excessive / cpml

Cirno's Perfect Math Library: Just about everything you need for 2D/3D games. Hopefully.
https://excessive.github.io/cpml/
Other
304 stars 41 forks source link

cpml.mat4.from_transform() computes the wrong rotation matrix. #83

Open leonardus opened 1 year ago

leonardus commented 1 year ago

cpml/modules/mat4.lua:

local rx, ry, rz, rw = rot.x, rot.y, rot.z, rot.w

local rm = new {
  1-2*(ry*ry+rz*rz), 2*(rx*ry-rz*rw), 2*(rx*rz+ry*rw), 0,
  2*(rx*ry+rz*rw), 1-2*(rx*rx+rz*rz), 2*(ry*rz-rx*rw), 0,
  2*(rx*rz-ry*rw), 2*(ry*rz+rx*rw), 1-2*(rx*rx+ry*ry), 0,
  0, 0, 0, 1
}

Something about how that rotation matrix is calculated is wrong. Because when I replace it with my own function, that gets the rotation matrix the same way mat4.from_quaternion computes the matrix (using q:to_angle_axis()), I get the correct results that I was expecting:

function dbg_from_transform(trans, rot, scale)
    local sm = cpml.mat4.new{
        scale.x, 0,       0,       0,
        0,       scale.y, 0,       0,
        0,       0,       scale.z, 0,
        0,       0,       0,       1,
    }

    local rm = cpml.mat4.from_angle_axis(rot:to_angle_axis())

    local rsm = rm * sm

    rsm[13] = trans.x
    rsm[14] = trans.y
    rsm[15] = trans.z

    return rsm
end
leonardus commented 1 year ago

Specifically, it seems like the issue is that the rotation matrix being generated is transposed. Here are the matrix results using translation of (0, 0, 0), rotation of (0.4301481, 0.7169136, 0.2867654, 0.4677318), and scale of (1, 1, 1):

Native (cpml.mat4.from_transform()): [ -0.192, +0.348, +0.917, +0.000, +0.885, +0.465, +0.009, +0.000, -0.424, +0.814, -0.398, +0.000, +0.000, +0.000, +0.000, +1.000 ]

Custom implementation using cpml.mat4.from_angle_axis(rot:to_angle_axis()): [ -0.192, +0.885, -0.424, +0.000, +0.348, +0.465, +0.814, +0.000, +0.917, +0.009, -0.398, +0.000, +0.000, +0.000, +0.000, +1.000 ]