All commit authors signed the Contributor License Agreement.

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

We don't have any guarantees about reference counting,

I know that very well. But…

6 years ago, there was a difficult to debug issue with SageMath sagemath/sage#24986 , and the developers concluded using that hacky reference counting solution is the best available workaround.

Turns out it causes additional issues later: sagemath/sage#39224

There exists a better solution, see sagemath/cysignals#215 (comment) , but while it is not yet implemented (there are a lot of places that need to be changed in the code base) I try to make the best of what is available.

If the developers originally implemented the proper solution, we wouldn't be in this situation. But removing the workaround entirely before implementing the proper solution seems impractical at this point.

(in the meantime, I'll try to figure out how to write test case and blurb, but I'd appreciate any help if possible)

(in the meantime, I'll try to figure out how to write test case and blurb, but I'd appreciate any help.)

Feel free to tag me if you need any help. The devguide should have everything you need, though :)

Feel best if I leave this as draft until I figure out what's the root cause of the change in behavior Python 3.11 and Python 3.12.

Bisecting results in 1e197e6 being the first bad commit.

Seems to make sense…? If more references to frames are added then it's more likely to create cyclic references…?

issue: #96421
pull request: #96319

More investigation shows the frame object has f_back cleared when the function exits in Python 3.11, while this is not the case in Python 3.12.

I haven't fully understood what the linked pull request does yet, but maybe if only ensure the shim frame is cleared without clearing the actual frame…?

@markshannon Any idea about this?

from contextlib import contextmanager

class MyException(Exception):
	def __del__(self):
		print(self.__traceback__.tb_frame.f_back)
		print("deleted")

@contextmanager
def f():
	try:
		yield
	except MyException as e:
		global g
		g = e
		print(e.__traceback__.tb_frame.f_back)

with f():
	raise MyException()

print(g.__traceback__.tb_frame.f_back)  # ⟵ this prints None in Python 3.11 and <frame…> in Python 3.12

print("done")

I think that this PR (which likely will be closed in favor of more general solution) is not the best place for discussion.

This PR is stale because it has been open for 30 days with no activity.