`codecs._Decoder` is incorrectly annotated as taking `input: bytes`, while at runtime it receives `memoryview`

[denballakh]

How to check runtime bevahiour:

# main.py
import codecs

def decode(input: bytes, errors: str = 'strict') -> tuple[str, int]:
    print(type(input))
    return '', len(input)

def encode(input: str, errors: str = 'strict') -> tuple[bytes, int]:
    raise NotImplementedError

def search(name: str) -> codecs.CodecInfo | None:
    if name == 'test':
        return codecs.CodecInfo(encode, decode)
    return None

codecs.register(search)
import a

# a.py:
# coding: test
a = 1

When i run this code, decode prints <class 'memoryview'>, which is inconsistent with codecs._Decoder annotation.
Annotating input as memoryview doesnt help, because codecs.CodecInfo accepts only function that takes input arg of type bytes.

[Avasam]

Linking #9006 since it relates to potentially incorrect bytes annotation in stdlib.

[srittau]

PR to fix this welcome!

[Akuli]

With #12610, decode functions should be defined like this:

from typing_extensions import Buffer
# alternatively: from collections.abc import Buffer (new in python 3.12)

def decode(input: Buffer, errors: str = 'strict') -> tuple[str, int]:
    # do something with input
    # use bytes(input) or bytearray(input) if needed