From: Sam Gross <colesbury@gmail.com>
Date: Mon, 15 Apr 2024 16:54:56 +0000 (-0400)
Subject: gh-117688: Fix deadlock in test_no_stale_references with GIL disabled (#117720)
X-Git-Tag: v3.13.0b1~416
X-Git-Url: http://git.ipfire.org/gitweb.cgi?a=commitdiff_plain;h=520cf2170ea08730e142d591e311b7ab8a6afe63;p=thirdparty%2FPython%2Fcpython.git

gh-117688: Fix deadlock in test_no_stale_references with GIL disabled (#117720)

Check `my_object_collected.wait()` in a loop to give the main thread a
chance to merge the reference count fields. Additionally, call
`my_object_collected.set()` in a background thread to avoid deadlocking
when the destructor is called asynchronously via the eval breaker
within the body of of `my_object_collected.wait()`.
---

diff --git a/Lib/test/test_concurrent_futures/executor.py b/Lib/test/test_concurrent_futures/executor.py
index 6a79fe69ec37..3049bb748614 100644
--- a/Lib/test/test_concurrent_futures/executor.py
+++ b/Lib/test/test_concurrent_futures/executor.py
@@ -83,24 +83,34 @@ class ExecutorTest:
         # references.
         my_object = MyObject()
         my_object_collected = threading.Event()
-        my_object_callback = weakref.ref(
-            my_object, lambda obj: my_object_collected.set())
-        fut = self.executor.submit(my_object.my_method)
+        def set_event():
+            if Py_GIL_DISABLED:
+                # gh-117688 Avoid deadlock by setting the event in a
+                # background thread. The current thread may be in the middle
+                # of the my_object_collected.wait() call, which holds locks
+                # needed by my_object_collected.set().
+                threading.Thread(target=my_object_collected.set).start()
+            else:
+                my_object_collected.set()
+        my_object_callback = weakref.ref(my_object, lambda obj: set_event())
+        # Deliberately discarding the future.
+        self.executor.submit(my_object.my_method)
         del my_object
 
         if Py_GIL_DISABLED:
             # Due to biased reference counting, my_object might only be
             # deallocated while the thread that created it runs -- if the
             # thread is paused waiting on an event, it may not merge the
-            # refcount of the queued object. For that reason, we wait for the
-            # task to finish (so that it's no longer referenced) and force a
-            # GC to ensure that it is collected.
-            fut.result()  # Wait for the task to finish.
-            support.gc_collect()
+            # refcount of the queued object. For that reason, we alternate
+            # between running the GC and waiting for the event.
+            wait_time = 0
+            collected = False
+            while not collected and wait_time <= support.SHORT_TIMEOUT:
+                support.gc_collect()
+                collected = my_object_collected.wait(timeout=1.0)
+                wait_time += 1.0
         else:
-            del fut  # Deliberately discard the future.
-
-        collected = my_object_collected.wait(timeout=support.SHORT_TIMEOUT)
+            collected = my_object_collected.wait(timeout=support.SHORT_TIMEOUT)
         self.assertTrue(collected,
                         "Stale reference not collected within timeout.")