[Mono] Reentrant Monitor.Wait unexpected behavior

[eduardo-vp]

There's a current bug that can happen when using synchronization contexts that allow message pumping:

• The main thread calls Monitor.Wait on lock A.
• SynchronizationContext.Wait pumps a message and the main thread calls Monitor.Wait on lock B.
• A background thread calls Monitor.Pulse on lock A but lock B steals the signal.

runtime/src/libraries/System.Threading/tests/MonitorTests.cs

Lines 518 to 560 in d343974

	public static void ReentrantWaitFromSyncContextTest()
	{
	// Since we set the SynchronizationContext for the current thread, we run
	// this test in a background thread to avoid affecting other tests.
	ThreadTestHelpers.RunTestInBackgroundThread(() =>
	{
	object lockA = new();
	object lockB = new();
	bool innerWaitResult = true; // should become false (lockB is never pulsed)
	var pumpCtx = new ReentrantWaitSyncContext(() =>
	{
	lock (lockB)
	{
	innerWaitResult = Monitor.Wait(lockB, 500);
	}
	});
	SynchronizationContext.SetSynchronizationContext(pumpCtx);
	var t = new Thread(() =>
	{
	Thread.Sleep(100);
	lock (lockA) { Monitor.Pulse(lockA); }
	}) { IsBackground = true };
	t.Start();
	bool outerResult;
	var sw = System.Diagnostics.Stopwatch.StartNew();
	lock (lockA)
	{
	outerResult = Monitor.Wait(lockA, FailTimeoutMilliseconds);
	}
	long elapsed = sw.ElapsedMilliseconds;
	Assert.True(outerResult, "Outer Monitor.Wait should have been pulsed");
	Assert.True(elapsed < FailTimeoutMilliseconds / 2,
	$"Outer Monitor.Wait took {elapsed}ms — signal was likely stolen by inner wait");
	Assert.False(innerWaitResult,
	"Inner Monitor.Wait(lockB) should return false (lockB was never pulsed)");
	t.Join(FailTimeoutMilliseconds);
	});
	}

[dotnet-policy-service]

Tagging subscribers to this area: @agocke, @VSadov
See info in area-owners.md if you want to be subscribed.