extensions import static_assert, is_subtype_of

class A: ...
class B(A): ...

class C[T]:
    def f(self) -> None:
        self.x: T | None = None

static_assert(not is_subtype_of(C[B], C[A]))
static_assert(not is_subtype_of(C[A], C[B]))

Constructors: excluding `init` and `new`

We consider it invalid to call __init__ explicitly on an existing object. Likewise, __new__ is only used at the beginning of an object's life. As such, we don't need to worry about the variance impact of these methods.

from ty_extensions import static_assert, is_subtype_of

class A: ...
class B(A): ...

class C[T]:
    def __init__(self, x: T): ...
    def __new__(self, x: T): ...

static_assert(is_subtype_of(C[B], C[A]))
static_assert(is_subtype_of(C[A], C[B]))

This example is then bivariant because it doesn't use T outside of the two exempted methods.

This holds likewise for dataclasses with synthesized __init__:

from dataclasses import dataclass

@dataclass(init=True, frozen=True)
class D[T]:
    x: T

# Covariant due to the read-only T-typed attribute; the `__init__` is ignored and doesn't make it
# invariant:

static_assert(is_subtype_of(D[B], D[A]))
static_assert(not is_subtype_of(D[A], D[B]))

Union Types

Union types are covariant in all their members. If A <: B, then A | C <: B | C and C | A <: C | B.

from ty_extensions import is_assignable_to, is_subtype_of, static_assert

class A: ...
class B(A): ...
class C: ...

# Union types are covariant in their members
static_assert(is_subtype_of(B | C, A | C))
static_assert(is_subtype_of(C | B, C | A))
static_assert(not is_subtype_of(A | C, B | C))
static_assert(not is_subtype_of(C | A, C | B))

# Assignability follows the same pattern
static_assert(is_assignable_to(B | C, A | C))
static_assert(is_assignable_to(C | B, C | A))
static_assert(not is_assignable_to(A | C, B | C))
static_assert(not is_assignable_to(C | A, C | B))

Intersection Types

Intersection types cannot be expressed directly in Python syntax, but they occur when type narrowing creates constraints through control flow. In ty's representation, intersection types are covariant in their positive conjuncts and contravariant in their negative conjuncts.

from ty_extensions import is_assignable_to, is_subtype_of, static_assert, Intersection, Not

class A: ...
class B(A): ...
class C: ...

# Test covariance in positive conjuncts
# If B <: A, then Intersection[X, B] <: Intersection[X, A]
static_assert(is_subtype_of(Intersection[C, B], Intersection[C, A]))
static_assert(not is_subtype_of(Intersection[C, A], Intersection[C, B]))

static_assert(is_assignable_to(Intersection[C, B], Intersection[C, A]))
static_assert(not is_assignable_to(Intersection[C, A], Intersection[C, B]))

# Test contravariance in negative conjuncts
# If B <: A, then Intersection[X, Not[A]] <: Intersection[X, Not[B]]
# (excluding supertype A is more restrictive than excluding subtype B)
static_assert(is_subtype_of(Intersection[C, Not[A]], Intersection[C, Not[B]]))
static_assert(not is_subtype_of(Intersection[C, Not[B]], Intersection[C, Not[A]]))

static_assert(is_assignable_to(Intersection[C, Not[A]], Intersection[C, Not[B]]))
static_assert(not is_assignable_to(Intersection[C, Not[B]], Intersection[C, Not[A]]))

Subclass Types (type[T])

The type[T] construct represents the type of classes that are subclasses of T. It is covariant in T because if A <: B, then type[A] <: type[B] holds.

from ty_extensions import is_assignable_to, is_subtype_of, static_assert

class A: ...
class B(A): ...

