Description

ngoldbaum

opened on Oct 15, 2026

Issue body actions

Originally reported by @krassowski in #158973.

A daemon thread can be stopped while holding the warnings lock. During interpreter finalization, garbage collection triggers a ResourceWarning, deadlocking the finalizing thread:

import threading
import time
import warnings

ready = threading.Event()
warnings.simplefilter("always", ResourceWarning)

class WarnAtShutdown:
    def __init__(self):
        self.self_ref = self  # Defer __del__ until shutdown GC.

    def __del__(self):
        warnings.warn("late", ResourceWarning)

def showwarning(message, *args, **kwargs):
    if str(message) == "hold":
        ready.set()
        time.sleep(0.1)

warnings.showwarning = showwarning
threading.Thread(
    target=lambda: warnings.warn("hold", ResourceWarning),
    daemon=True,
).start()
assert ready.wait(4)
WarnAtShutdown()

Expected: the process exits normally. Actual: it hangs during interpreter shutdown.

• CPython versions tested on: 3.14.8, 3.15.0.
• Operating systems tested on: macOS.

Linked PRs

• gh-159485: Fix warnings lock deadlock during interpreter shutdown #159488

Activity

1. ngoldbaum

   self-assigned this

   on Oct 15, 2026
2. ngoldbaum

   added

   type-bugAn unexpected behavior, bug, or errorAn unexpected behavior, bug, or error

   interpreter-core(Objects, Python, Grammar, and Parser dirs)(Objects, Python, Grammar, and Parser dirs)

   3.14bugs and security fixesbugs and security fixes

   3.15bugs and security fixesbugs and security fixes

   3.16new features, bugs and security fixesnew features, bugs and security fixes

   on Oct 15, 2026
3. ngoldbaum

   added a commit that references this issue on Oct 15, 2026

   pythongh-159485: Fix warnings lock deadlock during interpreter shutdown

   37baa8f
4. ngoldbaum commented on Oct 16, 2026

   

   ngoldbaum

   on Oct 16, 2026

   ContributorAuthor

   More actions

   I updated the reproducer script in the issue to avoid an unbounded wait (it now just does time.sleep(0.1) like test_threading does). This is more realistic and is fixed by @ZeroIntensity's InterpreterGuard approach. There's no need to guard against waiting forever in a callback like my original reproducer was arguing for. There's still a deadlock even without that and the InterpreterGuard fixes it.
5. ngoldbaum

   added 2 commits that reference this issue on Oct 16, 2026

   pythongh-159485: Hold an interpreter guard while using the warnings lock

   fdec193

   pythongh-159485: Use the public interpreter guard API for the warning…

   538a842
6. kevinprimeraquecreo-stack

   added this to Project_VKand Project_VKon Oct 18, 2026
7. github-project-automation

   moved this to In progress in Project_VKon Oct 18, 2026

Issue actions