gh-115832: Fix instrumentation version mismatch during interpreter shutdown (#115856)

A previous commit introduced a bug to `interpreter_clear()`: it set
`interp->ceval.instrumentation_version` to 0, without making the corresponding
change to `tstate->eval_breaker` (which holds a thread-local copy of the
version). After this happens, Python code can still run due to object finalizers
during a GC, and the version check in bytecodes.c will see a different result
than the one in instrumentation.c causing an infinite loop.

The fix itself is straightforward: clear `tstate->eval_breaker` when clearing
`interp->ceval.instrumentation_version`.

1 files changed, 30 insertions, 0 deletions

diff --git a/Lib/test/_test_monitoring_shutdown.py b/Lib/test/_test_monitoring_shutdown.py
new file mode 100644
index 00000000000..3d0fbec029d
--- /dev/null
+++ b/Lib/test/_test_monitoring_shutdown.py
@@ -0,0 +1,30 @@
+#!/usr/bin/env python3
+
+# gh-115832: An object destructor running during the final GC of interpreter
+# shutdown triggered an infinite loop in the instrumentation code.
+
+import sys
+
+class CallableCycle:
+    def __init__(self):
+        self._cycle = self
+
+    def __del__(self):
+        pass
+
+    def __call__(self, code, instruction_offset):
+        pass
+
+def tracefunc(frame, event, arg):
+    pass
+
+def main():
+    tool_id = sys.monitoring.PROFILER_ID
+    event_id = sys.monitoring.events.PY_START
+
+    sys.monitoring.use_tool_id(tool_id, "test profiler")
+    sys.monitoring.set_events(tool_id, event_id)
+    sys.monitoring.register_callback(tool_id, event_id, CallableCycle())
+
+if __name__ == "__main__":
+    sys.exit(main())