bullet 2.82 invalid normal direction on collision/sweep test

[shelomentsev]

Sweep test vs btBvhTriangleMeshShape on some triangles returns normal in opositional direction.
I start digging this issue, I also found that collision boxes fall through that triangles(invalid contact normal).
If I change position one of triangle vertex then sweep return valid normal(same as box contact normal)
Here is source code tho show that issues(modified BasicDemo)
https://github.com/shelomentsev/bullet2_bug

[antarespilot]

I've having this problem as well. Boxes falls through about half of my triangles.

It works using 2.81, but not with 2.82. This is a rather serious regression I would say.

[erwincoumans]

It should be fixed in latest github revision, did you try it?

[antarespilot]

Hi, just did, and everything works fine again. :-) Tested using double-precision-mode.

I had not tried compiling from this repo in a while (using Linux), great to see all the things that have happened here during the last month or so.

[shelomentsev]

I just did too, bug not fixed.
If I set m_fixContactNormalDirection = 0, it fix bug