https://github.com/python/cpython/blob/master/Lib/random.py#L157
If bytearray is passed as a seed, the function will change it.
It either has to be documented, or the implementation should change.

I don't see a bug here. As documented, "a str, bytes, or bytearray object gets converted to an int and all of its bits are used." The various input types are converted to an integer and it doesn't really matter how it is done as long as it is consistent. FWIW, the docs link to the source code so it can be examined if needed.

I find the following behaviour very confusing:

>>> import random
>>> a = bytearray("1234", "utf-8")
>>> b = bytearray("1234", "utf-8")
>>> a == b
True
>>> random.seed(a)
>>> a == b
False
>>> a
bytearray(b'1234\xd4\x04U\x9f`.\xabo\xd6\x02\xacv\x80\xda\xcb\xfa\xad\xd1603^\x95\x1f\tz\xf3\x90\x0e\x9d\xe1v\xb6\xdb(Q/.\x00\x0b\x9d\x04\xfb\xa5\x13>\x8b\x1cn\x8d\xf5\x9d\xb3\xa8\xab\x9d`\xbeK\x97\xcc\x9e\x81\xdb')
>>> b
bytearray(b'1234')

The function doesn't document it will change the input argument

The problem is that random seed will do

            if isinstance(a, str):
                a = a.encode()
            a += _sha512(a).digest()
            a = int.from_bytes(a, 'big')

and that will modify the bytearray in place.

>>> a = bytearray("1234", "utf-8")
>>> a += b"digest"
>>> a
bytearray(b'1234digest')

IMHO, seed shouldn't modify the input. Since str, and bytes are immutable that will only happen when passing a bytearray which is not consistent.

One solution to these would be making a copy of the parameter using copy.deepcopy and then performing operations on that copy. (The downside of this solution is that it will change the behavior.)

Okay, I now understand your report and will prepare a fix.

New changeset e733e99 by Miss Islington (bot) in branch '3.9':
bpo-44018: random.seed() no longer mutates its inputs (GH-25856) (GH-25864)
e733e99

New changeset 2995bff by Miss Islington (bot) in branch '3.10':
bpo-44018: random.seed() no longer mutates its inputs (GH-25856) (GH-25872)
2995bff