Bug report

Bug description:

The list/tuple fast path in bytearray.__init__() resizes the object to len(arg) before converting the items. When an exact int is outside range(256) it raises ValueError without undoing the resize, so an object that outlives the failed call (a re-initialized bytearray, or a subclass catching the error from super().__init__()) keeps the full length, with bytes that were never written.

b = bytearray(b'abc')
try:
    b.__init__([1, 2, 300])
except ValueError:
    pass
print(len(b), b)

This prints a length of 3; the last byte is leftover heap content (0xcd on a debug build). Expected 0 bytearray(b''), which is what a non-int item ([1, 2, 1.5]) or an iterator (iter([1, 2, 300]), which leaves b'\x01\x02') gives. With [300] * n the whole object is stale data from a recently freed block. The buffer branch has the same shape when PyBuffer_ToContiguous() fails.

The fast path dates back to gh-91149 (3.11).

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-160179: Leave bytearray empty when __init__() fails midway #160180
• [3.15] gh-160179: Leave bytearray empty when __init__() fails midway (GH-160180) #160202
• [3.14] gh-160179: Leave bytearray empty when __init__() fails midway (GH-160180) #160204