RFR: 8377850: AbstractPrimaryTimer: concurrent addAnimationTimer() causes lost timers and hangs

Florian Kirmaier fkirmaier at openjdk.org
Tue Feb 17 21:44:35 UTC 2026


On Thu, 12 Feb 2026 23:03:31 GMT, Florian Kirmaier <fkirmaier at openjdk.org> wrote:

> Add `synchronized(lock)` around array mutations in `addPulseReceiver`/`removePulseReceiver`/`addAnimationTimer`/`removeAnimationTimer` and around snapshot-taking in `timePulseImpl`. Iteration remains outside the lock. `updateAnimationRunnable()` is also called outside the lock to avoid nested locking.
> 
> This preserves the existing copy-on-write design - the lock just ensures it works correctly across threads. Performance impact is minimal: the lock only covers field reads/writes, not the per-frame iteration.
> 
> **Testing:**
> New `AbstractPrimaryTimerThreadSafetyTest` with `testConcurrentAddAnimationTimer` - 8 threads add timers simultaneously, repeated 100 times. Fails 100% without fix. Reuses `AbstractPrimaryTimerStub` from existing tests.
> 
> All existing animation tests pass.
> 
> In JPro, this often caused a Deadlock during startup.
> This might have caused many bugs, which are very hard to reproduce.

PR is probably wrong, due to a missunderstanding on my side.
For some reason, it made my deadlock disapear.
But the actual problem was somewhere else, which I identified (independent of the JFX codebase).
Will probably close the PR and ticket in the coming days, after I've verified i didn't miss anything.

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

PR Comment: https://git.openjdk.org/jfx/pull/2074#issuecomment-3917226967


More information about the openjfx-dev mailing list