Matrix3D multiplication operator bug

General JUCE discussion
1

I believe multiplication operator in Matrix3D  has a bug.

mat and m2 are swapped. It should be like this:


        return Matrix3D(
            mat[0] * m2[0] + mat[1] * m2[4] + mat[2] * m2[8] + mat[3] * m2[12],
            mat[0] * m2[1] + mat[1] * m2[5] + mat[2] * m2[9] + mat[3] * m2[13],
            mat[0] * m2[2] + mat[1] * m2[6] + mat[2] * m2[10] + mat[3] * m2[14],
            mat[0] * m2[3] + mat[1] * m2[7] + mat[2] * m2[11] + mat[3] * m2[15],
            mat[4] * m2[0] + mat[5] * m2[4] + mat[6] * m2[8] + mat[7] * m2[12],
            mat[4] * m2[1] + mat[5] * m2[5] + mat[6] * m2[9] + mat[7] * m2[13],
            mat[4] * m2[2] + mat[5] * m2[6] + mat[6] * m2[10] + mat[7] * m2[14],
            mat[4] * m2[3] + mat[5] * m2[7] + mat[6] * m2[11] + mat[7] * m2[15],
            mat[8] * m2[0] + mat[9] * m2[4] + mat[10] * m2[8] + mat[11] * m2[12],
            mat[8] * m2[1] + mat[9] * m2[5] + mat[10] * m2[9] + mat[11] * m2[13],
            mat[8] * m2[2] + mat[9] * m2[6] + mat[10] * m2[10] + mat[11] * m2[14],
            mat[8] * m2[3] + mat[9] * m2[7] + mat[10] * m2[11] + mat[11] * m2[15],
            mat[12] * m2[0] + mat[13] * m2[4] + mat[14] * m2[8] + mat[15] * m2[12],
            mat[12] * m2[1] + mat[13] * m2[5] + mat[14] * m2[9] + mat[15] * m2[13],
            mat[12] * m2[2] + mat[13] * m2[6] + mat[14] * m2[10] + mat[15] * m2[14],
            mat[12] * m2[3] + mat[13] * m2[7] + mat[14] * m2[11] + mat[15] * m2[15]);
 

2

So, if this were just an ordinary 4x4 matrix, then yes, I would agree -- the version you wrote down is the one found in math textbooks and the one I know.

However, I hesitate to just go ahead and fix this. The Matrix3D class is a 4x4 matrix but representing 3D space. Apparently this is how you represent 3D graphics in homogeneous coordinates / projective geometry for computer graphics. Unfortunately I am not really familiar with the topic, but this seems to be different from the classic cartesian cordinates / euclidean space. I can imagine that in this realm, maybe special conventions apply, which is why it is implemented the way it is?

Any graphics expert here who knows this?

3

Using "homogeneous coordinates" is a trick, which allows the translation to be handled like the other transforms, i.e. as a matric multiplication. The multiplication however follows the normal rules, hence the formular above should be performing a 4x4 matrix multiplication.

 

4

Transformation matrix is usually calculated like this:

matrix = projection * view * model;

Unfortunately due to this bug one can not use multiplication operator to calculate matrix.

 

 

5

If all else fails, use GLM (http://glm.g-truc.net/0.9.6/index.html)

6

Thanks all for your comments, they are consistent with other things I have found today on the topic.

So yes, this operator* should be just a normal matrix multiplication, like the one we are all familiar with, and therefore was implemented the wrong way around.

I fixed it on the newest tip.