The body flips wierdly to the other side(the side of the body becomes bottom or sth simillar) after about full circle and than rotates normaly (tough its flipper wrong way).
The idea is your Irrlicht node is connected to your bullet 'body'. So you should alter your bullet 'body' in your simulation to get your irrlicht node to do as you like.
And so, I believe you aren't supposed to directly set physics components after the initial setup stage for the simulation, so you cant just rotate the bullet body's rotation directly. It's a taboo to do so, but if you're trying to, say, cap the velocity of an object, you have no choice.
In your case, if you are trying to rotate it, I'd say you need to look into an 'addAngularVelocity' function and use the Y-component of the vector you give it as the speed to rotate bullet body. I suppose addTorque you could achieve the same thing.
__________________________________
...you'll never know what it's like, living your whole life in a dream...
EDIT: ok somehow got this working tough torque isn't what I would like to use to rotate my player as I want it to be controlled by mouse like First person game.
In one of my FPS controller tests with irrBullet, I found it worked pretty well to use the Irrlicht FPS camera for basically all movement as the head. Even rotation around the Y axis.
Then a force is applied on the body in the direction the camera is pointing (except its X rotation).
Here's the input handler for it (the code is not entirely finished, as you can see, but the movement part works):
Yes, with my above code you just apply a force in the direction the camera is facing.
I don't think there's any sense in rotating a perfectly symmetrical capsule when you can simply apply the force in the direction the camera is pointing (make sure you set velocity.Y to 0 so it doesn't apply any upward or downward force).