readline.c: Missing NULL check on PyLong_FromLong

Bug report

Bug description:

Original gist: https://gist.github.com/devdanzin/60aed55f44fba133ee2d34d46aa8d197

readline.c: Missing NULL check on PyLong_FromLong in setup_readline

Summary

PyLong_FromLong(0L) at lines 1418-1419 stored in begidx/endidx without NULL check. On OOM during module init, readline.get_begidx() does Py_NewRef(NULL) → segfault.

Reproducer

import _testcapi, sys
if "readline" in sys.modules: del sys.modules["readline"]
for n in range(1, 80):
    if "readline" in sys.modules: del sys.modules["readline"]
    _testcapi.set_nomemory(n, 0)
    try:
        import readline
        _testcapi.remove_mem_hooks()
        break
    except MemoryError:
        _testcapi.remove_mem_hooks()
    except ImportError:
        _testcapi.remove_mem_hooks()
    except:
        _testcapi.remove_mem_hooks()
# Assertion: obj != NULL in PyStackRef_FromPyObjectSteal

The original gist said we should check the result of PyLong_FromLong(0L), but I think this is a false positive, because 0 is small int and should not failed while creating a Python int object from it.

However there are other line of code in this file for calling PyLong_FromLong with begidx/endidx withoud check the result:

https://github.com/python/cpython/blob/d333e5aa59f6cc20198502f4ff50b14eabde9b8b/Modules/readline.c#L1330-L1331

I guess the LLM found a real bug but reported it at the wrong place. Either way, we should fix it.

CPython versions tested on:

CPython main branch

Operating systems tested on:

No response

Linked PRs

• gh-154386

Read Modules/readline.c around setup_readline and the PyLong_FromLong calls near lines 1330-1331; first run the supplied _testcapi memory-hook reproducer and inspect linked PR gh-154386. Done means the affected out-of-memory path no longer permits a NULL object to reach readline accessors, with the relevant CPython tests passing.