NaN crashes? [SOLVED]

[ejtttje]

Hi, I'm upgrading my app to 2.77 from 2.76, and now I keep getting NaN-based crashes. It seems to be focused on whenever I have a btGeneric6DofConstraint active in the dynamics world, after the simulation update my graphics chokes on the NaNs coming out of the positions and orientations. (I haven't tried other constraints, not sure if this is specific to btGeneric6DofConstraint or common to all constraints.)

However, if I disable adding the constraint to the dynamics world, then everything runs fine, except of course everything falls apart.

So, just a quick question if anyone has any ideas what relevant issues might be in the new version so I can focus my debugging? Thanks!

EDIT: Also, the problem persists in the current SVN.

[ejtttje]

Followups: it's not immediately returning NaN after all. Just within the first few physics steps, things 'explode', e.g. [position] {quaternion}

Code: Select all

[-276.442 -1150.68 15453.3]ᵀ {0.365876 0.506803 -0.250501 0.73928}
[1.55066e+09 -5.01065e+09 8.94692e+10]ᵀ {0.988438 -0.0544533 -0.0913721 0.108057}
[9.03737e+15 -2.9872e+16 5.26062e+17]ᵀ {0.82155 -0.109615 -0.556292 0.0598202}
[5.35732e+22 -1.7708e+23 3.11846e+24]ᵀ {0.968297 -0.0706576 -0.21874 0.0977865}
[3.17579e+29 -1.04972e+30 1.84861e+31]ᵀ {0.980996 -0.0219184 0.149094 0.122216}
[nan nan nan]ᵀ {nan nan nan nan}

That's just for the 'base frame' of my mechanism, the supposedly constrained components are similarly exploding to the four corners of the world (I don't think they are moving together, different directions)
Anyway, I'll keep at it.

[ejtttje]

The problem was with btContactSolverInfoData::m_damping. Previously (2.76) I had this set to 50 during the program initialization. This high m_damping value appears to cause constraints to explode in 2.77, if I leave this at the default (1) everything is happy.

I'm not sure why I had that parameter so high, my simulation seems to run the same in 2.76 with the default value. Perhaps I had been testing different values and didn't see any effect from that one and forgot to remove the test value *shrug*