Bullet 2.83

[scrawl]

Yesterday I was pondering the idea to improve the physics code before I re-implement it in the OSG branch. It just so happens, that today Bullet 2.83 was released, including the fix for the issue with btScaledBvhTriangleMeshShape we had.

The practical benefits would be reduced memory usage and faster loading times, since we no longer need to duplicate the vertex data when creating a scaled instance of a collision shape.

Just did a test run with btScaledBvhTriangleMeshShape, everything seemed to work fine, but I would like to get some more testing (playtesting) if possible.
https://github.com/OpenMW/openmw/pull/563

[psi29a]

two things

1) bullet will stick around, even when the switch is made to OSG right?

2) https://packages.debian.org/source/sid/bullet <-- currently 2.82, I can file a bug to get it bumped to 2.83 but I'm not sure about the turnaround time. This might delay the release for Debian and Ubuntu.

[Zini]

#2 is an important point. I am not overly happy with updating a dependency immediately before a release. Shouldn't we move that to 0.27.0? I guess that would also give psi29a time to sort out the Ubuntu(Debian situation.

[psi29a]

If this could be pushed back until 0.37, that would give Scott and I enough time to deal with problem.

If you feel, otherwise, no worries. We'll have it up in our PPA so at least Ubuntu can have a release.

Update: I've managed to create packages, so getting this in our PPA should be simple. I've emailed my diff to the bullet maintainers at Debian, waiting for their response.

[Zini]

0.37.0 then. It's not urgent and we will get enough testing anyway.

[corristo]

Updated Bullet to 2.83, built current master with PR merged (0c2786908912d61e4ceb18e7fe1a7991e402a0da), experiencing consistent crashes in exteriors.

Stacktrace:

Code: Select all

Exception Type:        EXC_BAD_ACCESS (SIGSEGV)
Exception Codes:       KERN_INVALID_ADDRESS at 0x0000000000000000

Thread 0 Crashed:: Dispatch queue: com.apple.main-thread
0   ???                           	000000000000000000 0 + 0
1   openmw                        	0x000000010890ba51 btCompoundShape::updateChildTransform(int, btTransform const&, bool) + 167
2   openmw                        	0x00000001080b8ec1 (anonymous namespace)::animateCollisionShapes(std::map<OEngine::Physic::RigidBody*, OEngine::Physic::AnimatedShapeInstance, std::less<OEngine::Physic::RigidBody*>, std::allocator<std::pair<OEngine::Physic::RigidBody* const, OEngine::Physic::AnimatedShapeInstance> > >&, btDynamicsWorld*) + 1649
3   openmw                        	0x00000001080b8835 MWWorld::PhysicsSystem::stepSimulation(float) + 85
4   openmw                        	0x0000000107fe5e83 MWWorld::World::doPhysics(float) + 51
5   openmw                        	0x0000000107fe7179 MWWorld::World::update(float, bool) + 121
6   openmw                        	0x000000010827410e OMW::Engine::frameRenderingQueued(Ogre::FrameEvent const&) + 2046
7                                 	0x000000010aca8af3 Ogre::Root::_fireFrameRenderingQueued(Ogre::FrameEvent&) + 79
8                                 	0x000000010aca91ab Ogre::Root::_updateAllRenderTargets(Ogre::FrameEvent&) + 43
9                                 	0x000000010aca913e Ogre::Root::renderOneFrame(float) + 80
10  openmw                        	0x0000000108279821 OMW::Engine::go() + 1921
11  openmw                        	0x000000010825d04f main + 1743
12  openmw                        	0x0000000107c686b4 start + 52

[Sslaxx]

Don't suppose someone could temporarily revert the 2.83-specific changes could they, at least until the PPA situation is updated?

[scrawl]

Updated Bullet to 2.83, built current master with PR merged (0c2786908912d61e4ceb18e7fe1a7991e402a0da), experiencing consistent crashes in exteriors.

Didn't crash for me, but now I see why it might have. Could you try again with commit 60fffec3452c68ef2675a7c573af0436c051a6d4 please.

[corristo]

Didn't crash for me, but now I see why it might have. Could you try again with commit 60fffec3452c68ef2675a7c573af0436c051a6d4 please.

Fixed, thanks!

[psi29a]

Got a reply back from the Debian maintainer:

Hi Bret,

thanks for the info. I'll see if I can attend to Bullet within the next
couple of days.

Cheers,

Markus