Bullet's Voronoi Simplex Solver

[langford]

Well it is a convex sweep test, from the most outer API, but just before the call to rayTest is this comment /// unfortunately the implementation for rayTest and convexSweepTest duplicated, albeit practically identical

In any case the problem I describe is real and is exactly on topic to the GJK simplex numerical issues in this thread. The failures ultimately occur from btGjkPairDetector::getClosestPoints()

There is sample code and a whole bunch of repro cases in the issue I wrote up

http://code.google.com/p/bullet/issues/detail?id=250

[Erwin Coumans]

Well it is a convex sweep test

Ah, the convex sweep test indeed uses GJK, but the btCollisionWorld::rayTest doesn't.

Thanks a lot for the reproduction case. I checked it, and this particular case is fixed now: Bullet detected a degenerate GJK case, falls back into EPA but finds no penetration. The second GJK run (without all the degeneracy checks) reports a valid positive distance.

Can you please check latest Bullet 2.75-rc4 and see if it fixes your issues?
http://code.google.com/p/bullet/issues/detail?id=250

If you have other failures, please report them in the issue tracker.
Erwin