On Fri, 3 Jul 2026 12:59:36 GMT, Jose Pereda <[email protected]> wrote:

>> This PR captures the `currentValue` of the fired value change event in 
>> `ExpressionHelper$Generic::fireValueChangedEvent`,  before entering the 
>> notification loop, in order to pass it down to all the change listeners. 
>> 
>> This prevents a potential issue that could happen if all the change 
>> listeners are removed before the notification loop ends: the `currentValue` 
>> is nullified, and the pending listeners would unexpectedly receive null, 
>> instead of the value that was correctly sent to the previous listeners.
>> 
>> Some unit tests have been included, all of them but one pass before and 
>> after the proposed fix, and the last one fails before the fix and passes 
>> after it. 
>> 
>> The scenario that describes this failing test is: 
>> - A property has attached a few InvalidationListeners and a few 
>> ChangeListeners (so the Generic path is followed, and not the SingleChange). 
>> - At some point the property changes to a non-null `newValue`, and all the 
>> listeners get notified, one by one, in the same order as they were 
>> registered. 
>> - If, for any reason, and _while_ the different change listeners are being 
>> notified, all the change listeners are removed,  the pending listeners will, 
>> unexpectedly,  receive `null`, though the listeners that were notified 
>> before that happens received the expected `newValue`.
>> 
>> For simplicity, the test simulates that exact moment by removing all the 
>> change listeners when `listenerA` receives `newValue` from the `changed` 
>> notification. Then the `currentValue` gets nullified, and then the pending 
>> listeners in the notification list `curChangeList` (`listenerB` and 
>> `listenerC`, as they were registered after `listenerA`) get `null`. There 
>> could be other triggers, as long as they happen on the same thread, as part 
>> of the same call stack that started `fireValueChangedEvent`.
>> 
>> With the fix, by capturing the `currentValue` into `newValue` before 
>> entering the loop, then `B` and `C` will still receive `newValue`, 
>> regardless any possible change done to `currentValue` in the process.
>> 
>> ---------
>> - [X] I confirm that I make this contribution in accordance with the 
>> [OpenJDK Interim AI Policy](https://openjdk.org/legal/ai).
>
> Jose Pereda has updated the pull request incrementally with one additional 
> commit since the last revision:
> 
>   assign back to helper

We have two choices for this PR:

1. Proceed with this PR for jfx27. Even though it is at best an incomplete fix 
for the problem, it does avoid an NPE in a specific use case
2. Abandon this PR and wait until jfx28 when the underlying problem will be 
fixed by #1081

Presuming this PR has no ill effects, that is, there are no cases where it 
misbehaves or has worse behavior than today, then 1 seems a reasonable option. 
Paraphrasing @mstr2, If we think this is "_more correct_" than what we have 
today, then we can proceed on the principle "that _more correct_ is better than 
_less correct_".

Either way, we should continue to push on #1081 and get it in early in jfx28 -- 
some time this month would be great. I note that if we go with option 1, we 
would likely want to revert it as part of #1081, since it would no longer be 
necessary.

@mstr2 @hjohn What is your recommendation?

-------------

PR Comment: https://git.openjdk.org/jfx/pull/2200#issuecomment-4981506678

Reply via email to