05.01.2017, 23:12
(
Last edited by IllidanS4; 06/02/2017 at 07:37 PM.
)
Added GetVehicleRotationQuatFixed which returns correct values for unoccupied vehicles. GetVehicleMatrix must be provided by a plugin, like YSF. If you are using VehicleMatrix.inc, it could fix a couple of cases, but it will still be mostly garbage values. A plugin is required to read the matrix from the server's memory.
Why GetVehicleMatrix? Internally, the server stores the vehicle's rotation as a 3Ч3 rotation matrix, the same way as the game does. For some reason, this matrix is not exposed to scripts, only the conversion to a quaternion. Unfortunately, unoccupied vehicle updates damage the third row of the matrix and invalidate the values, changing them to near-zeros (mostly denormal floats). I have no idea why that is, I guess some sort of a game/server bug. The internal conversion to the quaternion sadly uses the third row (though it doesn't have to, two rows of the matrix are completely sufficient), returning complete bogus in case it was damaged.
Now that I know the third row is actually redundant and is just a cross product of the other rows, if I can obtain the internal matrix, just fixing the third row via this method is sufficient and produces correct quaternions. Due to the imprecise floating point calculations, the reconstructed row is a bit inaccurate, but still really close to the correct one. It should fix 100 % of wrong quaternions.
Why GetVehicleMatrix? Internally, the server stores the vehicle's rotation as a 3Ч3 rotation matrix, the same way as the game does. For some reason, this matrix is not exposed to scripts, only the conversion to a quaternion. Unfortunately, unoccupied vehicle updates damage the third row of the matrix and invalidate the values, changing them to near-zeros (mostly denormal floats). I have no idea why that is, I guess some sort of a game/server bug. The internal conversion to the quaternion sadly uses the third row (though it doesn't have to, two rows of the matrix are completely sufficient), returning complete bogus in case it was damaged.
Now that I know the third row is actually redundant and is just a cross product of the other rows, if I can obtain the internal matrix, just fixing the third row via this method is sufficient and produces correct quaternions. Due to the imprecise floating point calculations, the reconstructed row is a bit inaccurate, but still really close to the correct one. It should fix 100 % of wrong quaternions.