On Mon, 12 Oct 2026 15:54:37 GMT, Marius Hanl <[email protected]> wrote:
>> Many tests already fail if you get this wrong (it was a bad commit on my
>> side), it perhaps wasn't 5000 tests that failed, but it was a lot :)
>>
>> I think there even already is a targetted test (but it is so fundamental
>> that many others will fail as well).
>
> I digged deeper into the vetoing and the `logNonConvergence` the last days
> and I really like it.
> Digging into the code, I have one question. I wrote a test which demonstrates
> my question:
>
>
> @Test
> void test() {
> SimpleIntegerProperty property = new SimpleIntegerProperty(0);
> List<Change> firstListenerChanges = new ArrayList<>();
> List<Change> secondListenerChanges = new ArrayList<>();
> List<Change> thirdListenerChanges = new ArrayList<>();
>
> property.addListener((_, old, current) -> {
> System.out.println("L1");
> firstListenerChanges.add(new Change("L1", old, current));
>
> if (current.intValue() == 2) {
> property.set(1);
> }
> });
> property.addListener((_, old, current) -> {
> System.out.println("L2");
> secondListenerChanges.add(new Change("L2", old, current));
> });
> property.addListener((_, old, current) -> {
> System.out.println("L3");
> thirdListenerChanges.add(new Change("L3", old, current));
>
> if (current.intValue() == 1) {
> System.out.println("Set 2");
> property.set(2);
> System.out.println("Set 3");
> property.set(3);
> }
> });
>
> property.set(1);
>
> assertEquals(3, property.get());
> assertConsistentChangeSequence(firstListenerChanges, 0, 3, Set.of(0,
> 1, 2, 3));
> assertConsistentChangeSequence(secondListenerChanges, 0, 3, Set.of(0,
> 1, 2, 3));
> assertConsistentChangeSequence(thirdListenerChanges, 0, 3, Set.of(0,
> 1, 2, 3));
> }
>
>
> It prints out:
>
> L1
> L2
> L3
> Set 2
> L1
> L1
> Oct 12, 2026 5:52:27 PM com.sun.javafx.binding.ListenerList logNonConvergence
> WARNING: IntegerProperty [value: 1] was modified during the invocation of
> multiple listeners, and the values set do not seem to be converging; the
> listener
> test.javafx.beans.ObservableValueTest$$Lambda/0x000000005826df70@2a7686a7
> changed the value from 1 to 2, which was then reset to 1 by another listener;
> to avoid this warning ensure listeners are not making conflicting updates, or
> avoid changing the value in listeners
>
> Set 3
> L1
>
>
> Shouldn't listener 2 (L2) and L3 also be notified in that case?
Yes, they should indeed be notified -- an earlier iteration (when `progress`
was still used for two way communication up and down the call stack) had this
correct, but it has regressed and no test caught it.
The problem is that when a nested loop exits, normally `progress` will be
exactly set to the value that it had when the parent loop called it -- but when
non-convergence happens, it exits the loop hard, and leaves `progress` at some
different value. The fix is to restore `progress` always before exiting a
nested loop.
-------------
PR Review Comment: https://git.openjdk.org/jfx/pull/1081#discussion_r4251642807
