Hinge constraint bugged in bullet 2.75

[Dragonlord]

I came across some nasty problem with the new 2.75 version of bullet. In my engine constraints are specified as 6-dofs but internal code tries to use more specific bullet constraints if the limits match certain criteria. For example for a single limited degree of freedom a hing constraint is used. This worked well so far but suddenly fails in bullet 2.75 . The constraint is set up along the z axis as the swing axis and with the orientation adjusted so that the swing limits are [-x..x] where x is the maximum swing of the door. Hence with a door having 160 degrees limit this would be [-80..80]. Now bullet 2.75 develops a strange behavior. When the door moves from the positive swing limit over to the negative swing limit passing the zero point the door flips around one of the locked axes (x or y). This behavior had not been present in 2.73 . Can you examine the changes made if some of them broke the hinge constraint? For the time being I simply disallow hinge constraints using 6-dof instead which though increases the number of broken constraints to 2 ( cone-twist has been broken already in 2.73 exhibiting explosion behavior with ConstraintDemo like values ). It's a bit silly to have specialized constraints if they are broken.

[Erwin Coumans]

Can you file an issue in the google issue tracker, with a reproduction case (so your feedback doesn't get lost in forum discussions)? http://code.google.com/p/bullet/issues/list

Thanks!
Erwin

[Dragonlord]

I'd love to but the Collada exporter is not compiling at all so I'm a bit out of ideas on how to produce a test-case.

[Erwin Coumans]

In that case please file an issue without a reproduction case (but with description and a link to your forum posting).

A future Bullet will have built-in serialization, so no hassle with COLLADA compilation required anymore.
Cheers,
Erwin

[Dragonlord]

I think I wrestled it down. I somehow managed to get it to compile. Placed the code in a separate SCons environment so I can rip it out again once that serialization works ( as the setup is really, really ugly right now :O ). When I manage to get a snapshot I'll attach the output file if this would be of help.