crash during PhysicsCollisionObject_finalizeNative about minie

[stephengold]

for the record:

        Collection<PhysicsJoint> joints = physicsSpace.getJointList();
        for (PhysicsJoint joint : joints) {
            physicsSpace.removeJoint(joint);
        }
        Collection<PhysicsCollisionObject> pcos = physicsSpace.getPcoList();
        for (PhysicsCollisionObject pco : pcos) {
            physicsSpace.removeCollisionObject(pco);
        }

[stephengold]

I've been busy with other bugs. If you could provide a simple Java app to trigger the crash, that might be very helpful.

[stephengold]

I have a possible fix: stephengold/Libbulletjme@489383e

[ertugrulcetin]

Once you release a test version, I can test it on my machine.

[ertugrulcetin]

Seems like it's fixed! Thank you so much, good job!

[stephengold]

Thanks for the debug crash log ... that's helpful!

To self: findLinearSearch() on the world's collision-object array caused a segmentation violation. Was the world in some inconsistent state?

[ertugrulcetin]

Could be, I was using re-init fn in here (several times in a row) clear fn does the real job. Maybe I'm doing something wrong. The goal is to re-initialize the app without restarting it.

[stephengold]

Sorry, I don't understand Clojure yet. I've been studying the C++ code to figure out how an inconsistency might've arisen. No clarity yet. I may build a modified library and ask you to re-test...

[ertugrulcetin]

Sure, I'll re-test it. From my understanding, whatever we do in Java, Clojure in any other JVM language that uses jMonkeyEngine this crash should not happen right? It's the C++ code that has the problem? Or user code?

[stephengold]

I'm convinced this is a Minie bug, not a bug in JME or your application. It's likely you can work around it by clearing out physics spaces before they get GC'ed, but Minie shouldn't require that.

I expect the ultimate fix will probably involve changes to Minie's native libraries.

[ertugrulcetin]

I see. Also, I get JVM crash on bullet and bullet-native deps too (That's why I switched to Minie in the first place). One question regarding that, how can I manually clear physics spaces?

[ertugrulcetin]

@stephengold is there any update? Also, is there anything that I can help with? (testing, Java-related stuff). Sorry, I don't know C++ :(

[stephengold]

The new "Minie-3.1.0-test4" build incorporates a possible workaround for this issue. Try it and let me know whether (or not) it makes any difference for your application.

[ertugrulcetin]

Sure, will try ASAP.

[ertugrulcetin]

@stephengold I was going to write manual clearing physics space worked, but it failed last minute. Here is the JVM crash log: https://gist.github.com/ertugrulcetin/a28089705073b17cbf51137a2bba3ca1 it took longer compared to previous crashes.
jmonkey version: 3.3.2-stable, Minie version: 3.1.0-test4

In clear function, those are the steps:

• Detach all children from rootNode
• Clear all lights from rootNode
• Get all states from stateManager using getStates method (reflection) and call .detach method, exclude these AppStates: AudioListenerState, DebugKeysAppState, FlyCamAppState, ResetStatsState, StatsAppState
  • If BulletApp state then call clear-physics
• invoke terminatePending method from stateManager (reflection)
• call simpleInitApp method

[ertugrulcetin]

UPDATE: False alarm. It turns out I forgot to add manual clear fn to some other part that required it. It works, thanks!

It does not work if I don't clear physics spaces manually tho.

[stephengold]

By commenting out the physicsSpace.destroy() in "TestIssue13.java" I have reproduced this issue on my Linux system. Therefore, this issue is not OSX-specific.

[stephengold]

The new "Minie-3.1.0" build incorporates the (proposed) fix. Try it and let me know whether (or not) it makes any difference for your application.