Add `bytearray.__bytes__` to support `typing.SupportsBytes`

[ktbarrett]

Bug report

Bug description:

from typing import SupportsBytes

a = bytearray()
assert isinstance(a, SupportsBytes)  # Fails
bytes(a)  # No failure

bytes.__bytes__ was added in 3.11 so that bytes could match SupportsBytes, so I would think this should work as well.

CPython versions tested on:

3.14

Operating systems tested on:

Linux

[brianschubert]

Linking #68422 for related discussion

[picnixz]

SupportsBytes is meant to expose __bytes__. bytearray doesn't have a __bytes__ method. This means explicitly adding a __bytes__ method for the conversion. While T.__T__ makes sense for primitive types with self being returned, adding a __bytes__ method means specializing the construction bytes(x).

I don't know whether it will be faster or slower but the reason by bytes(bytearray(a)) succeeds is because bytes works for any iterable of 8-bit ints, which bytearray is a special case thereof not because there is a __bytes__ method.

[serhiy-storchaka]

Isn't it because bytearray supports the buffer protocol?

The current behavior is correct. isinstance() returns False because bytearray does not implement the SupportsBytes protocol. Most of other types that are accepted by bytes() (array.array, memoryview, int, list, etc) also do not implement the SupportsBytes protocol.

[ktbarrett]

I want to express "any type that can be used in the expression bytes(obj). Is collections.abc.Buffer the best way to state that? I understand SupportsBytes tests __bytes__ which is why I suggested that as the solution as that was the solution to a similar problem before.

[brianschubert]

I want to express "any type that can be used in the expression bytes(obj). Is collections.abc.Buffer the best way to state that?

For that, you can use Iterable[SupportsIndex] | SupportsIndex | SupportsBytes | Buffer, which is what bytes.__new__ accepts in typeshed