On Thu, 23 Apr 2026 19:02:29 GMT, Thorsten Fischer <[email protected]> wrote:
> Hi all, > this is the topic I opened on the mailing list for this ticket: > https://mail.openjdk.org/archives/list/[email protected]/thread/XJB272UTZSNRYTQTHSALXYY4VO23DCIU/ > > Additional notes to the problem: > - The error happend with a dated driver, but updating it to a recent version > did not fix the problem. > - I could not reproduce the problem with a similar notebook with the same > old(!) driver, but newer GPU revision. > - I can verify that the fix below seems to fix the problem: the application > runs now for 5 days straight, which did never happen before with that > machine. The application runs on the FX 17, so my 'long term test' is with > that branch. > > Notes to the patch: > - This patch reduces the GL calls between frames. > - During normal rendering, currentDrawable is now null between present() and > createGraphics() (which sets it back to a real drawable), but it is then > never read. Outside that window, the only reader is clearContext() on the > shutdown path, which already has a null guard. currentDrawable is only > written in invalidateCurrentDrawable() and makeCurrent(), and only read in > makeCurrent() and clearContext(). > - On the shutdown path (clearContext), dummy.swapBuffers() would no longer be > called, which I believe is fine. > - ES2SwapChain.dispose() on master calls drawable.dispose(), which requires > the drawable to be non-current. Since present() no longer parks the context > on the dummy, that step moves into the dispose path: > context.makeCurrent(null) before drawable.dispose(). > > --------- > - [x] I confirm that I make this contribution in accordance with the [OpenJDK > Interim AI Policy](https://openjdk.org/legal/ai). Code change invalidates only the Java level current drawable after swap. I see no harm in doing this and i see that it maintains the current drawable state properly. Also i don't see need for similar change in other pipelines like Metal/D3D, as we manage things differently on other pipelines. I tested Ensemble8 with this change using ES2 pipeline on my M1 Mac. Played around with animating demos in multi monitor setup and i didn't observe any issue. LGTM. ------------- Marked as reviewed by jdv (Committer). PR Review: https://git.openjdk.org/jfx/pull/2159#pullrequestreview-4335357623