# type[T] is covariant in T
static_assert(is_subtype_of(type[B], type[A]))
static_assert(not is_subtype_of(type[A], type[B]))

static_assert(is_assignable_to(type[B], type[A]))
static_assert(not is_assignable_to(type[A], type[B]))

# With generic classes using type[T]
class ClassContainer[T]:
    def __init__(self, cls: type[T]) -> None:
        self.cls = cls

    def create_instance(self) -> T:
        return self.cls()

# ClassContainer is covariant in T due to type[T]
static_assert(is_subtype_of(ClassContainer[B], ClassContainer[A]))
static_assert(not is_subtype_of(ClassContainer[A], ClassContainer[B]))

static_assert(is_assignable_to(ClassContainer[B], ClassContainer[A]))
static_assert(not is_assignable_to(ClassContainer[A], ClassContainer[B]))

# Practical example: you can pass a ClassContainer[B] where ClassContainer[A] is expected
# because type[B] can safely be used where type[A] is expected
def use_a_class_container(container: ClassContainer[A]) -> A:
    return container.create_instance()

b_container = ClassContainer[B](B)
a_instance: A = use_a_class_container(b_container)  # This should work

Inheriting from generic classes with inferred variance

When inheriting from a generic class with our type variable substituted in, we count its occurrences as well. In the following example, T is covariant in C, and contravariant in the subclass D if you only count its own occurrences. Because we count both then, T is invariant in D.

from ty_extensions import is_subtype_of, static_assert

class A:
    pass

class B(A):
    pass

class C[T]:
    def f() -> T | None:
        pass

static_assert(is_subtype_of(C[B], C[A]))
static_assert(not is_subtype_of(C[A], C[B]))

class D[T](C[T]):
    def g(x: T) -> None:
        pass

static_assert(not is_subtype_of(D[B], D[A]))
static_assert(not is_subtype_of(D[A], D[B]))

Inheriting from generic classes with explicit variance

from typing import TypeVar, Generic
from ty_extensions import is_subtype_of, static_assert

T = TypeVar("T")
T_co = TypeVar("T_co", covariant=True)
T_contra = TypeVar("T_contra", contravariant=True)

class A:
    pass

class B(A):
    pass

class Invariant(Generic[T]):
    pass

static_assert(not is_subtype_of(Invariant[B], Invariant[A]))
static_assert(not is_subtype_of(Invariant[A], Invariant[B]))

class DerivedInvariant[T](Invariant[T]):
    pass

static_assert(not is_subtype_of(DerivedInvariant[B], DerivedInvariant[A]))
static_assert(not is_subtype_of(DerivedInvariant[A], DerivedInvariant[B]))

class Covariant(Generic[T_co]):
    pass

static_assert(is_subtype_of(Covariant[B], Covariant[A]))
static_assert(not is_subtype_of(Covariant[A], Covariant[B]))

class DerivedCovariant[T](Covariant[T]):
    pass

static_assert(is_subtype_of(DerivedCovariant[B], DerivedCovariant[A]))
static_assert(not is_subtype_of(DerivedCovariant[A], DerivedCovariant[B]))

class Contravariant(Generic[T_contra]):
    pass

static_assert(not is_subtype_of(Contravariant[B], Contravariant[A]))
static_assert(is_subtype_of(Contravariant[A], Contravariant[B]))

class DerivedContravariant[T](Contravariant[T]):
    pass

static_assert(not is_subtype_of(DerivedContravariant[B], DerivedContravariant[A]))
static_assert(is_subtype_of(DerivedContravariant[A], DerivedContravariant[B]))

26 KiB Raw Blame History

Variance: PEP 695 syntax

Covariance

Contravariance

Invariance

Bivariance

Mutual Recursion

Class Attributes

Mutable Attributes

Immutable Attributes

Final attributes

Underscore-prefixed attributes

Frozen dataclasses in Python 3.12 and earlier

Frozen dataclasses in Python 3.13 and later

NamedTuple

Properties

Implicit Attributes

Constructors: excluding __init__ and __new__

Union Types

Intersection Types

Subclass Types (type[T])

Inheriting from generic classes with inferred variance

Inheriting from generic classes with explicit variance

26 KiB

Raw Blame History

Constructors: excluding `init` and `new`