[heads up] segmentation fault with Python 3.15 when running test_badcompresscopy -> when zlib_Compress_copy initialization fails

[befeleme]

I am testing Python libraries in Fedora Linux 45 with Python 3.15 and have encountered a segmentation fault with python-zlib-ng.

This issue comes when running test_badcompresscopy, and it's related to how zlib_Compress_copy works.
In case initialization fails, goto error is executed:

python-zlib-ng/src/zlib_ng/zlib_ngmodule.c

Lines 802 to 804 in 07c7f5f

	if (!self->is_initialised) {
	PyErr_SetString(PyExc_ValueError, "Cannot copy flushed objects.");
	goto error;

That, as a first thing, calls LEAVE_ZLIB, which substitutes PyThread_release_lock((obj)->lock);

python-zlib-ng/src/zlib_ng/zlib_ngmodule.c

Lines 840 to 843 in 07c7f5f

	error:
	LEAVE_ZLIB(self);
	Py_XDECREF(return_value);
	return NULL;

This, however, cannot be successful, because the thread wasn't acquired in the first place: ENTER_ZLIB is called AFTER the initialization check, on line 810:

python-zlib-ng/src/zlib_ng/zlib_ngmodule.c

Line 810 in 07c7f5f

ENTER_ZLIB(self);

This happens with Python 3.15 and not before, because the implementation of PyThread_release_lock has changed to use PyMutex objects, which apparently behaves differently, aborting entirely: python/cpython#134745

To fix this, the goto error part could be replaced with just the decref and return, without the LEAVE_ZLIB part.

-        goto error;                                                                                       
+        Py_DECREF(return_value);                                                                          
+        return NULL;  

I have verified the fix works in my testing environment (the tests passes, package relying on this one also builds). I can submit it if you deem it correct for the issue.

Minimal reproducer:

import copy
from zlib_ng import zlib_ng

c = zlib_ng.compressobj()
c.compress(b"x")
c.flush()
copy.copy(c)

When run with Python 3.14 it raises ValueError: Cannot copy flushed objects., with Python 3.15 it's a segmentation fault.

[rhpvorderman]

Thanks. And also thanks for the reproducer. I will add this to the todo.