Bug report

Bug description:

A reference cycle between a str subclass and its re.finditer() iterator is not collected:

import gc
import re
import weakref


class Text(str):
    pass


def create_cycle():
    text = Text("aaa")
    text.iterator = re.finditer("a+", text)
    return weakref.ref(text)


ref = create_cycle()
for _ in range(3):
    gc.collect()
print("collected:", ref() is None)

Expected: collected: True

Actual: collected: False

This also affects Pattern.scanner() and bytearray subclass inputs. For bytes-like inputs, PyObject_GetBuffer() acquires a strong reference stored in state.buffer.obj. scanner_traverse() omits the owned references in both state.string and state.buffer.obj.

AI assistance: Codex helped investigate, test, and draft this report.

CPython versions tested on:

CPython main branch, 3.14, 3.15

Operating systems tested on:

macOS

Linked PRs

• gh-158095: Fix reference cycles in re scanners #158100
• [3.14] gh-158095: Fix reference cycles in re scanners (GH-158100) #161245
• [3.15] gh-158095: Fix reference cycles in re scanners (GH-158100) #161246